Determinants ofFinancially BurdensomeFamily Health ExpensesUnited States, 1980Series C, Analytical Report No. 6

I

Determinants ofFinancially BurdensomeFamily Health ExpensesUnited States, 1980Series C, Analytical Report No. 6

U.S. DEPARTMENT OF HEALTH AND HUMAN SERVICES

Published byPublic Health Service’

Centers for Disease ControlNational Center for Health Statistics

April 1988

National Medical Care Utilizationand Expenditure Survey

The National Medical Care Utilization and Expenditure

Survey (NMCUES) is a unique source of detailed nationalestimates on the utilization of and expenditures for varioustypes of medical care. NMCUES is designed to be directlyresponsive to the continuing need for statistical information

on health care expenditures associated with health servicesutilization for the entire U.S. population.

NMCUES will produce comparable estimates over time

for evaluation of the impact of legislation and programs onhealth status, costs, utilization, and illness,related behaviorin the medical care delivery system. In addition to nationalestimates for the civilian noninstitutionalized population, it

will also provide separate estimates for the Medicaid-eligiblepopulations in four States.

The first cycle of NMCUES, which covers calendar year1980, was designed and conducted as a collaborative effort

between the National Center for Health Statistics, Public HealthService, and the Office of Research and Demonstrations,Health Care Financing Administration. Data were obtainedfrom three survey components. The first was a national house-

hold survey and the second was a survey of Medicaid enrolleesin four States (California, Michigan, Texas, and New York).Both of these components involved five interviews over aperiod of 15 months to obtain information on medical care

utilization and expenditures and other health-related informa-

tion. The third component was an administrative records surveythat verified the eligibility status of respondents for the Medi-care and Medicaid programs and supplemented the householddata with claims data for the Medicare and Medicaid

populations.Data collection was accomplished by Research Triangle

Institute, Research Triangle Park, N. C., and its subcontractors,

the National Opinion Research Center of the University ofChicago, 111., and SysteMetrics, Inc., Berkeley, Calif,, underContract No. 233–79–2032.

Co-Project Officers for the Survey were Robert R,Fuchsberg of the National Center for Health Statistics (NCHS)and Allen Dobson of the Health Care Financing Administration(HCFA). Robert A. Wright of NCHS and Larry ,Corder ofHCFA also had major responsibilities, Daniel G. Horvitz of

Research Triangle Institute was the Project Director primarilyresponsible for data collection, along with Associate ProjectDirectors Esther Fleishman of the National Opinion ResearchCenter, Robert H. Thornton of Research Triangle Institute,and James S. Lubalin of SysteMetrics, Inc, Barbara Moserof Research Triangle Institute was the Project Director primar-ily responsible for data processing.

I?or sale by the Superintendent of Documents, U.S. Government Printing Office Washington, D.C. 20402

Copyright Information

All material appearing in this report is in the public domainand may be reproduced or copied without permission; citationas to source, however, is appreciated.

Suggested Cdation

Dicker, M., and Sunshine, J. H.: Determinants of financiallyburdensome family health expenses, United States, 1980.National Medical Care Utilization and Bpenditure Survey.Series C, Analytical Report No. 6. DHHS Pub. No. 86-20406.National Center for Health Statistics, Public Health Service.

Washington. U.S. Government Printing Office, Apr. 1988.

Library of Congress Cataloging-in-Publication Data

Dicker, Marvin.Determinants of financially burdensome family health

expenses.(National Medical Care Utilization and Expenditure Survey.

Series C, Analytical report ; no. 6) (DHHS publication ; no.88–20406)

Bibliography: p.1. Family—Medical care-United States-Costs-

Statistics. 2. Family—Medical car&United States-Costs—Statistical methods. 1.Sunshine, Jonathan H. Il. NationalCenter for Health Statistics (U.S.) Ill. Title. IV. Series:National Medical Care Utilization and Expenditure Survey(Series). Series C, Analytical report; no. 6. V. Series: DHHSpublication ; no. 88-20406. [DNLM: 1. Expenditures, Health-United States. 2. Financing, Personal—United States. W2 AN224n no. 6]RA407.3.D47 1988 338.4’33621 87-600309ISBN 0-8406-0382-7

Executive Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Overall Findings:TheSixMeasures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Overall Findings:The Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Methodsand Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Sourceofthe Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Definitionofthe Family . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .StandardizationforPart-YearFamilies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .The Definition ofaHealthExpense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .AdjustmentstotheSample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .TheTwo-PartModel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Sampling Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Non-Sampling Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Statistical Significanceand Hypothesis Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Measuring Financially Burdensome Health Expenses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .SixBasicMeasures: Definitions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Three Dollar Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Total ChargesforHealth Care Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Out-of-PocketExpensesforHealthCare Services . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . .Total Out-of-PocketHealth Expenses . . . . . . . . . . . . . . . . . . . . . . . . .

Three Ratio Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Total Charges for Health Care Services asa Percent of Family (or Societal) income

Out-of-Pocket Expensesfor Health Care Sewices as aPercentof Family Income .Total Out-of-Pocket Expensesas aPercent of Famiiylncome . . . . . . . . . . . .

WhatLevel ofExpense is Financially Burdensome?. . . . . . . . . . . . . . . . . . . .Six BasicMeasures: Data Presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Three Dollar Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Total Chargesfor Health Care Services . . . . . . . . . . . . . . . . . . . . . . . .Out-of-Pocket Expenses for Health Care Services . . . . . . . . . . . . . . . . . .Total Out-of-PocketHealth Expenses . . . . . . . . . . . . . . . . . . . . . . . . .Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Three Ratio Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Total Charges for Health Care Services as aPercent of Family income . . . . . . .Out-of-Pocket Expenses for Health Care Services as aPercent of Family Income .Total Out-of-Pocket Health Expenses as aPercent of Family income . . . . . . . .Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The Determinants of Financially Burdensome Health Expenses: A Regression Analysis . . . . .The lndexof Financially Burdensome Health Care Expenses . . . . . . . . . . . . . . . . .Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Procedures Followed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Older Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Younger, Lower-income Families. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Younger, Better-Off Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

1

12445666777

8889

101010101010101011111111121212131414141415171819191919202121

2224

...111

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Y,...”+26Explanato~Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ~~~..~ 26Individual Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .+,I”! 27

PatternsAmong Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...~t~ 29Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..-’. 31

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . s“.’..””” 32

Appendixes1. Technical Noteson Regression Methods . . . . . . . . . . . . . . . . . . . , . . . . . . . . . . . . . . . ..t . . . . . 36Il. Technical NotesonSuIveyand Nonregression Methods . . . . . . . . . . . . . . . . . . . . . . . ... . . . . ../4. 52111.DefinitionofTerms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ,.~.,~.” 62

ListofTextTables

A,

B.

c.

D.

E.

F.

G.

H.

J.

K.

L.

M.

Total charges forhealth care formultiple-person fami~es, byageand status relative tothepove~ level: United States,1980 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . s**+4

Out-of-pocket expenses forhealth care sewices formultiple-person families, byageand status relative to the povertylevel: United States,1980 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..~.~

Total out-of-pocket health expenses for multiple-person families, by age and status relative to the poverty level: UnitedStates, 1980 . . . . . . . . . . . . . . .. i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ., ..,,,.

Total charges forhealth care asa percent of annual income formultiple-person families, byageand status relative tothepovertylevel: UnitedStates,1980 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r.~.

Out-of-pocket expenses for health care services as apercent ofannual income for muitiple-person families, byage andstatus relativetothe povertylevel: UnitedStates,1980 . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..l.~..

Total out-of-pocket health expenses as apercent of annual income for multiple-person families, by age and statusrelativetothepovertylevel:United States, 1980 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Significant regression findings forthe financial burden index for older multiple-person famflies. . . . , . . . , . . . , , ,

Significant regression findings forthe financialburdenindexforyounger,l ower-income multiple-person families , , , t .

Significant regression findings for the~nancial burden indexforyounger, better-off multiple-person families . . . . . , .

Comparison of multiple correlation coefficients squared, bydependent variable and age and family status relative tothepovertylevel . . . . . . . . . . . . . ‘., . . . . . . . . . . . . . . . . . . . . . . . . . . ..<S...<O!<../ fit

Statistically significant variables from:a setof regressions on the index of financially burdensome family health careexpenses arranged by the number:of family socioeconomic populations in which each variable was statisticallysignificant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..r. r.. ..i..<

Statistically significant variables from a set of regressions on total family charges for health care arranged by thenumber of family socioeconomic populations in which each variab[e was statistically significant . , . , . . . . , . . . . .

Symbols

No families with these characteristics insample

* Potential reliability problem; statistic isbased on sample size of fewer than 50or has relative standard error greaterthan 30 percent

12

12

13

15

16

17

22

23

25

26

29

30

. . . Category not applicable

iv

Determinants of FinanciallyBurdensome Family HealthExpenses:

~ United States, 1980By Marvin Dicker, Ph.D.,National Center for Health Statistics,and Jonathan H. Sunshine, Ph.D.,Applied Management Sciences, Inc.

Executive SummaryThis report focuses on two questions of current inter-

est to policy makers. First, “What percent of U.S. familiesexperience financially burdensome health expenses?”and, second, “What are the determinants of financiallyburdensome health expenses among U.S. families?” Thefirst question is addressed by examining how the distribu-tion in. the United States of families with financiallyburdensome health expenses is affected by six differentpossible measures of financial burden. The second ques-tion is addressed by using multiple regression techniqueson one of the measures selected as a preferred measure.

The data used are from the family data files ofthe 1980 National Medical Care Utilization and Expendi-ture Survey (NMCUES). This report presents data onapproximately 5,000 multiple-person families inter-viewed in this longitudinal survey. It provides a separateanalysis for each of three socioeconomic family popula-tions that have consistently been of interest to policymak-ers. These are (1) older families (defined for this reportas all U.S. multiple-person families with a member 65years of age or over); (2) younger, lower-income families(defined as all U.S. multiple-person families below 200percent of the poverty level in 1980 and with all membersunder 65 years of age); and (3) younger, better-offfamilies (defined as all U.S. multiple-person familiesat 200 percent of the poverty level or higher in 1980and with all members under 65 years of age).

NO~ The authors are grateful for the support received during all stagesof the preparation of this report from our colleagues at both the NationalCenter for Health Statistics and Applied Management Sciences, Inc. At theNational Center for Health Statistics, Thomas Hodgson consulted and advisedon herdth economics and econometrics, and Cecelia Snowden consulted andadvised on research methods and multiple regression techniques. MargaretCooke, Pennifer Erickson, and Robert A. Wright reviewed several draftsof the manuscript and also made many useful suggestions. Editors in thePublications Branch provided valuable assistance during all stages of Wlsreport,

At Applied Management Sciences, Inc., Colleen Goodman, JoAnnKuchak, and Alfred J. Meltzer provided executive management, skillfullymaking the firm’s resources available to meet the changing needs of theproject. Daniel Cohn provided programming skills as the staff member princi-pally responsible for data processing. Dr. James Bethel acted as statisticalconsultant for the project.

Overall Findings: The Six Measures

Two general conceptual approaches have been usedin the literature to assess financially burdensome healthexpenses. The first approach measures financial burdenby the size of a family’s health bill in dollars. Thesecond approach focuses on a family’s ability to payits health bill, and it measures financial burden as aratio of health expenses to family income. There is no ‘agreement on which of the two approaches is preferableand also no agreement on which of several operationalmeasures in each category is the most appropriate. Inorder to shed light on this controversy, this report com-pares six potentially useful operational measures of finan-cial y burdensome health expenses. Three are dollarmeasures and three are ratio measures,

The three dollar measures are (1) total charges forhealth care (irrespective of who pays the bill or whetheror not the bill is paid), (2) out-of-pocket expenses forhealth care services (family-paid premiums for healthinsurance are not included), and (3) total out-of-pocketexpenses for health (the previous measure plus out-of-pocket premiums).

The three ratio measures use the three dollar measuresto construct measures involving a ratio of expenses tototal family income. This gives (1) the ratio of totalcharges for health care to family income, (2) the ratioof out-of-pocket expenses for health care services tofamily income, and (3) the ratio of total out-of-pocketexpenses for health to family income.

Given these measures, the question still remains asto what level of expense, or ratio of expense to income,constitutes a financially burdensome expense. The usualpractice in the literature has been to use several differentthresholds arbitrarily selected from the upper part ofthe particular expense distribution under examination,and this practice is followed in this report.

The overall finding for the six measures was thatdifferent results were found for the different measureseven when the same threshold was used. For example,when looking at dollar measures, the following wasfound for a threshold of $3,000 or more in 1980. Fortotal health care charges, 18 percent of all U.S. familieshad dollar charges this high.. For out-of-pocket expen-ses for health care services (excluding family-paidpremiums), 2 percent of all U.~. families had dollar

1

expenses this high. And, finally, for total out-of-pocketexpenses for health (including premiums), 3 percent ofall U.S. families had dollar expenses this high. As thisexample illustrates, the percent of families found at anygiven dollar threshold was generally highest when healthcosts were measured by total charges, lowest when healthcosts were measured by out-of-pocket expenses for healthcare services, and in between when health costs weremeasured by total out-of-pocket expenses.

Different results for the same threshold were alsofound when ratio measures were compared. For example,the following was found for a financial burden thresholdof health expenses of 10 percent or more of incomein 1980. For total health care charges as a percent offamily income, 29 percent of all U.S. families had billsof 10 percent or more of their income: For out-of-pocketexpenses for health care services as a percent of familyincome, 6 percent of all U.S. families had expensesof 10 percent or more. And, finally, for total out-of-pock-et expenses for health (including insurance premiums)as a percent of family income, 12 percent of all U.S.families had expenses of 10 percent or more. Also,the overall pattern between the different ratio measureswas similar to the overall pattern found for the dollarmeasures. That is, the percent of families reaching anygiven ratio threshold was highest for total charges forhealth care as a percent of income, lowest for out-of-pocket expenses for health care services as a percentof income, and in between for total out-of-pocket ex-penses for health as a percent of income.

When total out-of-pocket expenses for health as apercent of family income was used as a measure offinancial burden in 1980, as indicated above, twice asmany U.S. families were found to have spent 10 percentor more of their income on health than when the morecommon measure of out-of-pocket expenses for healthcare services as a percent of income was used. (Aspreviously pointed out, the difference between these twomeasures is that the first ‘measure includes premiumsfor health insurance and. public health care coverageprograms paid by the family, whereas the second doesnot.) The inclusion of family-paid premiums in out-of-pocket health expenses had the consistent effect of ap-proximately doubling the proportion of U.S. familiesat a given expense-to-income threshold in 1980. Thus,the inclusion of family-paid premiums is an importantfactor in assessing the out-of-pocket health expense$and the financial burden of health that U.S. familiesbear.

The above patterns associated with the different typesof measures were also generally found when each ofthe three socioeconomic family populations wasexamined separately. However, the three socioeconomicpopulations also differed from each other in the propor-tion of families found at different financial burdenthresholds. For dollar measures, the general finding wasthat a larger proportion of older families than of youngerfamilies was found at each dollar threshold, but that

the two younger family categories did not differ fromeach other in the proportion of their populations foundat the different threshold levels. For example, 4 percentof older families had out-of-pocket expenses for healthof $3,000 or more while only 2 percent of familiesin each of the younger family categories had expensesthis high. For ratio measures, both older families andyounger, lower-income families had a higher proportionof their populations at a given ratio threshold level thanyounger, better-off families. For example, 27 percentof older families and 20 percent of younger, lower-income families had total out-of-pocket expenses forhealth of 10 percent or more of income compared withonly 4 percent of younger, better-off families. In short,patterns of relative financial burden differed with familysocioeconomic status, family age, and type of measureused. For dollar measures, only the age of family mem-bers was associated with differences in the distributionof financial burden in the U.S. population. For ratiomeasures, both the age of family members and familyincome were associated with such differences.

Overall Findings: The Regression Analysis

Of the six measures discussed above, the ratio oftotal out-of-pocket expenses for health to family incomewas chosen as the preferred measure for investigatingthe causes of financially burdensome health expensesamong U.S. families. This measure seemed to be thebest indicator of the total financial burden related tohealth that a family had, and it was labeled the financialburden index.

Multiple regression analysis was used to investigatethe effect on this index of family social and demographiccharacteristics, family health and illness characteristics,family income and health insurance characteristics, andfamily geographic and urbanization characteristics, Re-gressions were run separately for each of the threesocioeconomic family populations with the financial bur-den index as the dependent variable and approximately45 of the above family characteristics as independentvariables. Because of the large number of independentvariables involved, a multiple-step regression processwas used, which will no~-~e,describedhere, The statisticalsignificance” of the findings was assessed by using aSAS multiple regression program (SURREGR) that takesinto account the complex sample design of NMCUESand by using a F test that was analogous to a multiplet-test.

When the relationship between all 45 family charac-teristics and the financial burden index was examinedusing multiple regression techniques, the overall findingwas that in 1980 the major determinants of financiallyburdensome family health expenses (as measured bythe index) were family income and the completenessand type of health insurance coverage (or public healthcare coverage program) the family had. This was the

2

finding in each of the three socioeconomic populations.By contrast, health qtatus variables such as the generalhealth status of family members, special family healthevents (such as death or institutionalization of a familymember), and the specific illnesses that family members

~ had were much less important as overall predictors ofa family’s health-related i-inancial burden as measuredby the index. This comparatively negative finding forhealth status variables was especially surprising giventheir importance in predicting total family health chargesin a separate multiple regression analysis using. the samedata set, the same three socioeconomic populations, andthe same 45 family characteristics as independentvariables.

Several health status variables, however, did showup as population-specific predictors of the financial bur-

den index. Among older families, heart and circulatorydiseases (as a general category) was a statistically signifi-cant predictoq among younger, lower-income families,family work-loss days was a predictor; and, finally,among younger, better-off families, both hospitalizationand family illness days in bed were predictors.

In addition to these health-status variables, otherpopulation-specific predictors of the financial burdenindex were race of family head (significant for younger,better-off families); residence in the South (significantfor both older families and younger, better-off families);age and education of family head (significant for bothyounger family populations); and, finally, family head-spouse structure (significant for both older families andyounger, lower-income families).

3

Introduction

Overview

A consistent goal of contemporary U.S. health policyhas been adequate health care at an affordable out-of-pocket cost. In pursuing this goal, the most recent em-phasis has been on relieving financially burdensomehealth care expenses among U.S. families (CatastrophicIllness Expenses, 1986). However, there has been alack of adequate data on which to base national policyinitiatives. This report attempts to supply some of theneeded data. It is the fifth in a series of reports fromthe 1980 longitudinal National Medical Care Utilizationand Expenditure Survey (NMCUES) that have presenteddata on health care expenses and the use of health careby U.S. families (Dicker, 1983; Dicker and Sunshine,1987; Sunshine and Dicker, 1987a; Sunshine and Dicker,1987b).

Although many aspects of the U.S. health care sys-tem can be studied using the individual as the unit ofanalysis, the family is the more appropriate unit forstudying financially burdensome expenses. This is be-cause the financial burden of health care, and decisionsconcerning the use and financing of health care, usuallyare family responsibilities rather than individualresponsibilities.

Given that the family should be the unit of anal-ysis, the NMCUES is a particularly good data source,as it was originally conceptualized with family analysisin mind. It has a distinct and carefully conceptualizedmethod of longitudinal family construction, a collectionof specially developed longitudinal and. cross-sectionalfamily variables, and a family public use data tape thatwill allow other researchers to reexamine and carryfufiher the exploratory research presented here (PublicUse Data Tape Documentation: Family Data, 1986).

At the present time there is no agreement amongresearchers as to what. constitute financially burdensomehealth expenses for a family. Therefore, the first partof this report uses frequency table~o explore multipleconceptualizations and multiple operational definitionsof such expenses. On the basis of this examination,an index of financially burdensome family health ex-penses (or, for short, the financial burden index) wasdeveloped. It is defined as the ratio of annualized totalout-of-pocket family expenses for health (for all family

members nd including all family-paid premiums) toTannualized total family income (from all sources and

all family members). The concept of total out-of-pocketexpenses in the above definition differs from the conceptof out-of-pocket expenses for health care services inthat total out-of-pocket expenses include all family-paidpremiums for health care coverage.

The second part of this report uses the financialburden index as the dependent variable in a series ofmultiple regression analyses designed to test hypothesesabout the causes of financially burdensome health ex-penses among families. In testing these hypotheses, alarge number of sociodemographic, health-related,socioeconomic, and geographic explanatory variables areexamined.

Because the focus of this report is on supplyingdata for use in making health policy, three populationsof particular interest to policymakers are examined,These are (1) older families (defined as all U.S. multiple-person families with a member 65 years of age or over);(2) younger, lower-income families (defined as all U.S.multiple-person families with all members under 65 yearsof age and with characteristics (income, family size,and so forth) that placed them below 200 percent ofthe poverty level in 1980); and (3) younger, better-offfamilies (defined as all U.S. multiple-person familieswith all members under 65 years of age and with charac-teristics (income, family size, and so forth) that placedthem at 200 percent of the poverty level or higher in1980). Data were tabulated separately on all three familypopulations and a separate regression analysis was donefor each population. The underlying assumption of thisapproach was that both the causes of and remedies forfinancially burdensome health care expenses may bepopulation-specific.

This report, therefore, attempts to answer three ques-tions: What is the proper measure of financially burden-some health care expenses, what percent of U.S. familieshad such expenses in 1980, and, what are the determin-ants of such expenses among different policy-relevantpopulations of multiple-person families? It differs fromother reports on these subjects in its focus on multiple-person families, its systematic examination of differentmeasures of financial burden, its examination of a largerand more varied collection of variables, its consistent

4

controlling for age and family socioeconomic statusthrough the use of separate populations for analysis,and its use of multiple regression analysis.

Background

Wyszewianski (1986a) makes a distinction betweenthe size of a health care expense and a family’s abilityto pay the bill. This distinction underlies the major dis-tinction among studies dealing with financially burden-some or catastrophic health care expenses. Studies thatfocus on the size of health care expenses usually measurefinancially burdensome expenses by the total charges(in dollars) for health care services. If these chargesare large (above a certain threshold) they are consideredto be catastrophic charges. Wyszewianski suggests thatthis approach reflects the most common concept of cata-strophic expenses as involving illnesses with very largeexpenses. By contrast, studies that focus on a family’sability to pay the bill usually measure financially burden-some expenses by the ratio of out-of-pocket health careexpenses to family economic resources (usually income).In these types of studies, families with relatively moder-ate or low total charges maybe found to have burdensomehealth care expenses if the families are both poor andhave inadequate or no health care coverage. What followsis a discussion of selected studies from both orientationsto illustrate some of the issues involved.

Most studies on the size of health care expensesfocus on the individual rather than on the family asthe unit of analysis. However, one study that focuseson the family and is illustrative of this type of researchis the Congressional Budget Office study, CatastrophicMedical Expenses: Patterns in the Non-Elderly, Non-Poor Population (Koretz, 1982). This study used datacollected between 1974 and 1978 from a sample ofnon-poor families enrolled in the Federal Blue Crossor Blue Shield health benefits plan. The analysis com-bined one-person and multiple-person families into oneanalytic aggregate, Although the data were from theyears 1974-78, the analysis was done in terms of 1982constant dollars. The expense variable of interest wastotal annual health care expenses (charges) as reportedto Blue Cross, regardless of who paid for them. Four“catastrophic” thresholds were used to designate familieswith “high cost” or “catastrophic illnesses. ” Familiesat or above a threshold were considered to be familieswith catastrophic expenses. These thresholds were totalannual health expenses of $3,000, $5,000, $10,000,and $20,000. Two findings from the report are as follows:

1. Among the population sampled, families exceedingany of the catastrophic thresholds in a single yearare relatively rare, but they account for a sizableproportion of total health expenses. For example,only 5 percent of the examined families exceeded$5,000 in total expenses in a given year, but those

2.

families accounted for half of all expenses for allfamilies.

Although only a small proportion of families havecatastrophic expenses in a given year, a much largerproportion have high expenses at least once overseveral years. For example, over a 3-year period,27 percent of the families exceeded the $3,000 cata-strophic threshold at least once, while in a singleyear only 11 percent did so.

The above findings are interesting and important,but they deal with the size of health expenses ratherthan with a family’s ability to pay. Studies of the sizeof health expenses, therefore, are of limited use forstudying the financial burden of health expenses on afamily, though they are often very useful for studyingthe financial burden of health expenses on a society.

Some other limitations of the Koretz study shouldbe noted. Because it combines one-person and multiple-person families into one category, it is not a study offamilies as this concept is generally understood by thepublic and many social scientists, for whom the term“family” usually refers to a multiple-person social unit.In the Koretz study, multiple-person families cannot bedistinguished from one-person families. As a conse-quence, the percent of multiple-person families that ex-ceeds a given catastrophic threshold cannot be ascer-tained. Second, as its title indicates, the Koretz studyis not a study of the full family population of the UnitedStates. Finally, the study makes a valuable contributionin its use of multi-year data, but it lacks a rigorousdefinition of a longitudinal family. Because family group-ings change over time, there is a need for clear andconsistent rules identifying which changes constitute thebeginning or end of a family and which are regardedas changes in composition within the ‘samefamily. (SeeAppendix II for more information on this point.)

Recently there have been several studies that havefocused on a family’s ability to pay health care bills.They may all be considered as part of the same researchproject as they all use the same data base and havebeen written by researchers from the University of Michi-gan School of Public Health (Berki, et al., 1985; Berki,1986; Wyszewianski, 1986b). These studies used datafrom the longitudinal 1977 National Medical Care Ex-penditure Survey (NMCES). This survey used a represen-tative sample of the civilian noninstitutionalized popula-tion of the United States. The variable of interest inthese analyses was the percent of family ‘income cons-umed by out-of-pocket expenses for health care services.This measure of health expenses did not include out-of-pocket premium payments families made for health carecoverage. The analyses from the Michigan group com-bined one-person families and multiple-person familiesinto one analytic aggregate.

As in the Koretz report, a series of catastrophicthresholds was specified. But unlike the Koretz report,in which dollar thresholds were used, these reports used

5

percent-of-income thresholds. The thresholds were out-of-pocket health care expenses of 5 percent of income,10 percent of income, and 20 percent of income. Familiesat or above these thresholds were considered to befamilies with “highly burdensome or financially cata-strophic health care expenditures” (Berki, et al., 1985,p. 4). Some findings and conclusions from these reportsare as follows;,

1.

2.

3.

4.

Of all ‘U.S. families; 20 percent had out-of-pocketexpenses of 5 percent or more of income in 1977,and 4 percent had expenses of 20 percent or more(Berki, et al., 1985).

Families with high out-of-pocket expenses relativeto income accounted for a disproportionate amountof the total expenses for all families. For example,the 20 percent of families with expenditures of 5percent or more of income accounted for 42 percentof total expenditures for all families (Berki, et al.,1985).

Two distinct types of families have out-of-pocketexpenses that are relatively high in relation to in-come. These are families with a high-cost illnesswhere the costs are so large that insurance doesnot keep the family from being burdened, andfamilies with relatively small expenses but in-adequate or no health care coverage and low income(Berki, et al., 1985).

Most families with high out-of-pocket health servicesexpenditures relative-to their ~ncome resemble themedically indigent and the uninsured (Wys-zewianski, 1986b)l

Although from the above it seems that some of thefamilies with high out-of-pocket expenses relative toincome overlap with some of the families with hightotal expenses (in dollars), the focus on a family’s abilityto pay the bill also directs the researcher’s attentionto poor families with low expenses and to the role ofhealth insurance in reducing out-of-pocket costs.

Some of the limitations of the Michigan group’sresearch are similar to those previously discussed forthe Koretz study. The Michigan studies always use aunit of analysis composed of both one-person and multi-ple-person families, and they do not present a rigorousdefinition of a longitudinal family. The consequencesofthese limitations are the same as in the Koretz study.

A third and novel approach to measuring financiallyburdensome or catastrophic health care expenses wasused in the report of the Secretary of the Departmentof Health and Human Services to the President, Cata-strophic Illness Expenses (1986). This approach com-bined the two orientations discussed above. The reportdefined catastrophic expenses as out-of-pocket expensesthat exceeded an absolute dollar threshold (of either$2,200 or $4,400) plus a percent-of-income threshold(of either 5 percent or 10 percent). Note that in thismeasurement approach the dollar thresholds refer to out-

of-pocket spending rather than to total charges as inthe Koretz study. The approach used in the Secretary’sreport sets a dollar floor (determined by the dollarthreshold) below which out-of-pocket health expensesare not considered catastrophic, although they may stillbe experienced by the family as financially burdensome.

The above discussion indicates that not only is thereno agreement in the literature on what should be thethreshold indicating financially burdensome or cata-strophic expenses, but there is also no agreement onwhat measure of expenses the term “catastrophic” shouldrefer to. In the above studies, it has referred to totalcharges for health care, out-of-pocket expenses for healthcare services, and a ratio of out-of-pocket expenses forcare to family income. Better conceptualization is neces-sary, and it is carried out in this report in the chaptertitled “Measuring Financially Burdensome HealthExpenses.”

Methods and Limitations

Source of the Data

As previously pointed out, the data presented inthis report are from the National Medical Care Utilizationand Expenditure Survey (NMCUES). In NMCUES, in-formation was collected on health problems, health carereceived, expenditures for care, health insurance, andrelated topics. Data were obtained throughout calendaryear 1980 from a sample of the U.S. civilian nonin-stitutionalized population. NMCUES included both anational household sample encompassing approximately6,800 families and four state Medicaid samples. Allinformation in this report is based on the national house-hold sample. Detailed technical information on the sam-ple, on estimation procedures, and on measurement pro-cedures can be found in Appendix II.

The NMCUES differs from most surveys of healthin that it was a panel (or longitudinal) survey. Altogether,there were either four or five interviews, approximately3 months apart, that were conducted with each familyin the sample from early 1980 to early 1981, In eachinterview, information on all family members wasgathered, usually from a single family respondent,

Definition of the Family

Because NMCUES is a longitudinal survey, coveringan entire year, the important concept. of ““”longitudinalfamily was developed to deal with the facts that thecomposition of a family can change over time and thatfamilies come into and go out of existence over time.The concept of longitudinal family used in this reportis presented in detail in Appendix II. Simplified, it isas follows:

ofby

At a point in time, a family is defined as a grouppersons sharing a common household and relatedblood, marriage, adoption, or a formal foster care

6

relationship. An unmarried student 17–22 years of ageliving away from home is also considered a part ofa family.

When an initially sampled family had a change inmembership during 1980, the prechange and postchangegroups were considered the same family if and onlyif the “majority” of members of the prechange groupbecame members of the postchange group, and the“majority” of members of the postchange group hadpreviously been members of the prechange group. Forthe purpose of counting a “majority,” persons movinginto or out of the sample universe—namely, the universeof civilian noninstitutionalized persons residing in theUnited States—were omitted from the count. For exam-ple, persons who were born, or who had died, or whohad moved into or out of an institution, or into or outof the military were omitted from the count.

Standardization for Part-Year Families ,

One problem with analyzing data from a longitudinalsurvey is that some families enter and leave the surveyuniverse during the time covered by the survey. Thishas two consequences. First, the number of differentfamilies in the longitudinal universe is larger than thenumber of families that would be found in a cross-sectional survey. Second, a number of families (about12 percent in NMCUBS) did not exist for the full surveyyear (Dicker and Casady, 1984).

If each family that ever existed during the year weretreated equally as one unit, the count of families, whichwould be equal to the gross total number of distinctfamilies that ever existed during the year, would belarger than the average number of families that existedat a single point in time (the average cross-sectionalestimate). Also, if each family that ever existed duringthe year were treated as one unit, measures of healthspending and use of health care by families would notbe comparable, as some counts of family spending anduse of care would be for a whole year and some forless than a whole year.

Consequently, the following standardizing proce-dures were chosen. The population of families was time-adjusted so that, for example, a family that had existedfor only half a year was counted as only half a unit.Therefore, in this report the total number of familiesin any category represents the total number of familyyears for that category. (Alternatively, this can be consid-ered the average daily number of families in that categoryduring the year 1980.) Moreover, the counts for anyhealth behavior event were adjusted to represent annualrates for that event. For example, a family in the surveyfor half the year with $150 in total charges is representedas a half family year unit with total charges at an annualrate of $300 per year. Because these concepts are awk-ward to use in writing, families will be generally dis-cussed in the following text as if they represented oneunit each, and the expenses will be discussed as if they

were actual expenses rather than annual rates. It shouldbe noted, however, that the term, “family,” as usedin the text, means family years and that all health ex-penses are rates per family year.

The Definition of a Health Expense

Annualized family health expenses are the healthexpenses used in this report. Depending on the analysis,the expenses can be either total charges for care orout-of-pocket expenses. Ideally, these measures wouldinclude expenses for all types of health care. However,the actual analysis is limited by the type of expensedata collected in NMCUES, which did not cover alltypes of health expenses. The data used in this reportinclude expenses for the following types of health serv-ices: inpatient hospital care, inpatient physician care,outpatient hospital and emergency room care, ambulatoryphysician care, dental care, acquisitions of prescriptionmedicines, care from other independent medical provid-ers (such as chiropractors, speech therapists, faith heal-ers, and psychologists), and the acquisition of healthcare supplies and services (such as eyeglasses, orthopedicitems, hearing aids, ambulance services, and diabeticitems) .’In previous reports (Sunshine and Dicker, 1987a;Sunshine and Dicker, 1987b), the expense measure usedin this report has been labeled expenditures for “allhealth care combined.” However, this measure does notinclude expenses for nonprescription medicines, nursinghomes, and other types of long-term care institutions.

When relevant in this report, family-paid premiumsfor private health insurance or public health care coverageprograms were also added to the above list of expenses.When premiums are included, the inclusion is notedin the text.

Adjustments to the Sample

As previously pointed out, this report covers onlymultiple-person families (defined as families with anaverage family size of 1.5 members or more duringthe survey year). This is the type of family that thegeneral public and most social scientists mean whenthey use the concept “family.” (See the discussion inDicker and Casady, 1985). Moreover, ai- the reviewof the literature indicates, this social unit has not beentreated separately by most previous researchers examin-ing financially burdensome health expenses. Also, tohave included one-person families in the analysis wouldhave meant having a separate analysis for that type ofsocial unit. This would have excessively increased boththe size of the report and the amount of time neededto complete it. Thus, one-person families (defined asfamilies with an average size of less than 1.5 membersduring the survey year) were excluded from the analysisin this report.

The NMCUES sample consisted of 4,888 respondingmultiple-person families. Of these, 43 (or 0.9 percent)

7

were families with military heads. Because NMCUESwas a survey of the noninstitutionalized civilian popula-tion, another family member (Usuaily the spouse) wasimputed as the head of family in these families. Thisimputation produced many anomalies in the data (seePublic Use Data Tape Documentation: Family Data,1986, pp. 22–23). Consequently, it was decided to ex-clude these families from the analysis. This gave a basicsample of 4,845 multiple-person families that were un-equivocally representative of the civilian family popula-tion of the United States.

Another problem particular to this report resultedfrom the use of ratio measures composed of health ex-penses in the numerator and family income in the de-nominator. Some families reported either a zero incomeor a very low income (defined as an income less than$1,000 or under 20 percent of the poverty level). Asthe above type of ratio measure would be undefinedfor families with a zero income, and as a very lowreported income is probably not a good measure of theactual financial resources available to the families report-ing such incomes, some adjustment was necess~. Twotypes of adjustments have been used in the literature.The first imputes a minimum income for such families(Duan, et al., 1982). The second leaves such familiesout of the analysis (Berki et al., 1985). Each has itspositive and negative elements. This report follows Berkiand leaves these families out of the analysis when ratiomeasures are used. Therefore, when such measures areused, 21 multiple-person families with zero or very lowincomes (0.4 percent of the total) are excluded, givinga basic sample of 4,824 civilian families. It is believedthat this exclusion does not fundamentally distort theanalysis presented here.

The Two-Part Model..

Another adjustmentusing a two-part model

to the sample is the result ofrecommend~d bv Duan. et al.

(1982) for an~lyzing the determinants of ~ealth spending(or of use of health services). The first part of thismodel identifies what distinguishes populations withhealth spending (or use of health services) from popula-tions without health spending (or use of health services).The second part of the model identifies the determinantsof the amount of spending or use in populations withspending or use. This model has been used in previousNMCUES reports on family health care spending anduse of health services. (See Dicker and Sunshine, 1987;Sunshine and Dicker, 1987a; Sunshine and Dicker,1987b.)

The appropriate model to use depends on the researchquestion under examination. In the descriptive tablesin this report, Tables A through F, the research questionis”, “What is the proportion of the total U.S. familypopulation (and the proportion of.the three socioeconomicsubpopulations) that have financially burdensome healthexpenses when different measures of such expenses are

8

used?” All families are included in this analysis, andthe two-part model was not used. In contrast, in themultiple regression analyses, the research question was“What are the causes of financially burdensome healthexpenses among families with expenses?” Here the two-part model is appropriate. Following Duan, et al, (pp.20-24), the population for this second research questionwas limited to families with positive (nonzero) healthexpenses. This eliminated the need to impute dollaramounts for families with zero expenses in order toavoid having to calculate the logarithm of zero, whichis undefined. Of the 4,824 families remaining in thesample after the adjustments discussed previously hadbeen made, 181 families had zero total out-of-pocketexpenses. (Total out-of-pocket expenses, as previouslystated, includes family paid premiums.) These 181families (about 3.7 percent of the adjusted sample) wereexcluded, leaving a total of 4,643 multiple-person, civil-ian families for the regression part of the study.

The exclusion of families with zero total out-of-pocket health expenses from the regression analyses dif-ferentially affected the three socioeconomic populations,Whereas only 1 percent of older families and only 2percent of younger, better-off families were excluded,11 percent of younger, lower-income families wereexcluded. Therefore, the reader is cautioned to interpretthe regression results as being statistically valid onlyfor family populations with nonzero total out-of-pockethealth expenses. As a result, the question actually investi-gated using multiple regression was “What are the causesof financially burdensome health expenses amongfamilies with nonzero total out-of-pocket healthexpenses?”

Sampling Error

Because the statistics shown in this report are basedon a sample of families rather than on information fromall families, they are subject to sampling error. Thestandard error is a statistic that measures such errors,Standard errors for most estimates in this report havebeen computed and are presented along with theestimates.

Non-Sampling Error

In addition, estimates presented in this report aresubject to nonsampling errors such as biased interviewingand reporting, misrecording of responses, undercover-age, and nonresponse. Extensive efforts were made tominimize these errors in the data collection and dataprocessing for the survey (see Bonham, 1983).

In terms of nonsampling error, it should be notedthat data in this report are derived from informationfurnished by a survey of households—that is, “consum-ers” of health care. Data reported by providers of care,for example, in surveys of physicians, hospitals, andnursing homes, are generally different from those re-

ported by households (Sunshine, 1982). Anderson andThorne (1985) specifically compared use of health careand expenditures on health care, as reported by familiesin NMCUES, with estimates underlying the nationalhealth accounts, which are generally provider-based.They reported good agreement on total U.S. use ofhealth care and on out-of-pocket expenditures for healthcare services after coverage differences-such as theomission of military and institutionalized persons fromNMCUES-are taken into account. However, they foundan approximate 10-percent difference between the na-tional health accounts and NMCUES in total chargesfor health care services. It is likely that total charges,as estimated in this report, underestimate the true amount.Appendix II provides further information on this prob-lem. (Also see the discussion in Sunshine and Dicker,1987b, on the imputation of total charges in the sectionon nonsampling error.)

Statistical Significance and Hypothesis Testing

All frequency tables in this report show not onlyan estimate of the percent of families with various highlevels of expenses, but also an estimate of the standard

error of these percents. Where the text indicates thattwo estimates differ, the difference has been tested bya multiple t-test using the Bonferroni inequality (seeLevy and Lemeshow, 1980, p. 296) and found significantat the 0.05 level. Standard errors were computed bythe SESUDAAN computer software package (Shah,1981), which takes into account the effect of theNMCUES complex sample design upon the standarderrors of statistics estimated from its data,

This report also uses multiple regression analysisto examine the relationship between the index of finan-cially burdensome family health care expenses and ap-proximately 50 independent variables that characterizefamilies. Even after stepwise regression produced smallerprefemed models, the preferred models still had a largenumber of variables, and an adjustment was made inestimating significance at the .05 level that was similarto the adjustment made with a multiple t-test using theBonferroni inequality. Using this adjustment, signifi-cance at the 0.05 level of probability was the equivalentof significance at the 0.0029 or lower level of probabilityas estimated by the SURREGR program (Holt and Shah,1982). For more details, see Appendix I.

9

Measuring FinanciallyBurdensome Health Expenses

In the “Background” section, two basic types ofmeasures of financially burdensome health care expenseswere identified. These were measures based on the dollarvalue of expenses and measures based on ratios of thedollar value of expenses to family (or societal) income.In keeping with this perspective, this section examinessix possible measures of financially burdensome healthcare expenses: three dollar measures and three ratio meas-ures. First, the measures are defined and their signifi-cance described. Then data from NMCUES for the threepopulations of interest will be presented for eachmeasure.

3.

Six Basic Measures: Definitions

Three Dollar Measures

1.

2.

10

Total charges for health care services-Total familycharges for health care services are defined as thetotal amount billed to families for health care serv-ices, whether these amounts are paid out-of-pocket,paid by health care coverage, or go unpaid. Totalcharges are the underlying expense that families andsocieties face. To the extent that they are so highas to be burdensome, they constitute a potential finan-cial (and health) problem for both individual familiesand for modern societies. Private health insuranceand public health care programs are social mecha-nisms designed to reduce the burden of total chargeson the family. These social mechanisms convert in-frequent, large, unexpected, burdensome expensesinto expenses that are routine, smaller, and moreplanned for. As a result, health expenses becomemore manageable. However, the existence of theseprograms usually does not completely eliminate allfamily health expenses. There are still the plannedexpenses associated with premium payments as wellas unexpected, unplanned expenses that result fromthe fact that health care coverage programs rarelycover all health care costs.

Out-of-pocket expenses for health care services—Out-of-pocket expenses for health care services aredefined as amounts paid by families for health careservices that are not reimbursed by public healthcare coverage programs or by private insurance.(Such expenses exclude premiums paid for thesepublic and private health care coverage programs.)Out-of-pocket expenses for health care services are

a measure of the expensesfamilies actually bear. They

for health care thattell what the family

actually had to ‘pay, wheth;r or not it had heal;hcare coverage and regardless of what the actualtotal charges by providers were. In large part, theseare unplanned expenses. They are therefore a meas-ure of what the cost problem is if, and when, thefamily uses health care services. They do not tellwhat this cost would be if there were no publichealth care programs or private health insurance.For this value, one must look to total charges, asdiscussed in number 1 above.

Total out-ofipocket health expenses—Total out-of-pocket expenses for health are defined as the totalamount families pay from their own resources forhealth care. They include both out-of-pocket ex-penses for health care services (as described innumber 2 above) and family-paid premiums for pri-vate insurance or public health care coverage pro-grams. While it is true that premium expenses areexpected, usually voluntary, and so presumably can-not constitute a financial disaster, premium expensesare genuine health expenses and should be includedwhen measuring a family’s total health cost burden.Indeed, if premium costs are large, they can bea substantial part of a health expenditure level that,in total, is quite burdensome. Moreover, in a veryreal sense, famiIies face a trade-off between pre-miums and out-of-pocket expenses for health careservices. If families forgo insurance, their premiumsare zero, but their out-of-pocket costs for healthcare services (or risk of these out-of-pocket costs)is high. At the other extreme, families might obtainvery extensive health care coverage—for example,through a health maintenance organization (HMO),They would thereby be sure of cutting out-of-pocketcosts for care almost to zero, but the premium fortheir coverage would then be relatively large. Thus,to obtain a realistic measure of the cost burden thatfamilies actually bear, it is desifable to include bothout-of-pocket costs for health care services and fam-ily-paid premiums.

Three Ratio Measures

4. Total charges for health care services as a percentoffamily (or societal) income—The financial burden

that a family (or society) experiences at any givenlevel of expenditure depends very heavily on itsincome. For example, an expense that is a severeburden for a low-income family might be a relativelyminor burden for a higher-income family with, say,five times the income. Hence, in order to measurethe burden that health expenses represent for thesetwo types of families, it is useful to examine theratio of these expenses to family income. Forfamilies, total charges as a percent of income measurewhat a family’s financial burden would be if it hadto pay its total cost for health care services out-of-pocket without the aid of private health insuranceor public health care coverage programs. For society,this measure tells what the aggregate financial burdenof health care is for the total society when comparedwith some measure of total societal income (GNP,net national product, and so forth).

5. Out-of-pocket expenses for health care services asa percent of family income—As with the previousmeasure, the argument for this measure is that theactual financial burden that any given expense gener-ates depends heavily on income. This measure, likemeasure number 2, focuses on the often unexpectedout-of-pocket spending for health care services thata family experiences after private health insuranceand public health care coverage programs have madetheir contribution to paying bills for health careservices.

6, Total out-of-pocket expenses as a percent of familyincome—This measure is derived from measurenumber 3, total out-of-pocket expenses for healthcare. It is the ratio of total out-of-pocket expenses(out-of-pocket expenses for health care services plusfamily-paid premiums) to family income. It is ameasure of the total actual financial burden of healthcare expenses on a family. It takes into accountthe tradeoff discussed earlier between planned pre-mium costs and generally unplanned out-of-pocketcosts for health care services. In addition, this meas-ure approximates the measure used in the federalincome tax to quantify how burdensome a family’shealth costs are.

Each of these measures is valuable in identifyingfinancially burdensome costs. None is the “right” meas-ure-or the “wrong” measure—in any universal sense.Some, however, are more useful than others for specificpurposes. For example, if a study’s objective is to modelthe effects of tax law changes, it should use the ratioof total out-of-pocket expenses for health care to totalfamily income, measure 6, or a variant thereof.

What Level of Expense is Financially Burdensome?

Apart from the question of what measure to use,identifying financially burdensome health expenses,aspreviously pointed out, involves anothe~basic question:

What level of expenses is financially burdensome? Totake examples from some of the studies cited above,is the threshold of burden $5,000, or $2,000, or some-thing in between? Is it 5 percent of income, 20 percentof income, or somewhere in between-or perhaps higheror lower? One approach to answering this question (men-tioned in Berki, et al., 1985) is to look to the Federaltax law, since it is the closest thing the United Stateshas to a nationwide standard. However, this approachdoes not provide a clear answer. As of 1987, the Federalincome tax threshold for an itemized deduction for medi-cal expenses became 7.5 percent of income, but beforethat it was 5 percent of income, and only a few yearsearlier it was 3 percent of income. Therefore, it appearsthat even for a specific purpose, such as tax law, theexpense level that is burdensome is the subject of chang-ing views. In short, there is no consensus, national orotherwise, as to what level of health care expenses isfinancially burdensome.

In light of this reality, a number of the studies haveevolved a useful pragmatic approach. (See, for example,Koretz, 1982 and Berki, et al., 1985.) This approachis based on the view that financially burdensome healthexpenses are-almost by definition—uncommon; and,therefore, possible thresholds to measure them shouldbe taken from the upper part of the distribution of ex-penses. The pragmatic approach then selects multiplethresholds, widely spaced from one another, and takenfrom the upper part of the distribution of expenses.For example, Berki et al. (1985) uses a range of thresholdvalues from 5 percent of income to 20 percent of income.The proportion of the U.S. population of families in-cluded by these thresholds ranges from the highest 20percent of U.S. families in out-of-pocket health carespending to the highest 4 percent of U.S. families. Thepurpose of a widely spread set of thresholds is to presentdata and show relationships across a range of valuesthat includes what most (if not all) observers wouldchoose as a cutoff for “burdensome” if they had tochoose a single cutoff. This approach offers readersspecific data and simultaneously illustrates relationshipsacross abroad range of values of interest.

Six Basic Measures: Data Presentation

In light of the preceding discussion of six possiblealternative measures of financially burdensome healthcare expenses, it seems appropriate to present basic dataon the distribution of U.S. families in 1980 by eachof the six measures. For each measure, a set of thresholdsis used that divides families into dichotomous distribu-tions according to whether or not their health expensesreached the various thresholds. Following the pragmaticapproach just described, the thresholds are chosen fromthe upper part of the distribution of families on eachof the six measuresof healthexpenses,and are widelyspread. The data for all U.S. multiple-person families

11

and for the three socioeconomic family categories thatare of concern in this report are found in Tables Athrough F.

Three Dollar Measures

1.

,,

Total charges for health care services-The 1980annual rate of total family charges for health careservices for multiple-person families is shown inTable A. This table shows that 18 percent of allU.S. multiple-person families had total charges in1980 of $3,000 or more, 9 percent had total chargesof $5,000 or more, 3 percent had total charges of$10,OOOor more, and only 1 percent had total chargesof $20,000 or more.

Two patterns emerge from the data on thesocioeconomic categories presented in the table.First, the percent of older families with high levelsof total charges was twice or more the percent foreither category of younger families. For example,18 percent of older families had total charges of$5,000 or more in 1980 compared with 7 or 8 percentof families in the two categories of younger families.Similarly, 8 percent of older families had totalcharges of $10,000 or more in 1980 compared with2 or 3 percent of families in the younger categories.

2.

The second pattern is that there are no statisticallysignificant differences between the two socio-economic categories of younger families in the pro-portion of families at or exceeding the various ex-pense thresholds shown in Table A. This suggeststhat both categories of younger families have thesame proportions of families with high total chargesfor health care.

Out-of-pocket expenses for health care services-The1980 annual rate of out-of-pocket spending for healthcare services for multiple-person families is shownin Table B. This table shows that 16 percent ofall multiple-person U.S. families had out-of-pocketexpenses for health care services of $1,000 or morein 1980, 4 percent had out-of-pocket expenses of$2,000 or more, and only about 1 percent had out-of-pocket expenses of $4,000 or more.

Because the amounts spent out of pocket forhealth care services are much less than the amountsbilled for total charges, the thresholds chosen forTable B tend to be lower than those chosen forTable A. There is, however, one comparablethreshold, $3,000 or more. A comparison of TablesA and B for this expense threshold indicates thatthe proportion of families with an expense at or

Table A

Total charges for health care for mutiple-pereon families, by age and status relative to the poverty level: United States, 1980

Number of Total charges

Sample families $3,000 $5,000 $10,000 $20,000Age and status of multiple-person family size in thousands or more or more or more or more

Percent of families (standard error)

All . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4,845 58,135 17.5 (0.6) 9.2 (0.5) 3.3 (0.3) 0.9 (0.1)

Older . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 860 10,809 26.9 (1.7) 17.8(1 .5) 7.7 (1.0) 2.3 (0.6)Younger:

Lower income . . . . . . . . . . . . . . . . . . . . . . . . . . 1,057 13,595 15.5 (1.2) 8.0 (0.9) 2.9 (0,5) ‘0,8 (0.3)Better off . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2,928 33,732 15.4 (0.7) 7.0 (0.6) 2.1 (0,3) ‘0.5 (0.2)

NOTES: Older families are families with member(s) e5 years of age or over. Younger families are families with no member 65 years of age or over. Lower-incomefamiliesarefamilieswith incomebelow 200 percent of the povetty level. Better-off families are families with inmme of 200 percent of the poverty level or more.

Table B

Out-f-pocket expenses for health care services for multiple-person families, by age and status refstive to the poverty IevekUntied States, 1980

Number of Out-of-pocket expenses

Sample families $1,000 $1,500 $2,000 $3,000 $4,000Age and status of multiple-person family size in thousands or more or more or more or more or more

Percent of families (standard error)

All . . . . . . . . . . . . . . . . . . . . . . . . . . . 4,845 58,135 15.5 (0.6) 8.0 (0.5) 4.2 (0.3) 1.6 (0.2) 0,7 (0.1)

Older . . . . . . . . . . . . . . . . . . . . . . . . . 860 10,809 20.7 (1.6) 10.6 (1.1) 5.8 (0.7) 2.4 (0.4) 0.8 (0,3)Youngen

Lower income . . . . . . . . . . . . . . . . . . . 1,057 13,595 12.0 (1.1) 6.4 (0.7) 3.8 (0.5) 1.8 (0.4) 1.2 (0.4)Better off . . . . . . . . . . . . . . . . . . . . . . 2,928 33,732 15.2 (0.8) 7.8 (0.6) 3.9 (0.4) 1.3 (0.2) 0.5 (0.1)

NOTES: Older families are families with member(s) 65 years of age or over. Younger families are families with no member 65 years of age or over, Lower-incomefamilies are families with income below 200 percent of the poverty level. Better-off families are families with income of 200 percent of the poverty fevel or more.

12

Table D

Total charges for health care as a percent of annual income for multiple-person families, by age and status relative to the poverty IevekUnited States, 1980

Total charges of—

Number of 10 percent 25 percent 50 percent 75 percent 100 percentSample families of income of income of income of income of income

Age and status of multiple-person family size in thousands or more or more or more or more or more

Percent of families (standard error)

All . . . . . . . . . . . . . . . . . . . . . . . . . . . 4,824 57,825 29.0 (0.8) 13.0 (0.6) 5.7 (0.4) 3.3 (0.3) 2.5 (0.3)

Older . . . . . . . . . . . . . . . . . . . . . . . . . 858 10,768 43.0 (1.9) 24.7 (1.7) 12.7 (1 .3) 8.6 (1 .2) 5.6 (0.9)Younger:

Lower income . . . . . . . . . . . . . . . . . . . 1,038 13,327 47.3 (1.7) 25.3 (1 .4) 10.8 (1.1) 6.1 (0.9) 5.3 (0.8)Better off . . . . . . . . . . . . . . . . . . . . . . 2,928 33,732 17.3 (0.7) 4.4 (0.5) 1.4 (0.3) 0.5 (0.1 ) ‘0.3 (0.1)

NOTES; Older families are families with member(s) 65 years of age or over. Younger families are families with no member 65 years of age or over. Lower-incomefamilies are families with income below 200 percent of the poverty level. Better-off families are families with income of 200 percent of the poverty level or more.Excludes very low income families—those repotiing annual income under $1,000 or less than 20 percent of the povetiy level.

their income, 6 percent of multiple-person familieshad total charges that amounted to 50 percent ormore of their income, and 3 percent of familieshad total charges that equaled or exceeded 100 per-cent of their income.

A comparison among the three socioeconomiccategories into which multiple-person families weredivided for this report shows that for most of thethresholds presented in Table D, older families andyounger, lower-income families had similar distribu-tions. That is, approximately the same proportionof older families and of lower-income youngerfamilies reached or exceeded a given threshold ofpercent of total charges to income. For example,approximately 25 percent of both older families andyounger, lower-income families had total chargesfor health care that amounted to 25 percent of theirincome or more. By comparison, a much smallerpercent of younger, better-off families were foundto have reached or exceeded each threshold. Forexample, only 4 percent of these families had totalcharges that reached or exceeded the 25-percent-of-income threshold.

A comparison of Table A with Table D suggeststhat the similarities found in Table D for olderfamilies and younger, lower-income families arefound because the health and socioeconomic factorsthat affect the numerators of the ratios in TableD are different from the health and socioeconomicfactors that affect the denominators. It appears, inparticular, that the relatively high proportion of olderfamilies at each percent-of-income threshold in TableD is the result of the higher total charges for healthcare these families had (Table A). In contrast, thesimilar, relatively high proportion of younger lower-income families at each percent-of-income thresholdin Table D appears to be the result of the lowerfamily incomes of these families, since their actualdollar charges for health care were relatively low(Table A).

Table D indicates that if families had to paytheir total charges for health care out of pocket,

5.

only a minority of all families would have financiallyburdensome charges. Looking at the two lowest per-cent thresholds, from 29 percent to 13 percent ofall families could be considered to have financiallyburdensome total charges. However, these propor-tions are not miniscule and probably are much largerthan desired either by the families or society.Moreover, the burden of total charges would fallmost heavily on older families and on younger,lower-income families. For these families, the twolowest thresholds in Table D show that from 47percent to 25 percent of the families had ratios thatindicate total charges that if paid out of pocket wouldbe financially burdensome to the families.

It should be noted, however, that family behaviorwould change considerably if families had to paytheir total charges out of pocket rather than havingmuch of these charges paid by health care coverageplans. Previous research has shown that use of careand total charges would be lower if care had tobe completely paid for on an out-of-pocket basis(Newhouse et al., 1981). ‘ t.. . ..

Out-of-pocket expenses for health care services asa percent of family income—The percent of familyincome accounted for by out-of-pocket expenses forhealth care services in 1980 is shown in Table Efor U.S. multiple-person families. This table showsthat 16 percent of U.S. multiple-person families hadout-of-pocket expenses for health care services thatamounted to 5 percent or more of their income in1980, 6 percent of multiple-person families had suchout-of-pocket expenses that amounted to 10 percentor more of their income, 2 percent of the familieshad out-of-pocket expenses that amounted to 20 per-cent or more of their income, and only 1 percentof families had out-of-pocket expenses for healthcare services totalling 25 percent or more of theirincome.

A comparison of Tables D and E indicates thatthe proportion of families at a particular percent-of-income threshold in 1980 was much smaller when

15

Table E

Out-of-pocket expenses for health care sem-ces as a percent of annual income for muftiple-person families, by age and status relativeto the poverty Ievqk United States, 1980

Out-of-pocket expenses of—

Number of 5 percent 10 percent 15 percent 20 percent 25 percent

Sample families of income of income of income of income of income

Age and status of multiple-person family size in thousands or more or more or more or more or more

Percent of families (standard error)

All . . . . . . . . . . . . . . . . . . . . . . . . . . . 4,824 57,825 16.3 (0.7) 6.4 (0.4) 3.4 (0.3) 2.1 (0.2) 1.4 (0,2)

Older . . . . . . . . . . . . . . . . . . . . . . . . . 858 10,766 29.9 (1 .9) 13.3 (1 .4) 6.9 (1 .0) 4.3 (0.8) 2,9 (0.6)

Younger:Lower income . . . . . . . . . . . . . . . . . . . 1,038 13,327 26.5 (1 .4) 11.8 (0.9) 7.5 (0.8) 5.1 (0.7) 3.4 (0.6)Better off . . . . . . . . . . . . . . . . . . . . . . 2,928 33,732 7.9 (0.6) 2.1 (0.3) 0.7 (0.2) *0.3 (0.1) ‘0.2 (0,1)

NOTES: Older families are families with mambar(s) 65 years of age or over. Younger families are families with no member 65 years of age or over. Lower-incomefamilies are families with income below 200 percent of the poverty level. Better-off families are families with income of 200 percent of the poverty level or more.Excludes very low income families—those reporting annual income under $1,000 or less than 20 percent of the poverty level.

the numerator for the ratio measure was out-of-pocketexpenses for health care services than when thenumerator for the measure was total charges forhealth care. For example, 29 percent of U.S. familieshad total charges for health care that amounted to10 percent or more of their income (Table D) com-pared with 6 percent that had out-of-pocket expensesfor health care services that were this high in propor-tion to income (Table E). The difference amountsto almost a fivefold decrease in the proportion offamilies that spent 10 percent or more of their incomeon health care. At higher expense thresholds, thedecrease in the proportion of families reaching orexceeding the threshold was even greater. For exam-ple, 13 percent of U.S. families had total chargesfor health care that amounted to 25 percent or moreof income (Table D) compared with only 1.4 percentthat had out-of-pocket expenses for health care serv-ices that were this high (Table E). This is almosta tenfold decrease in the probability of having healthcare expenses of 25 percent or more of income.These large decreases in the probability of havinghigh expenses for health care can be attributed tothe general availability in the United States of privateand public health care coverage programs that con-vert high total charges for health care into moderateor low out-of-pocket expenses for health careservices.

~is phenomenon is also found for the threesocioeconomic family categories that are of interestin this report. In each category, the relative riskof reaching or exceeding a particular percent-of-income threshold was much lower in 1980 for out-of-pocket expenses for health care services than fortotal charges. The size of the difference, h~wever,varied with the particular threshold and the popula-tion category. For example, 43 percent of olderfamilies had total charges for health care thatamounted to 10 percent or more of income (TableD) compared with 13 percent that had out-of-pocketexpenses for health care service that were this high

16

(Table E). This was approximately a threefold de-crease. For younger, better-off families, the corre-sponding statistics were 17 percent (Table D) and2 percent (Table E). This was an eightfold decrease.For younger, lower-income families the relevantstatistics were 47 percent (Table D) and 12 percent(Table E), a fourfold decrease.

As the above statistics indicate, and as was previ-ously found when examining total charges as a per-cent of income, when out-of-pocket expenses forhealth care services are analyzed as a percent ofincome, both older families and younger, lower-income families are much more likely to have hada high financial burden for health care costs thanyounger, better-off families. Moreover, the propor-tion of families with high levels of burden appearsto be about the same for younger, lower-incomefamilies and for older families. This was true evenat very high ratios. For example, at the expense-to-income threshold of 20 percent or more, 4 percentof older families and 5 percent of younger, lower-in-come families spent this proportion of income outof pocket for health care services compared withless than 1percent of younger, better-off families.

It is not obvious from a comparison betweenthe dollar data in Table B and the ratio data inTable E why the above distributions are the waythey are. A higher proportion of older families hadout-of-pocket expenses for health care services of$1,000 or more and $1,500 or more than did youngerfamilies (Table B). This suggests that the high pro-portions of older families _with high expense-to-income ratios result from the higher dollar levelsof out-of-pocket expenses for health care servicesthat they have. In contrast, the high proportionsof younger, lower-income families with high ex-pense-to-income ratios appears to result from thelower incomes of these families. This appears tosupport some of the conclusions of Berki et al,(1985). On the other hand, there were no consistent,statistically significant differences between older and

6.

younger families in the proportions of families withdollar expenses of $3,000 or more and $4,000 ormore (Table B). This suggests that a simple explana-tion involving only lower income for younger, lower-income families and only higher expenses for olderfamilies is not a complete explanation for the distribu-tion found in Table E.

Total out-of-pocket health expenses as a percent of

expenses therefore runs the risk of seriously under-stating the full financial burden of out-of-pockethealth care costs on U.S. families. In absolute num-bers, for example, Table E indicates there were 3.7million U.S. multiple-person families with out-of-pocket health care expenses of 10 percent or moreof income in 1980. Table F, in contrast, yields theestimate that there were 6.9 million such families.

When Table F (showing total out-of-pocket ex-family income—Total out-of-pocket health expensesare the sum of out-of-pocket expenses for healthcare services and family-paid premiums for privateand public health care coverage(s). Statistics on thismeasure are presented in Table F. This table showsthat 29 percent of U.S. multiple-person families hadtotal out-of-pocket expenses for health care thatamounted to 5 percent or more of their income,12 percent of multiple-person families had total out-of-pocket expenses of 10 percent or more of income,4 percent of families had total out-of-pocket expensesof 20 percent or more of income, and two percentof families had total out-of-pocket expenses of 25percent or more of income.

A comparison of the distributions in Table Fand Table E indicates that the addition of out-of-pocket premium payments for health care coverageapproximately doubles the proportion of familiesfound at each of the ratio thresholds. For example,Table F shows 12 percent of families had total out-of-pocket expenses that were 10 percent or more oftheir income compared with only 6 percent offamilies in Table E. For expenses of 20 percentor more of income, Table F shows 4 percent ofU.S. multiple-person families at or above thisthreshold compared with 2 percent in Table E. Theapproximate doubling effect is also found for thethree socioeconomic categories of families found inthe tables. Clearly, the addition of out-of-pocketpremiums to other out-of-pocket health care expenseshas a sizable impact on the distribution of familiesaccording to thresholds of financial burden. To omitout-of-pocket premiums from out-of-pocket health

penses as a percent of income) is compared withTable D (showing total charges as a percent of in-come), the expense-reducing effect of private healthinsurance and public health care programs is foundto be strongly attenuated versus the effect foundwhen Table E (showing out-of-pocket expenses onlyfor health care services) is compared with TableD. For example; consider the threshold of expensesof 10 percent or more of income. Table D showsthat 29 percent of U.S. multiple-person families areat this threshold, whereas Table F shows 12 percentof families at this threshold. This is less than athreefold decrease in the proportion of families reach-ing the threshold compared with the fivefold decreasefound when” the comparison was between totalcharges (Table D) and out-of-pocket expenses forhealth care services (Table E). At the higher thresholdof expenses of 25 percent or more of income, TableD shows 13 percent of families at this thresholdcompared with 2.2 percent for Table F. This is onlya sixfold decrease in the probability of reaching thethreshold compared with the nearly tenfold decreasefound when the comparison involved only the out-of-pocket expenses for health care services shown inTable E. This attenuation of the effect of healthcare coverage programs in reducing high out-of-pocket costs is also found when the families in thethree socioeconomic categories in the tables areexamined separately. The preceding comparison in-volving Tables D, E, and F suggests that whilehealth care coverage programs do have a major effectin relieving financially burdensome out-of-pocketcosts for health care, the effect of these programs

Table F

Total out-of-pocket health expenses as a percent of annual income for multiple-person families, by age and status relative to thepoverty level: United States, 1980

Total out-of-pocket expenses of—

Number of 5 percent 10 percent 15 percent 20 percent 25 percentSample families of income of income of income of income of income

Age and status of multiple-person family size in thousands or more or more or more or more or more

Percent of families (standard error)

All . . . . . . . . . . . . . . . . . . . . . . . . . . . 4,824 57,825 29.3 (0.9) 12.0 (0.5) 5.8 (0.4) 3.6 (0.3) 2.2 (0.2)

Older, . . . . . . . . . . . . . . . . . . . . . . . . 858 10,766 54.6 (1.7) 27.2 (1.7) 14.2 (1.5) 8.2 (1.1) 5.1 (0.8)Younger:

Lower income . . . . . . . . . . . . . . . . . . . 1,038 13,327 40.7 (1.7) 20.3 (1.2) 11.6 (1.1) 7.6 (0.9) 4.6 (0.7)Better off,,..,,.,.,,..,.. . . . . . . 2,928 33,732 16.8 (0.8) 3.8 (0.4) 0.9 (0.2) 0.5 (0.1) ‘0.4 (0.1)

NOTES. Older families are families with member(s) 65 years of age or over. Younger families are families with no member 65 years of age or over. Lower-incomefamilies are familieswith incomebelow 200 percent of the poverty level. Better-off families are families with income of 200 percent of the poverty level or more.Excludes very low income families-those reporting annual income under $1,000 or less than 20 percent of the poverty level.

,.

17&

appears much weaker after family-paid premiumsfor private health insurance and public health carecoverage programs are added to the other out-of-pocket health care expenses of families.

The relationships in Table F among the threesocioeconomic populations show a finding that hasbeen consistent regardless of which of the ratio meas-ures is examined. This finding is that when healthexpenses are calculated as a percent of income, botholder families and younger, lower-income familiesmore often have a high financial burden for healthcare than younger, better-off families. For example,27 percent of the older families and 20 percent ofthe younger, lower-income families had total out-of-pocket health care expenses of 10 percent or moreof income in 1980 compared with only 4 percentof the younger, better-off families. At the thresholdof expenses of 20 percent or more of income, thecorresponding statistics were, respectively, 8 per-cent, 8 percent, and less than 1 percent. Clearly,Tables D, E, and F indicate that no matter whichratio measure is used, older families and younger,lower-income families are much more at risk forfinancially burdensome health care expenses thanare younger, better-off families. At the thresholdof total out-of-pocket expenses of 20 percent of in-come or greater, this risk is more than eight timesgreater for the former categories of families thanfor the latter category.

Although both older families and younger,lower-income families are shown in Table F to havebeen at relatively high risk of having financiallyburdensome health care expenses in 1980, the distri-butions for these two types of families diverge sig-nificantly in Table F at the lower percent-of-incomethresholds. Older families were more at risk of ex-ceeding the 5-percent threshold or the 10-percentthreshold than were younger, lower-income families.At higher thresholds, these differences disappearedand both socioeconomic categories of families wereabout equally at risk in 1980. (See the distributionsin Table F for 15 percent or more of income, 20percent or more of income, and 25 percent or moreof income.)

Table C indicates that when total out-of-pocketdollar expenses are examined, at thresholds of$1,000 and $1,500 the younger, better-off familiesare more at risk of reaching or exceeding thethreshold than are the younger, lower-incomefamilies. At higher dollar thresholds, the twocategories of younger families had about the same

proportion of families reaching or exceeding thethreshold. In contrast, Table F shows that at everypercent-of-income threshold the younger, lower-income families were at far greater risk of financiallyburdensome total out-of-pocket expenses than werethe younger, better-off families. This suggests thatamong younger families, the lower income of thelower-income families outweighs the higher dollarexpenses of the better-off families in creating finan-cially burdensome total out-of-pocket expenses.

The generally higher dollar total out-of-pocketexpenses of the older families (Table C) suggestthat it is their high health expenses that principallycreate a high financial burden index for them (TableF). However, lower income probably also plays arole here, particularly at the higher thresholds. At$4,000, the highest dollar threshold of total out-of-pocket expenses in Table C, the proportion of olderand younger families reaching or exceeding thethreshold was about the same—approximately 1 per-cent. Thus, a relatively high proportion of olderfamilies would not be expected at the highest levelsof the financial burden index (Table F) unless lowincome was also a factor. In short, as with the ratiomeasure for out-of-pocket expenses for health careservices (Table E), a simple explanation involvingonly low income for lower-income families and onlyhigh expenses for older families is not the wholeexplanation of the patterns found for the ratio meas-ure of total out-of-pocket expenses.

Stiwma~-The three ratio measures suggest that fi-nancially burdensome expenses can result from high dol-lar expenses for health care, from low family income,from combinations of these two factors, and possiblyfrom other sources. These findings support the conclu-sions of Berki et al. (1985) that families with financiallyburdensome health care costs may differ greatly fromone another in terms of their relationship to the healthcare system. Some may enter the circle of families withfinancially burdensome expenses by having memberswith devastating illnesses that result in large expenses,Others may have members with only moderate illnessand moderate expenses, but they may lack the financialresources to pay readily for even a modest amount ofhealth care. It is also probable that other aspects ofthe health care system, such as type of insurance avail-able, affect the distribution of financially burdensomeexpenses. In the next section, regression analyses arecarried out to test whether the above hypotheses remainviable and to search for other possible determinants offinancially burdensome health expenses,

18

The Determinants of FinanciallyBurdensome Health Expenses:A Regression Analysis

The Index of Financially Burdensome Health CareExpenses

If only one of the measures discussed in the precedingchapter were to be chosen as the single best measureof financially burdensome family health care expenses,it would be the ratio of total family out-of-pocketexpenses to total family, income. Because total out-of-pocket expenses include both family-paid premiums forhealth insurance coverage and out-of-pocket costs forhealth care services, this measure, as previously pointedout, takes into account the trade-off families may makebetween out-of-pocket expenses for health insurancecoverage and out-of-pocket expenses for physicians’care, hospital care, and so forth. By including income,it relates these out-of-pocket expenses of all types toa family’s ability to pay (as measured by total familyincome from all sources and from all family members).In the remainder of this report, this ratio is labeledthe index of financially burdensome family health careexpenses or, for short, the financial burden index.

Because the index is the best available measure ofthe burden of family health care expenses, an extensivestudy of the determinants of the level of the index amongU.S. multiple-person families in 1980 was undertaken.This study follows.

Methods

Multiple regression analysis was used to identifythe statistically significant determinants of the level ofthe index and to estimate the effect of different familycharacteristics on this level. Appendix I presents a techni-cal description of the analytic procedures followed. Forthe readers’ convenience, a summary of this materialfollows.

Multiple regression analysis is a statistical techniquefor estimating the effect on a single dependent variableof each of a set of independent (or causal) variables.The effect of each independent variable is estimatedwhile controlling for the effect of all of the other independ-ent variables in the set. Multiple regression analysisreadily incorporates a large number of independent vari-ables, including both continuous and categorical vari-ables, but requires assumptions to be made about the

functional form of the relationship among the variables.When multiple regression analysis using many independ-ent variables shows a statistically significant associationbetween the dependent variable and a particular independ-ent variable, the analyst may assume that a relationshipexists between the two variables and that it possiblyis a causal one, at least in the population sampled.One reason for this is that the analysis controls forthe effects of all the other independent variables in thevariable set. However, misleading results can still occur,particularly if causally important variables are omittedfrom the analysis.

The Model

The set of independent variables assumed to causechanges in the dependent variable and the functionalform of this hypothesized relationship are usually referredto as the model. Note that a regression analysis examineshow a particular set of independent variables organizedinto a particular model affect the value of the dependentvariable for a particular population. That is, the analysisis both model-specific and population-specific, althoughinferences are often made to a broader population andto other models.

Variables used. The model used in this report toanalyze the relationship between family characteristicsand differing levels of the index of financially burden-some family health care expenses was derived from ageneral conceptualization of the health care system. Thisconceptualization suggests that generalized health status,specific health conditions (illnesses), and special healthevents (births, deaths, hospitalizations, and so forth)interact with family demographic factors to produce afamily potential for the use of health care services. Theactual use of care and the final level of out-of-pocketexpenses results from a further interaction of the abovehealth factors with social factors such as socioculturaluse patterns, family economic status, prices of healthcare, general economic conditions, and family healthcare coverage. Therefore, variables were selected forthe model that were representative of the above typesof health and social factors. It was assumed that a prop-erly selected set of such variables would include somevariables that would affect levels of the index of finan-cially burdensome family health expenses.

19

Besides this generaI conceptualization of the healthcare system, the review of the literature reported aboveand the findings in the section on measurement suggestedtwo hypotheses to be tested. First, Berki et al. (1985)and Wyszewianski (1986) suggested that the level offamily income should be an important determinant offinancially burdensome health expenses. Second, Berkisuggested that high cost illnesses should also be a deter-minant. To test these hypotheses, the model includedvariables for family income and types of illnesses thatfamily members could have.

Testing hypotheses depends on controlling for vari-ables that under alternative hypotheses could be the causeof the outcome actually found. A number of the independ-ent variables in the model are of this type. Finally,the extensive literature on the importance of healthinsurance in affecting both total expenditures and out-of-pocket payments suggests that health insurance variablesbe included in the model.

The independent variables selected could easily bearranged in categories of an Andersen-Newman modelas presented in Buczko (1986). An Andersen-Newmanmodel calls for health status variables (such as perceivedhealth status), enabling variables (such as income), andpredisposing variables (such as age). The model usedhere includes these types of variables, and one of thestrengths of the NMCUES is that it allowed for theinclusion of all these types of variables in the model.The specifics of making the model operational are foundin Appendix Table I. This table gives the actual opera-tional form of the dependent variable and of each ofthe 47 independent variables used in the regressionanalyses reported here.

It should be noted that many of the variables inTable I are imperfect indicators of the underlying con-cepts that they represent. As a consequence, an actualvariable can fit into the underlying conceptual schemein more than one way. For example, the variable D9,which identifies families with a black head of family,may fit into the scheme in at least three ways. Forone, it may be a demographic factor affecting healthstatus. (Black persons, for example, have particularlyhigh rates of hypertension.) Second, if racially baseddiscrimination exists, D9 would denote a smaller supplyof care available. Third, it may mark sociocultural differ-ences in habits and preferences in the use of healthcare: Note that D9 does not represent overall economicdifference associated with the different races, for suchdifferences are controlled for by the use of income asan independent variable in the regressions.

The fictional form of the relationship hypothesizedto exist between the dependent variable, the financial

‘burden index, and the independent variables is multiplica-tive. That is, it was assumed that a specified changein an independent vaiiable will. multiply the index bya constant amount. For example, having a family memberwith heart disease might multiply a family’s index by

1.3—that is, increase it by 30 percent—as comparedwith what the index would be if no member had suchan illness. (The multiplicative model was chosen in pref-erence to an additive model for reasons detailed in Appen-dix I, which also describes how an additive model wouldwork.) This hypothesized form of the functional relation-ship between the dependent variable and the independentvariables calls for the dependent variable and severalof the continuous independent variables to be used inlogarithmic form. (Again, Appendix I explains why thisis so.) Finally, the model does not take into accountinteraction effects between variables. In order to takesuch effects into account, special variables to measureinteractions would have to be included in the model,

Procedures Followed

Regression analysis was carried out separately forthe three socioeconomic multiple-person family popula-tions focused upon in this study. As previously stated,these are older families (families with a member 65years of age of over); younger, lower-income families(families with no member 65 years of age or over andwith family income below 200 percent of the povertylevel); and younger, better-off families (families withno member 65 years of age or over and with familyincome equal to or greater than 200 percent of the povertylevel).

There were several steps in the analysis, as detailedin Appendix I. In brief, the steps were as follows, First,a small number of the initial 47 independent variableswere excluded from each regression as not suitable.For example, a variable primarily used to distinguishbetween families with all members 65 years of age orover and families with only some members 65 yearsof age or over was omitted from the regressions forthe two younger family populations, The initial exclusionleft 43 to 45 independent variables in the regressions,with the number depending on the family populationinvolved.

Next, stepwise regression was carried out using PCSAS (SAS Institute, 1985). A major reason for usingstepwise regression was to eliminate possible multicol-Iinearity (strong correlation) among variables, as severalvariables were sometimes used to operationalize a singlebasic concept. For example, four different sets of vari-ables were used separately to operationalize the conceptof family general health status. These were (1) totalfamily bed days due to illness, (2) total family workloss days due to i[lness, (3) a family-level scale of re-ported health status, and (4) a family-level scale of limita-tions in main activity. The result of the stepwise regres-sions was a much smaller preferred regression modelfor each of the three socioeconomic family populations,These preferred models contained 17 to 24 independentvariables.

20

However, PC SAS does not properly estimate var-iances of regression coefficients for samples with a com-plex survey design, such as that found in the NationalMedical Care Utilization and Expenditure Survey(NMCUES). Therefore, the three preferred models werererun as (ordinary, nonstepwise) regressions using SUR-REGR (Holt and Shah, 1982). SURREGR is a regressionprogram that appropriately estimates variances in a sam-ple with a complex design, but it cannot carry out step-wise regression analysis.

The results of the SURREGR regressions were usedto identify which independent variables were statisticallysignificant. These results are shown in detail in AppendixTables II, III, and IV. Text Tables G, H, and J (onefor each of the three family populations) show the statisti-cally significant independent variables in each preferredmodel and the estimated effect of each significant vari-able on the financial burden index. Only about one-thirdto one-half of the independent variables in the preferredmodels were found to be statistically significant.

A SURREGR multiple regression on the full 43-to 45-variable models was carried out for each familypopulation in order to check that the PC SAS stepwiseregressions did not omit a statistically significant variablebecause of their deficiencies in variance estimation. Re-sults are shown in Appendix Tables V, VI, and VII.No omissions were found by this procedure.

Findings

This section presents findings on the determinantsof the financial burden index for multiple-person familiesin the civilian noninstitutionalized population of theUnited States during the year 1980. It discusses eachof the three family populations in turn.

Older Families

Statistically significant results from the regressionanalysis for older multiple-person families are shownin Table G, with more details of the regression analysisfound in Appendix Table II. The multiple correlationcoefficient (R2) for the regression equation “forthis olderpopulation was 0.53, which means the independent vari-ables shown in Table II explained 53 percent of thevariance in the dependent variable (which was the.naturallogarithm of the financial burden index). Seven of ‘the22 variables in the preferred model were found “to bestatistically significant determinants of the finan”c~albur-den index for older, multiple-person families.

A particularly strong inverse association was foundbetween the financial burden index and family incomeamong older families. The F statistic, which measuresthe statistical significance of this association, is far largerfor family income than for any other independent vari-able. Moreover, the numerical value of the regression

coefficient for the family income variable—minus O.85—is such that a large income difference has a large effecton the level of the financial burden index. The coefficientmeans that each 1-percent increase in family incomeproduced in 1980 approximately a 0.85 percent decreasein the level of the index. This is equivalent to sayingthat if one of two otherwise similar families had twicethe income of the other in 1980, the financial burdenindex of the higher-income family would have beenabout 45 percent less than that of the lower-incomefamily. Another implication of the regression coefficientmerits noting, A coefficient of minus 0.85 means thatin 1980 for each 1-percent increase in family income,total family out-of-pocket health expenses increased byapproximate y 0.15 percent—that is, much less thanproportionately.

Older families with a head but no spouse presentduring 1980 had a financial burden index about 32 percentlower than otherwise comparable families with differenthead-spouse structures. (These latter families weremostly head-and-spouse families, but they also includedthe 4 percent of older families that had an unstablehead-spouse structure during the year.) As with all regres-sion coefficients, this is the estimated effect of this vari-able afler controlling for the influence of all other vari-ables in the regression. In particular, since family sizewas included in the regression, smaller family size isnot the explanation for the lower index found for head-only families.

Older families with a member having heart or cir-culatory disease in 1980 had a significantly higher finan-cial burden index level than similar families that didnot have any members with these diseases. The indexfor the former families averaged 24 percent greater,and it should again be noted that this is the differencethat was found after controlling for the effects of otherfactors included in the regression—in particular, aftercontrolling for general health status and hospitalization.In contrast to this statistically significant finding fora category of illness, no statistically significant effectswere found for hospitalization or for variables measuringgeneral health status.

Very low values of the financial burden index werefound for older families in 1980 with three types ofhealth care coverage: Medicaid coverage only, Medicareand other public coverage, and an unknown source ofcoverage. The families with Medicaid-only coverage typ-ically had a financial burden index level about 67 percentlower than that of otherwise comparable families notin a coverage catego~ explicitly listed in the regression.(Most of these “not listed” families had Medicare andprivate coverage.)

Most of the families in the “Medicare and otherpublic” coverage category probably had both Medicareand Medicaid coverage, since about one in six elderlypersons in the United States has Medicaid coverage inaddition to Medicare. The financial burden index levelfor families in the “Medicare and other public” coverage

21

Table G

Significant regression findings for the financial burden index for older mutiple-person families

Significant factor Effect (all other factors assumed constant)

Fami[ystructure . . . . . . . . . . . . . . . . .

Type of illness: Heart and circulatorydiseases . . . . . . . . . . . . . . . . . . . .

Family income ... . ..~ . . . . . . . . . .

Type of insurance:Medicaid . . . . . . . . . . . . . . . . . . .

Medicare and other public coverage . . . . .

Unknown coverage source . . . . . . . . . .

Region . . . . . . . . . . . . . . . . . . . . .

The regression coefficient for families with a head but no spouse was – 0.384, This implies amultiplication by 0.68. Thus, families with a head but no spouse had a financial burden indexapproximately 32 percent lower than other families. (These others were predominantly head-and-spouse families.)

The regression coefficient for families with member(s) having heart or circulatory disease was0.218. This implies a multiplication by 1.24. Thus, these families had a financial burden indexapproximately 24 percent higher than families with no members having these diseases.

The regression coefficient for family income was – 0.853. This means that each 1.O-percentincrease in family income was associated with approximately a 0.85-percent decrease in thefinancial burden index.

The regression coefficient for families with health care coverage exclusively from Medicaid was– 1.119. This implies a multiplication by 0.33. Thus, families with health care coverageexclusively from Medicaid had a financial burden index approximately 67 percent lower thanfamilies not in a coverage source category explicitly listed in the regression. (Overwhelmingly,the unlisted families had both Medicare and private insurance.)

The regression coefficient for families with health care coverage from Medicare plus other publicsources was – 2.162. This implies a multiplication by 0-12. Thus, families with health carecoverage from Medicare plus other public sources had a financial burden index approximately88 percent lower than families not in a coverage source category explicitly listed in theregression. (Overwhelmingly, the unlisted families had both Medicare and private insurance,j

The regression coefficient for families ith health care coverage from unknown coverageY.

source(s) was – 1.151!~his implies a multiplication by 0.32. Thus, families with health carecoverage from unknown coverage source(s) had a financial burden index approximately 68percent lower than families not in a coverage source category explicitly listed in the regression.(Families with unknown coverage source were predominantly families all of whose members hadpart-year or no coverage. The unlisted families predominantly had full-year coverage of allmembers, with coverage from both Medicare and private insurance,)

The regression coefficient for families residing in the South was 0.288, This implies amultiplication by 1.33. Thus, families residing in the South had a financial burden indexapproximately 33 percent higher than families living elsewhere in the U.S.

NOTES: For further details of the regression, see Appendix Table Il. The probability for the 0.05 level of significance for the preferred model for older families using themultiple F-test discussed in Appendix I is 0.0023.For an explanation of the above interpretations of the regression coefficients, see Appendix 1.

catego~ was about 88 percent lower than the level forotherwise comparable older families.

Because of coding conventions established earlierin the processing of NMCUES data, families reportedas having an unknown coverage source were generallyfamilies in which no members had full-year health carecoverage. The 3 percent of older families with an un-known coverage source had an index about 68 percentless than that for otherwise similar older families notin a listed coverage category.

Finally, the regression shows older families residingin the South in 1980 with a higher index level thanotherwise comparable families living elsewhere in theUnited States. The index for families in the South wasabout one-third higher.

Younger, Lower-Income Families

Statistically significant results from the regressionanalysis for younger, lower-income multiple-person

22

families are shown in Table H, with more details ofthe regression analysis found in Appendix Table 111.For this population of families, the independent vm’ablesshown in Table III explained 27 percent of the variancein the dependent variable (which was the naturalIogarithm of the financial burden index). In total, 9of the 17 variables in the preferred model were foundto be statistically significant determinants in 1980 ofthe level of the financial burden index for younger,lower-income families.

Two demographic variables were found to be signifi-cant. First, families with a head but no spouse presentduring 1980 had lower index levels than families withother head-spouse structures. (Most of these “other”families had a head and spouse, but they also includedthe 4 percent of younger, lower-income families withan unstable head-spouse structure.) The head-onlyfamilies had a financial burden index approximately 39percent lower than that for otherwise comparable familieswith a different head-spouse structure.

Table H

Significant regression findings for the financialburden index for younger, fower4ncome muftiple-person families

Significant factor Effect (all other factors assumed constant)

Family structure . . . . . . . . . . . . . . . . .

Age . . . . . . . . . . . . . . . . . . . . . . .

Education . . . . . . . . . . . . . . . . . . . .

Geheral health status: family work-loss daysdueto illness . . . . . . . . . . . . . . . . .

Family income . . . . . . . . . . . . . . .

Type of insurance:Medicaid . . . .. . . . . . . . . . . . .

Other public and private coverage . . . .

Other public coverage . . . . . . . . . . .

Unknown coverage source . . . . . . . . . .

The regression coefficient for families with a head but no spouse was -0.501. This implies amultiplication by 0.61. Thus, families with a head but no spouse had a financial burden indexapproximately 39 percent lower than other families. (These others were predominantly head-and-spouse families.)

The regression coefficient for age of head was 0.025. This means that each additional year ofage of the family head was associated with an increase of approximately 2.5 percent in thefinancial burden index.

The regression coefficient for education of head was 0.049. This means that each additional .year of education of the family head was associated with an increase of approximately 5 percentin the financial burden index.

The regression coefficient for family work-loss days due to illness was 0.117. This means thateach 1-percent increase in the quantity (family work-loss days + 1) was associated with anincrease of approximately 0.12 percent in the financial burden index.

The regression coefficient for family income was – 0.556. This means that each 1-percentincrease in family income was associated with approximately a 0.56 percent decrease in thefinancial burden index.

The regression coefficient for families with hedlth care coverage exclusively from Medicaid was– 1.674. This implies a multiplication by 0.19. Thus, families with health care coverageexclusively from Medicaid had a financial burden index approximately 81 percent lower thanfamilies not in a coverage source category explicitly listed in the regression. (Overwhelmingly,the unlisted families were covered by private insurance only.)

The regression coefficient for families with health care coverage from private insurance plus.public source(s) other than Medicare alone was – 0.394. This implies a multiplication by 0.67.Thus, families with health care coverage from private insurance plus public source(s) other thanMedicare alone had a financial burden”index approximately 33 percent lower than families not ina coverage source category explicitly-listed in the regression. (Overwhelmingly, the unlistedfamilies were covered by private insurance only.)

The regression coefficient for families with health care coverage solely from public sources otherthan (1) Medicare with or without other public programs or (2) Medicaid alone was – 1.972. Thisimplies a multiplication by 0.14. Thus, families with health care coverage from these “otherpublic? sources had a financial burden index approximately 86 percent lower than families not ina coverage source category explicitly listed in the regression. (Overwhelmingly, the unlistedfamilies were covered by private insurance only.)

The regression coefficient for families with an unknown health care coverage source (whichwere predominantly families all of whose members had part-year or no coverage) was – 0.564.This implies a multiplication by 0.57. Thus, families with ,an unknown health care coveragesource had a financial burden index approximately 43 percent lower than families not in acoverage source category explicitly listed in the regression. (Overwhelmingly, the unlistedfamilies were covered by private insurance only, with at least some members having full-yearcoverage.)

NOTES: For further details of the regression, see Appendix Table 111.The probabilii for the 0.05 level of significance for the preferred model for younger, lower-Income families using the multiple F-testdiscussedin Appendix I is 0.0029.For an explanation of the above interprefations of the regression coefficients, see Appendw 1.

Second, the financial burden index increased in 1980with increasing age of the family head. Each additionalyear of age of the family head increased the index byapproximate! y 2.5 percent. While this may sound likea small effect, it becomes large with substantial agedifferences. For example, in 1980afamily with a head20 years older than the head of an otherwise similarfamily would have had a financial burden index about

65 percent higher than that of the family with the youngerhead.

Among younger, lower income families in 1980,more education was associated with a higher financialburden index. Each additional year of education com-pleted by the head of the family increased the indexby about 5 percent. Thus, a family with a head whocompleted high school would have had a financial burden

23

index some 20 percent higher than that of an otherwisesimilar family with a head who dropped out of schoolafter completing only eighth grade.

General family health status had an effect in 1980on the financial burden index of younger, lower-incomefamilies. Family work-loss days due to illness was thegeneral health status variable that was found to be statisti-cally significant. However, if this variable had beenomitted from the regression, it is quite possible thatsome other variable measuring general family healthstatus would have been significant in its place. Theestimated effect of work-loss days was such that each1-percent increase in the quantity (family work-loss daysdue to illness plus 1) increased the level of the financialburden index by approximately 0.12 percent.

In contrast to the significant effect on the financialburden index that was found for general health status (asmeasured by work-loss days), no significant effect on theindex was found for hospitalization in 1980 or for thepresence in 1980 of any of the disease categories usedas variables in the regression. (These disease catego-ries included cancer and heart or circulato~ disease.)

Even within the younger, lower-income family popu-lation (all of which, by definition, had a 1980 incomebelow 200 percent of the poverty level), higher-incomefamilies had a lower financial burden index in 1980 thanlower-income families. Unlike the situation with olderfamilies, however, for younger, lower-income familiesthe statistical significance of this finding, as measuredby the F test, was not stronger than that for all otherstatistically significant variables. The regression coeffi-cient of minus 0.56 for the family income variable meansthat each l-percent increase in family income producedan approximate O.56-percent decline in the index. Thisis equivalent to saying that if one of two otherwise similaryounger, lower-income families had twice the incomeof the other in 1980, the financial burden index of thehigher-income family would have been about 32 percentlower than that of the poorer family. A regression coeffi-cient of minus 0.56 also implies that each l-percent in-crease in family income in 1980 was associated withan increase in total family out-of-pocket health expensesof approximately 0.44 percent, which was again lessthan proportionate.

For younger, lower-income families in 1980 the mostprominent features of the regression findings were thenumber of statistically significant health insurance vari-ables and their very large effect on the level of the financialburden index. No fewer than four insurance coveragevariables were significant. These insurance variables werecoverage by medicaid only, coverage by public programsother than medicare or medicaid, coverage by both privateinsurance and public programs (other than medicare),and, finally, coverage by a “source unknown.”

For the one in eight younger, lower-income familieswith coverage from medicaid only in 1980, the indexwas about 81 percent lower than for families not in alisted coverage category. (These families not listed over-whelmingly had coverage only from private insurance.)

In other words, the financial burden index for medicaid-only families was only about one-fifth of that for similarfarnilies with (predominantly) private insurance only.

Less than 1 percent of younger, lower-income familieshad coverage solely from public programs other thanmedicare or medicaid in 1980, but for the few familiesthat had such coverage the index was reduced by about86 percent. That is, their financial burden index wasonly about one-seventh of that for comparable familiesdiffering only in source of health care coverage,

About one in four younger, lower-income familieshad coverage from a combination of private insuranceand public health care coverage programs (other thanmedicare) in 1980. For these families, the index wasapproximately one-third lower than for families not ina listed health coverage category.

Finally, the 28 percent of younger, lower-incomefamilies with “source unknown” coverage had an indexabout 43 percent lower than that for families not in alisted health care coverage category. (As noted, the“source unknown” families in this report generally werefamilies with partial or no health care coverage.)

Younger, Better-Off Families

Statistically significant results from the regressionanalysis for younger, better-off multiple-person familiesare shown in Table J, with more details of the regressionanalysis found in Appendix Table IV. The R2 for theregression equation for this population was 0,23, whichmeans the independent variables shown in Table IVexplained 23 percent of the variance in the dependentvariable (which was the natural logarithm of the financialburden index). In total, 10 of the 24 variables in thepreferred model were found to be statistically significantdeterminants in 1980 of the level of the financial burdenindex for younger, better-off families.

As with older families, a particularly strong statisticalassociation was found between the financial burden indexand family income. The F statistic, which measures thestatistical significance of this association, was far largerfor family income than for any other independent variablein the regression. Again, the numerical value of the regres-sion coefficient-minus O.87—is such that a large incomedifference had a large effect on the financial burden indexin 1980. Each 1-percent increase in family income in1980 produced an approximate 0.87-percent decrease inthe index. This means that if one of two otherwise similaryounger, better-off families had twice the income of theother in 1980, its financial burden index would havebeen about 45 percent less than that of the lower-incomefamily. A regression coefficient of minus 0.87 also impliesthat each l-percent increase in family income in 1980was associated with an increase in total family out-of-pocket health expenses of approximately 0.13 percent,a much less than proportionate increase.

Three sociodemographic variables were found to bestatistically significant for younger, better-off familiesin 1980. First, increased age of the family head was

24

Table J

Significant regression findings for the financial burden index for younger, better-off multiple-person families

Significant factor Effect (all other factors assumed constant)

Age . . . . . . . . . . . . . . . . . . . . . . .

Race . . . . . . . . . . . . . . . . . . . . . .

Education . . . . . . . . . . . . . . . . . . . .

Special health event: hospitalization . . . . . .

General health status: family illness daysinbre d . . . . . . . . . . . . . . . . . . . .

Family income . . . . . . . . . . . . . . . . .

Completeness ofhealth carecoverage . . . .

Type of insurance:Other public and private coverage. . . . . .

Othar public coverage . . . . . . . . . . . .

Region . . . . . . . . . . . . . . . . . . . . .

The regression coefficient for age of head of family was 0.015. This means that each additionalyear of age of the family head was associated with an increase of approximately 1.5 percent inthe financial burden index.

Theregression coeticient for famities with ablackhead of family was –0.335. This impliesamultiplication by 0.72. Thus, families with a black head of family had a financial burden indexapproximately 28 percent lower than that of families with heads of other races.

The regression coefficient for education of family head was 0.034. This means that eachadditional year ofeducation of the family head wasassociated withan increase of approximately3.5 percent in the financial burden index.

The regression coefficient for families with one or more hospitalizations (variable HI 3) was

0.256 and the regression coefficient for number of discharges (variable HI 4) was 0.069.Together, these imply a multiplication of the financial burden index by 1.38 for families with onedischarge. Thus, families with one discharge had a financial burden index about 38 percenthigher than families with no hospitalization.

The regression coefficient for family illness days in bed was 0.069. This means that each 1-percent increase inthequanti& (family illness days in bed +I)wasassociated with an increaseofapproximately 0.07 percent in the financial burden index.

Theregression coefficient for family income was –0.868. This means that each l-percentincrease in family income was associated with approximately a 0.87-percent decrease in thefinancial burden index.

The regression coefficient for families with no members having any health care coverage was– 0.529. This implies a multiplication by 0.59. Thus, families with no member having any healthcare coverage hadafinanciai burden index approximately4l percent lower than families not inacoverage completeness categoy explicitly listed in the regression. (Families in the unlistedcategories predominantly had all members with full-year coverage.)

The regression coefficient for families with health care coverage from private insurance pluspublic source(s) other than Medicare alone was –0.228. This implies amultiplication by O.80.Thusj families with these other public and private coverage mixes had a financial burden indexapproximately 20 percent lower than families not in a coverage source category explicitly listedin the regression. (Ovewhelmingly, theuntisted families were covered bypflvate insuranceonly.)

The regression coefficient for families with health care coverage solely from public sources otherthan (l) Me&care with orwithout other public programs or(2) Medicaid alone was –0.893. Thisimplies a multiplication by 0.41. Thus, families with health care coverage from these “otherpublic” sources had a financial burden index approximately 59 percent lower than families not inacoverage source category explicitly listed in the regression. (Overwhelmingly, the unlistedfamilies were covered by private insurance only.)

The regression coefficient for families residing in the South was 0.273. This implies a multi-plication by 1.31. Thus, families residing in the South had a financial burdan index approxi-mately31 percent higher than families residing elsewhere inthe U.S.

NOTES: For futiher details of theregression, see Appentix Table lV. TheprobaMlity of the 0.051evel ofsignificance forthepreferred model foryounger, betier-offfamilies using the multiple F-test discussed in Appendix I is 0.0021.For an explanation of the above interpretations of the regression coefficients, see Appendix 1.Variable numbers such as HI 4 refer to variables in Appendix Table 1.

associated with a higher financial burden index level. Hospitalization in 1980 had a significant effect onEach l-year increase in age produced an approximate the financial burden index for younger, better-off1,5-percent increase in the index. Second, families with families. Ahospitalization increased theindex by abouta black head of family had a relatively low financial 38 percent relative to what it would have been for familiesburden index. It was about 28 percent lower than that with no hospitalization. One generalized health statusfor comparable families with a head of family of another variable, family illness days in bed in 1980, also hadrace. Finally, education of the family head was associated a statistically significant effect on the financial burdenwith a higher index level, with each additional year index for younger, better-off families. The estimatedof education of the head increasing the index by about effect of bed days in 1980 was that each 1-percent in-3.5 percent. crease in the quantity (annual family bed days plus 1)

25

increased the financial burden index by approximately0..07 percent. In contrast, none of the variables involvingspecific categories of illness was statistically significant.

Three health insurance variables were found to bestatistically significant. These insurance variables were“no health care coverage, “ “coverage by private insur-ance and public sources (other than medicare),~’ and,finally, “coverage only by public sources other thanmedicare or medlcaid.” For the 2 percent of younger,better-off families in 1980 that had no health care cover-age, the financial burden index was about 41 percentlower than for otherwise comparable families (most ofwhom had full year health care coverage for all familymembers). For the one in seven younger, better-offfamilies with health care coverage from a combinationof private insurance and public programs (other thanmedicare) in 1980, the financial burden index was about20 percent lower than for otherwise comparable familiesnot in a coverage source category explicitly listed inthe regression. (These “not listed” families overwhelm-ingly had coverage solely from private insurance.) Fi-nally, less than 1 percent of younger, better-off familieshad coverage only from public programs other than medi-care or medicaid, but the financial burden index of thesefamilies was about 59 percent lower than that of otherwisecomparable families.

One regional effect was found. Younger, better-offfamilies residing in the South in 1980 were found tohave values on the index about 31 percent higher thanfor similar families living elsewhere in the United States.

Discussion

The financial burden index, it should be recalled,is the ratio of total family out-of-pocket health expensesto total family income. Total family out-of-pocket healthexpenses include both out-of-pocket expenses for health

~care and family-paid premiums for health care coverage.Separate regressions were carried out for three familypopulations that, combined, represented the whole

multiple-person family population of the United Statesin 1980. These three populations were older multiple-person families (those with a member 65 years of ageor over); younger, lower-income multiple-person ~families (those with all members under 65 years of ageand with income below 200 percent of the poverty level);and younger, better-off multiple-person families (thosewith all members under 65 years of age and with incomeof 200 percent of the poverty level or more).

Explanatory Power ~

R2, the multiple correlation coefficient squared, isa measure of the overall explanatory power of the entireregression. R2 is equal to the proportion of the variancein the dependent variable that is explained by all theindependent variables in combination. In order to providea better understanding of the relative explanatory powerof the regression equations reported in Tables II, III,and IV, R2 for these equations is compared here withthe R2 reported in other, similar regression studies. Asshown in the first column of Table K, R2Sin the regres-sions for the financial burden index ranged from 0,23to 0.53, depending on the socioeconomic family categoryinvolved.

This is a relatively high R2 compared with that re-ported in most studies. For example, three recent papersthat use NMCUES data or data from the similar 1977National Medical Care Expenditure Survey (NMCES)in regression equations similar to those presented herereport an R2 of 0.04 to 0.27 (Farley, 1986), 0.31 (Taube,Kessler, and Burns, 1986), and 0.18 to 0.20 (Buczko,1987). These studies, however, differ from this reportin that they deal with individuals, not families, withphysician visits, not total health care, and-except forthe Buczko study—with number of visits, not spending.

On the other hand, an analysis exactly parallelingthat described in this report, but using total chargesfor health care instead of the financial burden indexas the dependent variable, obtained higher R2S(Sunshineand Dicker, 1987c). As the second column in Table K

Table K

Comparison of mutiple correlation coefficients squared, by dependent variable and age and famify status relative to the poverty level

Dependent variable

Index of financially Index of financiallyburdensome family Total family burdensome family

health expenses health charges health expensesAge and status of family (preferred model)’ (preferred model)l’2 (full model)f

Older . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0.53 0.72 0.54

Younger:

Lower income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0.27 0.60 0.28Better off . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0.23 0.57 0.24

‘Dependent variable is natural logarithm of stated “statistic.‘Regression results not shown in this report.

NOTES: Older families are families with member(s) 65 years of age or over. Younger families are families with no member 65 years of age or over. Lower-incomefamilies are families with income below 200 percent of the poverty level. Better-off families are families with income of 200 percent of the pove~ level or more,

26

shows, R2S in the regression analyses of total chargeswas 0.57 to 0.72, depending on the family populationinvolved. Thus, factors not included in (or poorly meas-ured by) the 47 independent variables used in this reportaccount for more of the variance in the financial burdenindex than they do for total family charges for healthcare.

It is also interesting to compare the R2S of twoalternate versions of the regressions for the financialburden index. Preferred models with 17 to 24 indepen-dent variables are the source of the findings reportedand discussed in this chapter and are shown in AppendixTables II, III, and IV. However, full models with 43to 45 independent variables were also run using SUR-REGR, and are shown in Appendix Tables V, VI, andVII. (See Appendix I for more information on the twotypes of models.) The third column of Table K showsthe R2 for the full models. For all three family popula-tions, it is only 0.01 higher than the R2 for the preferredmodel. This is a typical finding for preferred modelsdeveloped with stepwise regression (as the models pre-sented here were), and shows that very little explanato~power is lost by using the preferred models rather thanthe full models.

Interestingly, for both the financial burden indexand total family charges for health care, the independentvariable sets used in this report account for more ofthe variance among older families than among youngerfamily populations. This suggests that factors not in-cluded in the regressions (or poorly measured by theincluded variables) are less important for determininglevels of these two health cost measures among olderfamilies than among younger families. Why this shouldbe true is not obvious.

Individual Variables

The regressions indicate a very strong role for familyincome as a determinant of the financial burden index.Income is the only independent variable statistically sig-nificant for all three family populations, and for twoof these populations its statistical significance, as meas-ured by the F statistic, is stronger than that of anyother statistically significant independent variable. Thisfinding accords with the Berki et al. (1985) findingthat one large population of families with a high levelof the financial burden measure used in that study iscomposed of families with low incomes. Their out-of-pocket expenses for health care services may not beparticularly large in dollar terms, but they are largerelative to the small incomes of the families. This findingmeans that a family’s ability to pay, as measured byincome, is important in determining the financial burdenthe family faces from health costs. It indicates that meas-ures of health costs that include an ability-to-pay factor(for example, the three ratio measures discussed in thesection “Measuring financially burdensome health ex-

penses”) are likely to show quite different patterns thanmeasures that assess health costs solely in dollar terms,without considering differences in ability to pay. How-ever, it should not be surprising that family incomeis so prominent an explanatory variable. Family incomeis the denominator of the financial burden index and,thus, family income helps determine the index’s level.

The regression coefficients for the family incomevariable show an interesting pattern. For older familiesand for better-off younger families, the coefficient indi-cates that for each 1-percent increase in income, totalout-of-pocket expenses for health care increase by about0.15 percent. In contrast, for lower-income youngerfamilies, total out-of-pocket expenses increase by about0.44 percent for each 1-percent increase in income. Thereare at least two plausible explanations for this difference.One is that total health spending by lower-incomefamilies is severely constrained by their low incomeand so rises rather rapidly as their income increases.Another explanation is that as income increases for lower-income families, these families no longer receive charitycare; and/or they lose the benefits of Medicaid and otherpublic programs that require very small or no out-of-pocket payments. Thus, their total out-of-pocket ex-penses increase relatively rapidly with increasing incomeeven though their total charges may not. The regressionanalyzing the determinants of total charges for lower-in-come younger families (Sunshine and Dicker, 1987c;regression not shown) supports the second of these twoexplanations.

One other point about income should be noted. Be-cause income is in the regression as an independentvariable, the regression estimates of the effects of otherindependent variables are made with the effect of incomecontrolled. Thus, the effects the regressions show forother variables are not the result of a confounding effectof income. This means that other variables affect theindex only by their effect on the numerator, which isthe family’s level of total out-of-pocket health expenses.While this point sounds obvious, it is important andeasily overlooked.

Consider, for example, the statistically significanteffect found for education of the family head amongboth of the younger family categories. Because the re-gressions control for the effects of income (and the effectsof the other independent variables in the models), itis reasonable to conclude that a genuinely causal effectof education has been identified here. Quite possibly,the effect of education found in this study operatedthrough differences in family valuation of health andhealth care. In contrast, an analysis that showed anapparent effect of education, but did not control forincome, could well be finding only a spurious effect,for education and income are strongly correlated.

Age of the family head, like education, is significantin the two younger family populations but not amongolder families. The effect of age probably results fromthe well-known poorer health, greater risk for costly

27

illness, and greater use of health care characteristic ofincreasing age. The health variables used in the regres-sions (measures of generalized health status, specificillness, and so forth) probably do not fully capture thisphenomenon.

The absence of a statistically significant effect ofage among the older family population may well resultfrom a limited variation in the age distribution amongthe members of older families. These families predomin-antly have heads 65 years of age or over, Wd fewof the heads or members are more than 20 years olderor younger than that. Moreover, the older families amongmultiple-person families tend to be among the youngerfamilies when all older families (both multiple-personand one-person) are taken into account. This wouldfurther restrict the age distribution among these familiesand reduce possible variation in the index associatedwith age.

A similar situation probably explains why, amonglower-income younger families, income is not far morestatistically significant than the other statistically signifi-cant independent variables. Among lower-incomeyounger families, variation in the income distributionis also quite limited. (The main limit is the 200 percent-of-poverty cutoff that defines the income top of thisfamily category, but very low-income families are alsoexcluded.)

The relatively modest role of hospitalization in theregressions is notable. Hospitalization is statistically sig-nificant for only one population—younger, better-offfamilies—and for that population one hospitalization wasfound to increase the level of the index by about 38percent above that for otherwise similar families withno hospitalization. In contrast, one descriptive studythat did not control for other factors (Kovar, 1986)showed elderly persons. with hospitalization having, onaverage, total charges for health care of more than 1,000percent of the total charges for elderly persons withno hospitalization. The regressions (previously men-tioned) for total family charges for health care that coverthe same three populations analyzed here and that, likethe present study, control for the effects of many vari-ables, also find an effect of hospitalization that is muchlarger than 38 percent (Sunshine and Dicker, 1987c;regressions not shown).

Given the difference between the findings of theregressions in this report and those of the parallel regres-sions for total charges, there seems to be a clear explana-tion for the modest role hospitalization plays in determin-ing financially burdensome health expenses amongfamilies. Inpatient hospital care is particularly well in-sured and so is much less important in total out-of-pocketexpenses than in total charges. Strong support for thisconclusion is provided by a comparison between totalcharges and out-of:pocket expenses for various typesof health care. Such a comparison is readily made byusing parallel data from two expenditure reports. One(Sunshine and Dicker, 1987a) deals with family out-of-

pocket expenses, and the other (Sunshine and Dicker,1987b) presents corresponding data on total familycharges. For multiple-person families, these sourcesshow that only 8 percent of the total charges for inpatienthospital care were paid out-of-pocket in 1980 comparedwith out-of-pocket expenses ranging from 20 to 65 per-cent of total charges for all the other types of healthcare services examined.

The lowered level of the index characteristic offamilies with incomplete or no health care coverageis puzzling. It is found (in one form or another) forall three socioeconomic populations in the study, (Recallthat families with “coverage source unknown” are gener-ally families with incomplete or no coverage.) However,’this insurance effect i,sthe reverse of what the literaturegenerally reports. The normal pattern is for limited orabsent coverage to lead to reduced total charges becauseof the reduced use of health care resulting from thehigh out-of-pocket payments for health care that are re-quired when coverage is limited or absent (Newhouseet al., 1981). Thus, a high index, not a low index,would be expected for families with incomplete or nocoverage. There is some evidence that persons withoutinsurance are disproportionately young persons in goodhealth who, because of their youth and good health,are likely to experience relatively low total charges forhealth care (Kaspar, Walden, and Wilensky, 1980;Wilensky and Walden, 1981). However, that does notseem to be a likely explanation of the findings of thisstudy, since the regressions control for health statusand age.

The relatively high levels of the financial burdenindex found in the South is a regional effect not previ-ously noted in the/literature. However, the phenomenonseems real. It is found for two population groups (olderfamilies and younger, better-off families), and the regres-sions in this study control for a large number of variables,making spuriousness unlikely, In principle, four explana-tions are possible. First, total charges for health caremay be relatively high in the South. Second, the propor-tion of total charges that have to be paid out-of-pocketmay be particularly high there. Third, families may haveto pay an unusually large part of the premiums for healthcare coverage; and, fourth, incomes may be relativelylow in the South. As indicated above, the fourth explana-tion is ruled out because the regressions control forincome. The first explanation is eliminated by the regres-sions for total charges (Sunshine and Dicker, 1987c;regression not shown), as these regressions do not showtotal charges to be high in the South. Thus, it seemsthat southerners have to pay a particularly large portionof health care cost out-of-pocket-either as out-of-pocketpayments for health care services or as family-paid pre.miums for health care coverage. As most health carecoverage in the United States is employment-based, andas this is particularly true for younger, better-offfamilies-one of the family categories that shows anelevated financial burden index in the South—it is reason-

28

able to conclude that employment-provided health carecoverage is relatively poor in the South. Among olderfamilies, however, the explanation is probably partiallydifferent. Medicaid coverage, for example, may be morerestrictive, so that older families in the South may haveto pay a larger amount out of pocket for health carethan similar families in other regions of the United States.

Head-only families were found to have a significantlylower index than head-and-spouse families among botholder families and younger, lower-income families. Thisfamily structure effect is difficult to explain, particularlybecause it is not found among younger, better-off familiesor in regressions for total family charges for health care.It is not a matter of smaller family size, the presenceof children, or coverage by Medicaid among head-onlyfamilies, as these factors are controlled for.

Finally, there is the lowered level of the index amongyounger, better-off families with a black head of family.This race-related finding should be regarded as likelyto be a genuine one—and possibly a sign of an importantsocial problem—because the regressions in this studycontrol for many possibly confounding variables suchas education, health status, family structure, income,and location of residence. Moreover, a similar findingfor the same family population was obtained in the regres-sions for total family charges for health care. In contrast,statistical associations with race that are found in analysesthat do not control for possible confounding factors—forexample, many tabular analyses-may only be artifactsof these confounding factors.

Race-related differences were not found among eitherolder families or younger, lower-income families. Thiswas true of regressions for both the financial burdenindex and for total family charges for health care. Older

families and younger lower-income families are usuallyregarded as more vulnerable than other family popula-tions and certainly have been a focus of attention forgovernmental health care coverage programs. The ab-sence of racially caused differences among these familycategories is a positive sign.

Patterns Among Variables

Perhaps the most important aspect of the findingsof this report is the wide variety of factors that arefound to be significant determinants of the level of thefinancial burden index. The study analyzed the effectsof family variables that can be’ classified into sevencategories: demographic, sociocultural, specific illnessesand special health events, general health status,economic, health insurance, and geographic. For all threepopulations analyzed, variables in at least five of theseseven categories were found to be significant. Thus,an important conclusion of the study is that the determi-nants of the level of the index are numerous and ofseveral types.

One way to assess the importance of the findingsfrom these regression analyses (other than by significancelevels and the size of regression coefficients) is toexamine how the statistically significant variables in theregressions were distributed among the three familypopulations. Table L shows that 16 ~f the variableswere significant in one or more of the three familysocioeconomic populations. However; only one variablewas statistically significant in all three populations, eightother variables were statistically significant in two ofthe populations, and another seven were statistically sig-nificant in only one population. Table L thus gives one

Table L

Statistically signitioant variables from a set of regressions on the index of financially burdensome family health care expenses arrangedby the number of family socioeconomic populations in which each variable was statistically significant

Statistically significant in-

3 2 1Variable populations populations population

Family income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, L, BHead-spouse structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, LCoverage by Medicaid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . O,L.Source ofhealth carecoverage unknown . . . . . . . . . . . . . . . . . . . . . . . ..’.... 0, LOther public coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L,BOther public and private coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L,BAgeofhead of family . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L, BEducation ofheadoffamily . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L,BRegion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, BMedicare andother public coverage.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0Heart andcirculatoty disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0Family work-loss days . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LFamily illness days in bed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BOneor more hospitalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BRaceofhead of family . ’. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BCompleteness ofhealth care coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B

NOTE: O = Older familiesL = Younger, lower-income familieeB = Younger, better-off families

29

measure of the importance of certain variables in affect-ing values of the financial burden index. This measureis the universality of the effects of the variables in theU.S. multiple-person family population. As noted, fromthis perspective family income is the most importantvariable because it is a statistically significant determin-ant of the index among all segments of the U.S. multi-ple-person family population.

Second in apparent importance are the eight variablesthat were found to determine values of the index amongtwo socioeconomic segments of the U. S. multiple-personfamily population. These eight variables were head-spouse structure, age of the head, education of the head,geographic region of residence, and four health carecoverage variables. Clearly, source of health insurancecoverage emerges from this list as a very importantgeneral category for determining levels of the financialburden index, as four of the eight variables are in thiscategory.

Further evidence of the importance of health insur-ance variables is the large difference they make in thelevel of the financial burden index. Public coverage(other than Medicare alone) usually reduces the indexby two-thirds or more from what it otherwise wouldbe.

Third in apparent importance are the seven variablesthat were significant for only one family population.These involve the race of the head, two insurance vari-ables (reinforcing the statement above that insuranceis very important for determining values of the index),and a set of four variables that more or less directlymeasure the family’s need for health services (presenceof heart or circulatory disease, work 10SSdays due toillness, illness days in bed, and hospitalization). Becausethese last four variables all have to do with health,

it can be concluded that health factors are significantin their effect on the financial burden index, but thattheir effect is scattered and relatively diffuse.

Indeed, the regressions for the financial burdenindex, as Table L shows, are conspicuous in the relativelysmall role of health variables. (These include generalizedhealth status, illness, and special health event variables.)The limited importance of health factors becomes particu-larly clear from a comparison of Table L with TableM. Table M presents a summary of the results of theregressions that used total family health charges as thedependent variable (Sunshine and Dicker, 1987c; regres-sions not shown). These regressions for total chargescover the same three family socioeconomic populationsand start from the same 47 independent variables asthe regressions summarized in Table L, However, inTable M, generalized health status, illness, and specialhealth event variables are major determinants of thedependent variable, total family charges for health care.Of the 15 variables in Table M, 8 come from thesecategories. By comparison, only 4 of the 16 variablesin Table L, which deals with the financial burden index,are of this type. Moreover, in the regressions for totalfamily charges for health care (Table M), six of theeight health variables were significant in two or morepopulations, with two of them being significant in allthree populations. In contrast, in the regressions forthe financial burden index (Table L), no health variablewas significant in more than one population. It appears,therefore, that health status (in the broad sense) of familymembers is the major determinant of total family healthcare charges, but that the major determinants of familyfinancially burdensome health expenses are family in-come and type and completeness of health insurancecoverage.

Table M

Statistically significant variables from a set of regressions on total family charges for health care arranged by the number of familysocioeconomic populations in which each variable was statistically significant

Statistically significant i~

3 2 iVariable populations populations population

Hospitalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, L, BFamily illness days in bed...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, L, BCancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, LHeart and circulatory disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, BAccidents, poisonings, and injuries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, BFamily income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0, BPerceived health status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L, BRegion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L, B

Completeness ofhealth carecoverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L, BLimitation in major activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LFamily work-loss daysdueto illness . . . . . . . . . . . . . . . . . ..- . . . . . . . . . . . . BPresence of child . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BAgeofhead of family . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B

Raceofhead of family . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BEducation ofhead of family . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B

NOTE: O = Older families ,L = Younger, lower-income familiesB = Younger, better-off families

30

This finding seems to be in partial conflict withthe conclusion of Berki et al. (1985) that families withcostly illnesses are one of the two major populationsof families with high proportions of family income con-sumed by out-of-pocket expenses for health care services.Although Table L shows some role for such illnessesas determinants of the financial burden index, it is asmall role, particularly when compared with the roleof illnesses as determinants of total family charges forhealth care (Table M). Quite likely, the explanationof the relatively small role of illness vtiables (and otherhealth variables) as determinants of the financial burdenindex is the same as the explanation given above forthe relatively small role of hospitalization. Health carecoverage probably pays for a relatively large percentof the cost of major illnesses, making their role in out-of-pocket expenses (which is the numerator of the financialburden index) relatively small.

Concluding Remarks

Table L indicates that different variables from themodel were found to be associated with the financialburden index in a statistically significant manner in dif-ferent socioeconomic populations. Although, withhindsight, some of the associations are obvious, othersare not. Some of the obvious associations are Medicarecoverage with the older population, Medicaid coveragewith lower-income populations (both older and lower-income younger families), and family work-loss dayswith a younger population (but, surprisingly, not withboth younger populations). What these associations

suggest is that the effects on the financial burden indexof the variables found to be statistically significant arenot universal, but rather are specific to particularsocioeconomic populations. To some extent, then, policyaimed at alleviating financially burdensome family healthcare expenses may have to encompass different solutionsfor different populations.

It should also be noted that there were a numberof variables in the preferred models that were not statisti-cally significant in any of the three socioeconomic familypopulations of interest. Whether these variables wouldbe statistically significant in other populations or witha larger sample cannot be assessed at this time. Moreover,it should be noted that a single regression analysis thatencompassed the entire U*S. multiple-person familypopulation might produce different findings about therelative importance of variables than do the three separateregressions reported here, So might an analysis that didnot involve the exclusions found in this section on theregressions. (The most important of these exclusionsare families with zero or very low income, familieswith zero as their total out-of-pocket health expense,one-person families, institutionalized persons, and ex-penses for long-term Carei)

Nonetheless, the preceding discussion stronglysuggests that to succeed, policies to alleviate financiallyburdensome health care expenses among U.S. multiple-person families must concentrate on improving healthcare coverage and/or on increasing income. These arethe two types of factors that this analysis has shownare the most important determinants of families’ financialburden for health care costs.

31

References

Aday, L. A., Fleming, G. V., and Andersen, R.: Access to medicalcare in the U.S.: Who has it, who doesn’t. Research Series.No. 32. Center for Health Administration Studies, University ofChicago. Chicago. Pluribus Press, 1984.

Andersen, R., and Benham, L.: Factors Affecting the RelationshipBetween Family Income and MdIcal Care Consumption, in H. E.Klarman, ed., Empirical Studies in Health Economics. Baltimore.The Johns Hopkins Press, 1970.

Anderson, J. M., and Theme, E.: Estimates of Aggregate PersonalHealth Care Expenditures in 1980. Comparison of tbe National HealthAccounts and the National Medical Care Utilization and ExpenditureSurvey Data. Working Paper No. 30, Division of Health InterviewStatistics, National Center for Health Statistics, Hyattsville, Md.,1985.

Arnold, S. F.: Theory of Linear Models. New York. John Wileyand Sons, 1981.

Berkl, S. E.: A look at catastrophic medical expenses and the poor.Health A~airs 5(4): 138–145, W]nter 1986.

Berki, S. E., .Wyszewianski, L., Magilavy, L. J., et al.: Familieswith high out-of-pocket health service expendhures relative to theirincome. Flrral Report on Phase II of Contract No. 233–8 1–3022.Rockville, Md. National Center for Health Services Research andHealth Care Technology Assessment, Public Health Service, Jan.30, 1985.

Binder, D. A.: On the variances of asymptotically normal estimatorsfrom complex surveys. Int.StatisticalRev.51 :279-292,1983.

Bonham, G. S.: Procedures and questionnaires of the National Medi-cal Care Utilization and Expendhure Survey. National Medical CareUtilization and Expenditure Survey. Series A, Methodological ReportNo. 1. DHHS Pub. No. 83-20001. Public Health Service.Washington. U.S. Government Printing Office, Mar. 1983.

Buczko, W.: Physician utilization and expenditures in a medicaidpopulation. Health Care Financ. Rev. 8(2):17–26, Winter 1986.

Campbell, D. T., and Julian, C. S.: Experimental and Quasi-Experi-mental Designs for Research. Chicago. Rand McNally College Pub-lishing Co., 1980.

Catastrophic Illness Expenses. Department of Health and HumanServices report to the President. U.S. Department of Health andHuman Services, Nov. 1986.

Cox, G., Parker, A. E., Sweetlarrd, S. S., et al.: Imputation ofMissing Item Data for the National Medical Care Utilization andExpenditure Survey. Prepared for the National Center for HealthStatistics under Contract No. HRA–233-79-2032. Research TrianglePark, N.C. Research Triangle Institute, 1982.

Cox, G., and Sweetland, S. S.: Imputation of Attribution-RelatedMissing Data for the National Medical Care Utilization and Expendi-ture Survey- Prepared for the National Center for Health Statisticsunder Contract No. HRA–233-79–2032. Research Triangle Park,N.C. Research Triangle Institute, 1982.

Dicker, M.: Health care coverage and insurance premiums of families,United States, 1980. National Medical Care Utilization and Expendi-ture Survey. Preliminary Data Report No. 3. DHHS Pub, No. 83-20000. Public Health Service. Washington. U.S. Government Print-ing Office, May 1983a.

Dicker, M.: Panel Surveys and Family Level Measures, Problemsand Solutions, in 1983 Proceedings of the American Statistical Associ-ation. Social Statistics Volume. Washington. American StatisticalAssociation, 1983b.

Dicker, M.: Demographic Analysis and a General Theory for Measur-ing Social Change in Organizations. The Case of the Family. Paperpresented at the annual meeting of the Southern Regional Demo-graphic Group. Orlando, Fla., Oct. 19, 1984.

Dicker, M., and Casady, R. J.: A Reciprocal Rule Model for DefiningLongitudinal Families for the Analysis of Panel Suwey Data, in1982 Proceedings of the American Statistical Association, SocialStatistics Volume. Washington. American Statistical Association,1982.

Dicker, M., and Casady, R. J-: Empirical Findings on the Distributionand Construction of Longitudinal Families, Part I, Modeling theUniverse, Inscope Changes, and Static and Dynamic Families, in1984 Proceedings of the American Statistical Association. SurveyResearch Methods Volume. Washington. American Statistical As-sociation, 1984.

Dicker, M., and Casady, R. J.: An Introductory Discussion of Issuesin the Methodology of Longitudinal Family Surveys, Paper presentedat the All Day Measurement Workshop of the Family Health Sectionat the annual meeting of the National Council on Family Relations.Dallas, Texas, 1985.

Dicker, M., and Sunshine, J. H.: Family Use of Health Care, UnitedStates, 1980. National Medical Care Utilization and ExpenditureSurvey. Series B, Descriptive Report No, 10. DHHS Pub, No, 87-20210. Hyattsville, Md. National Center for Health Statistics, PublicHealth Service, 1987.

Duan, N., Manning, W. G., Jr., Morns, C. N., et al,: A Comparisonof Alternative Models for the Demand for Medical Care. ContractNo- R–2754-HHS. Santa Monica, Calif. Rand Corporation, Jan.1982.

32

Farley, P. J.: Private insurance and public programs, Coverage ofhealth services. National Health Care Expenditures Study. Data Pre-view 20, DHHS Pub. No. (PHS) 85–3374. Rockville, Md. NationalCenter for Health Services Research and Health Care TechnologyAssessment, Mar. 1985.

Farley, P. J.: Hospital and ambulatory services for selected illness.Health Serv. Res. 21(5):587-634, Dec. 1986.

Feldstein, P. J.: Health Care Economics, 2d ed. New York. JohnWiley and Sons, 1983.

Hershey, J. C,, Luft, H. S., and Glanaris, J. M.: MaHng senseout of utilization data. Med. Care 13(10):838–852, Oct. 1975.

HoIt, M. M., and Shah, Et. V.: SURREGR, Standard Errors ofRegression Coeflcients From Sample Survey Data. Research TrianglePark, N.C. Research Triangle Institute, Apr. 1982.

Kaspar, J. A., Walden, D. C., and Wllensky, G. C.: Who arethe uninsured? National Health Care Expenditures Study. Data Pre-view 1. Hyattsville, Md. National Center for Health Services Re-search, Public Health Service, 1980.

Koretz, D.: Catastrophic Medical Expenses: Patterns in the Non-Elderly, Non-Poor Population. Washington, D,C. CongressionalBudget Office, Dec. 1982.

Kovar, M. G.: Expenditures for the medical care of elderly peopleliving in the community in 1980. Milbank Mere. Fund Q. 64( I):1OL132,1986.

Landis, J. R., Lepkowskl, J. M., Eklund, S. A., and Stehouwer,S. A.: A statistical methodology for analyzing data from a complexsurvey, the first National Health and Nutrition Examination Survey.Vital and Health Statistics. Series 2, No. 92. DHHS Pub. No (PHS)82–1366. National Center for Health Statistics, Public Health Service.Washington. U.S. Government Printing Office, Sept. 1982.

Levy, P. S., and Lemeshow, S.: Sampling for Health Professionals.Belmont, Calif. Lifetime Learning Publications, 1980.

Mallows, C. L.: Some comments on C(p). Technometrics 15:661–675,1973.

McCarthy, P. J.: Replication, an approach to the analysis of datafrom complex surveys. Vital and Health Statistics. Series 2, No.14. DHEW Pub. No. (PHS) 79–1269. National Center for HealthStatistics, Public Health Service. Washington. U.S. GovernmentPrinting Office, Apr. 1966.

McMillen, D. B., and Herriot, R.: Towards a Longitudinal Definitionof Households. SIPP Working Paper Series No. 8402. U.S. Bureauof the Census. Washington, D.C., 1984.

Moser, B., Whitmore, R., Frick, G. G., et al.: National MedicalCare Utilization and Expenditure Survey: Analytic Family File Con-struction Methodology Report. RTI Project No. 25 lU–I 898–14. Re-search Triangle Park, N.C. Research Triangle Institute, Oct. 1983.

Neter, J., and Wasserman, W.: Applied Linear Statistical Models.Homewood, Ill. Richard D. Irwin, Inc., 1974.

Newacheck, P. W.: Utilization and Expenditures for Medical CareServices Provided to Children with Activity Limitations. Grant MCJ–063468-01-0. Prepared for National Maternal and Child Health Re-source Center. Institute for Health Policy Studies, University ofCalifornia, San Francisco, 1985a.

Newacheck, P. W.: Prevalence and Severity of Chronic ConditionsAmong Children. Prepared for National Maternal and Child HealthResource Center. Institute for .Health Policy Studies, University ofCalifornia, San Francisco, 1985b.

Newhouse, J. P., Manning, W. G., Morris, C. N., et al.: Someinterim results from a controlled trial of cost sharing in health insur-ance. N. Engl. J. Med. 305:1501–1507, 1981.

Public Use Data Tape Documentation: Family Data. National MedicalCare Utilization and Expenditure Survey, 1980. Hyattsville, Md.National Center for Health Statistics, Public Health Service. Nov.1986.

SAS Institute, Inc.: SAS Users Guide, Basics, 1982 ed. Cary, N.C.SAS Institute, Inc., 1982.

SAS Institute, Inc.: SASISTAT Guide for Personal Computers, Ver-sion 6 Edition. Cary, N.C. SAS Institute, Inc., 1985.

Shah, B. V.: SESUDAAN, Standard Errors Program for ComputingStandardized Rates From Sample Survey Data. Research TrianglePark, N.C. Research Triangle Institute, Apr. 1981.

Sunshine, J.H.: Medicare, past; present, and future. Natl. J.I4(23):103WIO33, June 5,1982.

Sunshine, J. H.: How many Americans lack outside sources ofpayment for major health care costs? J. Health Hum. Resour. Adm.6(3):341–360, Wint& 1984.

Sunshine, J. H., and Dicker, M.: Family out-of-pocket expendituresfor health care: United States, 1980. National Medical Care Utiliza-tion and Expenditure Survey. Series B, Descriptive Report No. 11.DHHS Pub. No. 87–20211. National Center for Health Statistics,Public Health Service. Washington. U.S. Government Printing Of-fice, Aug. 1987.

Sunshine, J. H., and Dicker, M.: Total family expenditures for healthcare: United States, 1980. National Medical Care Utilization andExpenditure Survey. Series B, Descriptive Report No. 15. DHHS Pub.No. 87–20215. Natiomd Center for Health Statistics, Public HealthService. Washington. U.S. Government Printing Office, Sept. 1987.

Sunshine, J. H., and Dicker, M.: Determinants of total family chargesfor health care: A multiple regression analysis. 1987c (unpublished).

Taube, C. A., Kessler, L. G., and Bums, B. J.: Estimating theprobability and level of ambulatory mental health services use. HealthServ. Res. 21(2, Pt 11):321–340, June 1986.

Whitmore, R. W., Cox B. G., and Folsom, R. E.: Family UnitWeighting .Methodology for the National Household Survey Compo-nent of the National Medical Care Utilization and Expenditure Survey.Contract No. HRA–233–79–2032. Prepared for the National Centerfor Health Statistics. Research Triangle Park, N.C. Research TriangleInstitute, 1982.

Wllensky, G. R., and Walden, D. C.: Minorities, Poverty, andthe Uninsured. Paper presented at the 109th Annual Meeting ofthe American Public Health Association. Los Angeles, Nov. 1981.

Wyszewianski, L.: Financially catastrophic and high-cost cases: Defi-nitions, distinctions, and their implications for policy formation.Inquiry 23(4):382–394, Winter 1986a.

Wyszewianski, L.: Families with catastrophic health care expendi-tures. Health Serv. Res. 21(5):617-634, Dec. 1986b.

33

Appendixes

1.

Il.

111.

Technical Notes on Regression Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. Ir. ..~. ~Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. J......Technical DescriptionofMultiple Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ,.,.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...~..,The Regression Model inFinite Population Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .EstimationofRegression Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Estimating VariancesofRegression Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. ~ .. . . . . . . . . . . .Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...’... 4.14,..Variables Usedinthe Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Functional Formofthe Regression Equation . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Rationaleforthe Functional Form. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...4.... . . . .

lnterpretingthe Regression Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .interpreting MeansofVariables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Analytic Procedures Used..... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..4...... ,~..~..Weighting and Standardizingthe Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Selectingthe initial Variable Set... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ~identifying aCore Setof Variables Through Stepwise Regression . . . . . . . . . . . . . . . . . . . . . .. . . t . . .Estimating Statistical Significance Using SURREGR . . . . . . . . . . . . . . . . . . . . . .,....i,t...,t

AFurtherCheckUsing SURREGR, . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Technical NotesonSurveyand Nonregression Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ~i

Survey Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..s... ..6Sample Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ,’.~..~.

ResearchTriangle lnstituteSample Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .National Opinion ResearchCenterSample Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Collectionof Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ., . .. ~ . . . . . . . ..~..~. . . . . . .Imputation . . . . . . . . . ..-. . ..’ 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .ConstructionofLongitudinalFamilies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Construction and UseofFamilyWeights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Initial Family Weights ..,..... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...+..... 1.4.AdjustmentforUndercoverageand Nonresponse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .’......,

Estimators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Special Requirementsfor lmputationofFamily Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .ReliabilityofEstimates . . . . . . . . . . . . . . . . . . . . . . . . . ~ . . . . . . . . . . . . . . . . . . . . . . . . . . .

Standard Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Confidence intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...4Hypothesis Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . ...:.,

DefinitionofTerms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .’ . . . . . . . . .. . . . . . . . . . .

36

36373737373738

40414141414142

43

5253

5454

575757

59

60

34

List of Appendix Tables

1. Initial setofvariables usedinthe stepwise regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Il. SURREGR regression for the index of financially burdensome family health expenses for”older families . . . . . . . . . 47

Ill. SURREGR regression for the index of financially burdensome family health expenses for younger, lower-income

families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

IV. SURREGR regression for the index of financially burdensome family health expenses for younger, better-off families . . 48

V. SURREGR regression for the index of financially burdensome family health expenses for older families full model . . . . 49

V1. SURREGR regression for the index of financially burdensome family health expenses for younger, lower-income

families full model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

V1l. SURREGR regression for the index of financially burdensome family health expenses for younger, better-off families

full model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

35

Appendix 1Technical Notes on RegressionMethods

Introduction

Multiple regression analysis is a statistical methodfor examining the effect of a set of independent (orcausal) variables on a continuous dependent variable.It permits the analysis of a large number of independentvariables, both continuous and categorical, and providesseparate estimates for the effects of each. The set ofvariables used in the regression analysb, together withthe functional form of the equatioxthat relates the inde-pendent variables to the dependent variable, is calledthe regression model. By analyzing the coefficients inthe regression model, one can explore relationships be-tween the dependent and independent variables and makepredictions about the fiture behavior of the independentvariable.

This appendix first presents a technical descriptionof multiple regression analysis with special attentiongiven to the application of this technique to complexsurveys (such as the National Medical Care Utilizationand Expenditure Survey, NMCUES, which is the sourceof data used in this report). The appendix then presentsa series of somewhat less technical sections. These cover,in turn, the regression model used in this report, theinterpretation of regression coefficients and of meansof variables, and the analytic procedures followed inthis report.

Technical Description of Multiple Regressio~lAnalysis

Introduction

This section discusses the use of multiple regressionanalysis with special emphasis on its application to com-plex survey data and, in particular, on its use in analyzingthe NMCUES data of this study. The following topicsare covered:●

●

●

36

The basic structure of the regression model and howit is usually applied in finite population sampling,

Estimation of regression parameters using complexsurvey data, and

Estimation of variances of parameter estimates.

The Regression Model in Finite Population Sampling

In most statistical literature, the regression modelis stated as:

Yi = ~’~ + ei (1)

where yi is an observable random variable, ~i is a p X 1vector of independent variables, & is a p x 1 vectorof unobservable regression coefficients, and ei is a ran-dom variable with E(ei) = O and Var (ei) = C2, Thesequences yl, yz,... andel, e2,... are generally assumedto be independent and identically distributed. In the nor-mal regression model, it is further specified that el hasa normal distribution.

In finite population sampling, the regression modellooks similar but is formulated in a fundamentally differ-ent way. When sampling from a finite population, itis assumed that there is a population of N pairs ~i,~i):

p = {@l, 31),@z, 32),.,., ON, ~N)} (2)

from which a sample of n pairs is selected. Note thatyi is considered to be fixed in finite population samplingand the randomness is introduced by the selection proc-ess, whereas each yi in the infinite population regressionmodel is considered to be a random variable. The sameis true of the ei. In the finite population setting,expression “independent and identically distributed”no real meaning.

The regression coefficients are determined infinite population setting by the least squares equation:

Q= (x’x)-lx’~,

thehas

the

(3)

where X is the matrix whose rows are made up ofthe ~i’. In the infinite population situation neither theX’ X nor X’y exist in any meaningful sense, since eachwould consifi of divergent infinite sums. (Equation (3)would, however, be used in a sample from an infinitepopulation to estimate the regression coefficients,)

Finally, the error terms ei in the finite populationsetting are defined as the residuals from the least squaresequation in (3):

ei=yi— xi’ b.-— (4)

Thus, in sampling from a finite population, the ei can where U2 is the variance of the independent variablebe seen as a population of fixed values, from which y. When sampling from a finite population, this formulaa sample of size n is drawn. Again, this contrasts with is somewhat more complicated.the usual regression model, where each ei is considered Define the vector Ebyto be random. *

ui = > (yk — X~&) Xki wk. (12)k

Estimation of Regression Parameters

In equation (3), it was indicated that ~ is given Notice that ui is a Horvitz-Thompson type sum whose

by: variance can be calculated using the familiar rules ofstratified and/or cluster sampling. We can now writ:

& = (x’x)-’ X’y. a formula for the approximate covariance matrix of Qas:

The elements of the matrix X’X are given by:cov(~) = (x’x)- ‘ i(g) (x’x)-‘ (13)

Zv = ~ Xkixkj (5)k where ~(@ is an estimate of Cov(@. (See Binder (1983),

pp. 279–292.)while the elements of the vector X’~ are given by:

This method of variance estimation is often calledti = ~ xkiyk. (6) “Taylorizing” or “linearizing,” since the Taylor expan-

k sion is used to develop the linear approximation onwhich equation (13) is based. SURREGR (Holt and

To estimate ~, then, it is necessary to estimate zij Shah, 1982), the computer software package used inand ti. Leh the final steps of the regression analysis of this report,

uses this “Taylorizing” or linearizing technique.~i = sampling weight of ith sample unit, (7)

Unbiased estimates of ztiand ti are given by:

(8)

The Model

Introduction

In this section, the model used in the regression.. .analysis reported in this study is discussed. Thre= topics

and are covered in turn: first, the variables used in the regres-sion analysis; second, the functional form of the

Ati = ~ xkiyk~k. (9) hypothesized relationship between the dependent variable

k and the independent variables; and, third, the rationalefor using this-functional form.

It must be noted that while these estimates are unbiased,the estimate of& obtained by forming Variables Used in the Regression Analysis

~= ~–l; ‘ (10) The variables used in the regression analysis arelisted and described in Table I, which appears at the

is not. Nor is the variance of this estimate easy to calcu- .end of this appendix. The 48 variables in this tablelate, as the next section shows. were either taken directly from or constructed from the

variables on the NMCUES family data tape. The index

Estimating Variances of Regression Coefficients

When sampling from an infinite population, thecovariance matrix of ~ is given by:

covm(~ = (x’x) - 1U2 (11)

‘It should be noted that this is the “classical” point of view in survey sampling.Increasingly, superpopulation models are used in survey inference. Theyassert thatthe finite population under study is simply a large “sample”fromaninfinite“superpopulation.”

of financially burdensome family health care expenseswas used as the dependent variable because it is judged

-to be the best available measure of the burden of healthexpenses on families. The 47 independent variables werechosen from the’larger set of family variables availablefrom the NMCUES on the basis of previous researchand the desire to include the variables believed likelyto have the greatest explanatory power. In developinga regression model, the number of variables is limitedby the amount of data available; screening out unneces-sary variablesgreatlyfacilitatesthe analysis,The 48variables are discussed further in the text of this report.

37

in the section titled “The Model” in the chapter, “TheDeterminants of Financially Burdensome HealthExpenses.”

Functional Form of the Regression Equation

The functional form of the relationship underlyingthe regression equation used in this report is

where s is the number of independent variables trans-formed into their natural logarithm in the estimatingequation (see below). As in equation (l), yi is the depend-ent variable (in this report, it is the financial burdenindex, defined as the ratio of annualized total out-of-pocket health expenses to annual family income), theXOare the independent variables, the bi are the regressioncoefficients, and ei is the error term with E(ei) = O.(The notation EXP means that “e,” the base of naturallogarithms (=2.71 828.. .), is to be raised to the powerindicated by the expression in braces that follows EXP.)Expanding the products in equation (14) yields

Yi = (xilb’)(xz2b2)(Xi3b3) . . .

(EXP{bO})(EXP{b~+*xi,S+,})

(14a)

(EXp{bs+2xi,s+2})...(Exp{ei}).

Equation (14) and equation (14a), which are mathemati-cally equivalent, are not linear in the regression coeffi-cients (the bi) and so cannot be estimated in the fashiondescribed earlier in this appendix. However, these equa-tions have a number of desirable features, as describedbelow, and were selected because of these features.

One desirable feature is that they are easily trans-formed into equations that are linear in the regressioncoefficients. (For more on transformation of variablesin regression analysis, see Neter and Wasserman, 1974,pp. 123–127.) The transformation necessary to achievelinearity consists of taking the natural logarithm of bothsides of the equations. Taking the natural logarithm ofboth sides of equation (14) yields

In(yi) = b. >-fbjln(~u)) + (15)

Similarly, taking the natural logarithm of both sidesof equation (14a) yields

ln(yi) = b(, + bl ln(~il) +

b21n(~i2) + . . .

(15a)

+ . . . + ei.

Equation (15) and equation (15a), which are algebraicallythe same, are linear in the regression coefficients (thebj) and are the regression equation used in the analysesin this report. Their parameters and the variances ofthese parameters were estimated by the techniques de-scribed in the first section of this appendix.

A point to note here with respect to equation (15)is that in this equation, some of the original NMCUESvariables are transformed into their natural logarithms.That is, these variables are replaced by their naturallogarithms. Table I indicates the variables for whichthe natural logarithm, rather than the untransformed vari-able, was used.

AIso, equations (14) and(15) require any categoricalindependent variables to be expressed in numerical form.This is readily accomplished for variables which takeon only two values. .For these variables, it is accom-plished by assigning “1” to one of the values and “O”to the other. For example, sex of the head of the familywas coded as female = 1 and male = O. For categoricalvariables that take k>3 values, an extension of thisprocedure was used. Such categorical variables wererepresented in the regression equation by a series ofk – 1 “dummy” variables, each of which can take thevalue of “1” or “O.” Each k – 1 value of the originalcategorical variable was associated with one dummyvariable, which was assigned the value of O or 1 foreach family in the sample depending on whether thevalue (of the original categorical variable) was true(dummy = 1) or not (dummy = O). For example,the head-and-spouse structure of a family takes on threevalues: (1) head and spouse always present, (2) familyalways has only a head, and (3) changing head-and-spouse structure. Dummy variables, D1 and D2, werecreated from the second and third of these three values,while the first value was the omitted value.

In creating dummy variables, one of the originalk vaIues of the original variable must be omitted, sincethe kth dummy variable would be a linear co~binationof the first k– 1 dummy variables. (If one independentvariable in a regression is a linear combination of others,the matrix X’X (see equation (3)) cannot be invertedand the regression coefficients are thus undefined.) Typi-cally, the omitted value was the one regarded conceptu-ally as the “base case” or was the most common state.Table I shows the dummy variables that were used andindicates which value was omitted.

38

RationaIe for the Functional Form

The dependent variable. The functional form shownin equation (14) requires that the dependent variablebe used in logarithmic form when estimating the regres-sion model using equation (15). This functional formwas chosen for three reasons.

First, it is believed that the relationship betweenthe independent and dependent variables is primarilymultiplicative. For example, it was expected that thereduction in the financial burden index associated withMedicaid coverage would be multiplicative rather thanadditive. To simplify greatly, in an additive model, ifMedicaid coverage generally reduced the index from8 percent to 2 percent for some families, it should reduceit from 6 percent to O percent for other families andreduce it from 4 percent to minus 2 percent—an obviousabsurdity-for yet others. In contrast, a multiplicativerelationship would, following this example, generallyreduce the index to about one-fourth of its value-thatis, reduce it from 8 percent to 2 percent for the’ firstcategory of families, from 6 percent to 1.5 percent forthe second category, and from 4 percent to 1 percentfor the third. A multiplicative effect of this type is whatthe authors believed likely. When underlying relation-ships are multiplicative, a functional form with alogarithmic dependent variable in the regression model(that is, in equation (15)) is appropriate because sucha functional form is multiplicative in its untransformedversion (that is, equation (14)).

Second, a logarithmic dependent variable was usedbecause prior research in evaluating the appropriatenessof different functional forms for regression analysis ofmedical expenditure data indicates that a logarithmicdependent variable should be used in the estimating equa-tion. Duan et al. (1982) carried out an extensive analysisof residuals from various models and found that theyapproximated a normal distribution (as assumed by thelinear regression model) more closely when a logarithmicdependent variable was used than when an untransformeddependent variable was used in the regression model.

Third, the literature on health expenditures, perhapsbecause of the preceding two reasons, almost alwaysuses a logarithmic dependent variable in regression equa-tions. Some examples, which are similar to this reportin the databases they use or the subjects they investigate,are Taube et al. (1986) and Farley (1986). By followingthe literature, comparability of results is enhanced.

Itiepetient’ variables. Given a logarithmic depend-ent variable in the estimating equation (equation (15)),there remains the question of whether or not to transformthe (continuous) independent variables in the equationinto their logarithms. The alternatives of carrying outsuch a transformation or not doing so imply differentrelationships. The following paragraphs first describe.in nontechnical terms the relationships implied by eachalternative and then describe the choices made.

When the dependent variable is logarithmic, a

logarithmic independent variable in the estimating equa-tion implies constant elasticity. In nontechnical terms,this means that a 1 percent increase in the untransformedindependent variable produces a fixed percent increase(or decrease) in the untransformed dependent variable,with this increase (or decrease) called the elasticity.(Technically, elasticity is defined as (dyiidx~) (x~iyi),and the nontechnical description in the preceding sen-tence is usually a close approximation to elasticity asmeasured by this formula. ) A Iogatithmic independentvariable also implies that a fixed percent change in theuntransformed independent variable-for example, in-creasing it by 100 percent (that is, doubling it) producesa uniform percent change in the untransformed dependentvariable. In contrast, with a logarithmic dependent vari-able, using a continuous independent variable in non-transformed form in the estimating equation implies thatit is a unit increase in the nontransformed independentvariable (not a percent increase) that produces a uniformpercent change in the nontransformed dependentvariable,

Manipulation of equation (14) will show that theserelationships hold. [Independent variables used inlogarithmic form in the estimating equation (equation(15)) are the first s of the XU,and these are found inthe product term of equation (14). Independent variablesused without transformation in the estimating equation(equation (15)) are the remaining xv, and these are foundin the exponential term of equation (14). ]

In light of the different relationships that hold truefor logarithmic and untransformed continuous indepen-dent variables in the estimating equation (equation (15)),a choice was made between these two forms for continu-ous independent variables. The choice was based onbeliefs about the nature of the underlying relationships.For example, it was believed that if an increase from10 to 20 annualized family illness days spent in bed(“bed days”) resulted in, say, a doubling of the financialburden index, then a further increase from 20 to about40 bed days would be required to produce another doub-ling of the index. Hence bed days was used in logarithmicform in the estimating equation (equation (15)). Usingbed days untransformed would imply that an increasein bed days from 20 to about 30 would produce thesecond doubling of the index. In contrast, to take asecond example, it was believed that if an increase inthe age of the family head horn 30 to 40 years produceda given percent increase in the index (say, increasingit by 30 percent—that is, multiplying it by 1.3), thenan increase in age of the family head from 54 to approxi-mately 64 years would produce an equally large percentchange (that is, a 30 percent increase) in the index.Hence, age of the family head was used untransformed.If it were used in logarithmic form, an increase in thehead’s age from 54 to 72 years would be required togenerate the same effect (in percent terms) as the increasefrom age 30 to 40.

39

Based on previous research about the underlyingrelationships, the following four continuous variableswere used in logarithmic form in the estimating equation(equation (15)):

● Average family size.

● Annualized family bed days.

“ Annualized family work days lost because of illness(“work-loss days”).

● Family annual income.

Becau~e family bed days and family work-loss dayscan take on the value zero, and the logarithm of zerois undefined, the value of these variables was increasedby one before performing the logarithmic transformation.

The following three continuous independent vari-ables were used in untransformed form:

● Age of head of family.

● Years of education of head.

● Annualized number of hospital discharges for familymembers.

Using the PC SAS computer program (SAS Institute,1985), informal, exploratory tests were conducted onthe effect of the choice between no transformation ofthe above independent variables and a logarithmic trans-formation of them. There was little difference in R-squareand, in general, relatively littl~ difference in tests ofstatistical significance for each of these independent vari-ables when a logarithmic transformation was substitutedfor no transformation and vice versa. This would indicatethat significant factors can be detected using either thelogarithmic or untransformed dependent variable. Thiscan be explained partially by the fact that stronglogarithmic effects will also have roughly linear patterns.It should also be noted that the results of significancetests will generally be valid in regressions when usinglarge data sets because of the asymptotic normality prop-erties of least squares estimators. That is, even thoughthe residuals may not be normally distributed (one ofthe key assumptions on which the regression F testsare based), the F tests used in regression will be validfor large data sets. (See Arnold, 1981.) Thus the resultsof the significance tests may be similar even if thelogarithmic transformation improves the normality ofthe regression residuals.

Interpreting the Regression Coefficients

Because the estimating equation (equation (15)) in-volves variables in logarithmic form, interpretation ofthe regression coefficients is somewhat complex. Thereader may find the following explanation helpful ininterpreting the regression coefficients, which appearin Appendix Tables II, III, and IV.

When the dependent variable in a regression modelis in logarithmic form in the estimating equation, as

is the financial burden index in all the regressions in 4this report, three different types of independent variablescan be distinguished.

1. First are dummy variables. (Again, “dummy vari-able” designates a categorical variable that takes ononIy the values Oand 1.) The regression coefficient,b, for a dummy variable has the following interpreta-tion. The presence of the characteristic indicatedby the dummy variable is associated with multiplica-tion of the underlying, nonlogarithmic value of thedependent variable” by approximately antilog(b),where antilog(b) is the number whose logarithm isb. For example, in Table III the regression coefficientof D1, a dummy variable denoting families witha head but no spouse, is —0.50. The antilog of– 0.50 is 0.61. Thus, the regression coefficient of– 0.50 means that families with a head and no spousehave, other things equal, a financial burden indexabout 0.61 times as large as families with otherhead-spouse structures (or a financial burden index39 percent smaller than families with other head-spouse structures). Table H, which interpretsTable III, presents this finding.

2. Second are continuous independent variables usedin logarithmic form in the estimating equation, Forsuch variables, the regression coefficient, b, is theelasticity. This means that each 1-percent increasein the underlying, nonlogarithmic independent vari-able is associated with approximately a b-percentincrease in the underlying, nonlogarithmic dependentvariable. For example, 130, the natural logarithmof a family’s annual income, is a continuous indepen-dent variable used in logarithmic form in Table 111,Its regression coefficient there is – 0.56. This meansthat each l-percent increase in annual family income(the underlying, nonlogarithmic independent vari-able) is, other things equal, associated with approxi-mately a 0.56 percent decrease (a 0.56 percent de-crease is a – 0.56 percent increase) in the levelof the financial burden index (the underlying, non-Iogarithmic dependent variable). Again, Table H,which interprets Table III, presents this finding.

3. Finally, there are continuous independent variablesused in nontransformed (that is, nonlogarithmic)form. If the regression coefficient for such a variableis b, then each increase of one unit in the independentvariable is associated with a multiplication of theunderlying, nonlogarithmic form of the dependentvariable by approximately antilog(b), Again, TablesIII and H can serve to illustrate this point. Theage of the family head in years, D6, is a continuousindependent variable used in nontransformed formin the regression equation whose results are presentedin Table ~. Its regression coefficient in that equationis found to be 0.025. The antilog of 0.025 is 1.025,Hence, the interpretation of the regression result isthat each increase of one year (the unit in which

40

age is measured) in the age of a family’s head is,other things equal, associated with a multiplicationof the level of the family’s financial burden indexby approximately 1.025, which is an increase of2,5 percent. Again, Table H, which inteqrets theresults reported in Table III, presents this finding.

These interpretations of regression coefficients canbe demonstrated by suitable manipulation of equation(14) (or of equation (14a), which is mathematically equi-valent). However, in using these interpretations of~egres-sion coefficients, it should be noted that antilog(b), theantilog of the estimated regression coefficient, is a biasedestimator of the antilog of the regression coefficient,although statistical significance tests associated with thecoefficients are sound. Thus, the regressions correctlyindicate which variables are statistically significant. Ifextensive estimation using antilogs is to be carried out,corrections for the bias are available.

Interpreting Means of Variables

Because of the mixture of dummy variables, untrans-formed variables, and logarithmic variables in the regres-sions, readers may find helpful the following informationon interpreting the means of variables. (Means of thevariables used in the regressions are shown in TablesII, IfI, andIV.)

The mean value of a dummy variable is the proportionof the population that has the characteristic denoted bythe variable, For example, Table III shows that the meanof D1, the dummy variable denoting families with ahead but no spouse, is 0.42 for the U.S. populationof younger, lower-income multiple-person families in-cluded in this table. This means that 42 percent of thesefamilies have ahead-only family structure.

The mean value of an untransformed continuous vari-able is simply the (familiar) arithmetic mean of the vari-able for the population in question. For example, themean of D6 in Table III is 38.1. This variable measuresthe age of the family head in years and thus showsthat the (arithmetic) mean age of the family head forthe U.S. population of younger, lower-income multiple-person families included in the table is 38.1 years.

The mean value of a logarithmic continuous variableis the logarithm of the geometric mean of the variable.Taking the antilogarithm of the mean does not givethe arithmetic mean of the untransformed variable. Thegeometric mean often differs ve~ substantially from thearithmetic mean and should not be confused with it.

Analytic Procedures Used

Introduction

This section describes the analytic procedures usedin the regression analyses in this report. Several stepswere involved. These were weighting and standardizing

the data; selecting the initial variable set; finding a coreset of variables through stepwise regression; choosingcriteria for evaluating the statistical significance of vari-ables; and estimating the statistical significance of thecore variables with SURREGR (Holt and Shah, 1982),a computer program that takes account of the complexsample design of the NMCUES. These steps are de-scribed in turn in this section.

Weighting and Standardizing the Data

Before regression analysis (or other analysis) of thedata could begin, certain weighting and standardizingprocedures had to be carried out. These procedures aredescribed in more detail in Appendix II, but a summaryof them is included here in order to present in sequencethe procedures followed in the regression analysis.

Weighting of each case (family) in the data set beganwith a weight that previous reports on NMCUES familydata have called FWEIGHT. Described simply,FWEIGHT is the reciprocal of the sampling probabilityadjusted for undercoverage and nonresponse andsmoothed to agree with population totals from the March1980 Current Population Survey. For each case (family),FWEIGHT was multiplied by the proportion of the surveyyear (calendar year 1980) that the family was eligiblefor the survey. This time-adjusted weight, calledAWEIGHT, is the weight used in the regression analysesand in other analyses in this report. AWEIGHT differsfrom FWEIGHT only for families not in the samplefor a full year.

For these families, standardization of data on income,health spending, health care use, and other variablesthat measure rates was carried out. Data on these vari-ables covering the period the family was in the samplewere divided by the proportion of the year the familywas in the sample in order to derive an annualized rate.For example, a family in the sample for half the yearwith $10,000 of income and$150 of out-of-pocket healthexpenses recorded during this half year had its annualizedincome recorded as $20,000 and its annualized out-of-pocket health expenses recorded as $300. Annualizedstatistics, like these, were used in the regression analysesas the measure of all variables involving rates.

Selectingthe Initial VariableSet

Regressions were run separately for three categoriesof multiple-person families:●

●

w

older families, those with one or more membersage 65 or older.

Younger, lower-income families, those with nomember 65 or older and with income below 200percent of the poverty level.

Younger, better-off families, those with no member65 or older and with income of 200 percent of thepoverty level or more.

41

For each family catego~, a small number of the47 independent variables shown in Table I were omittedfrom the initial regressions because they were not applica-ble or relevant. Thus, the initial regressions involvedthe financial burden index as the dependent variableand slightly fewer than 47 independent variables. Theomitted independent variables and the reasons for theiromission are as follows. First, one dummy variable forsource of health care coverage was omitted for eachof the three populations, for the reason described abovein the section “Functional Form of the Regression Equa-tion.” The omitted variable was that representing themost common source of coverage—medicare plus privateinsurance, 138, for older families; and private insuranceonly, 134, for both younger family populations. Thedummy variable identifying families with all membersage 65 or older, D7, was omitted from the regressionsfor both younger family categories because such families,by definition, have no members age 65 or older. Thevariable measuring work days lost due to illness, H29,was omitted from the regression for older families be-cause many such families have no working members,which makes this variable meaningless for them. Thedummy variable for families with health care coverageentirely from “other public” sources, 140, was omittedfrom the regression for older families because none ofthe older families in the sample had such coverage.The dummy variable for families with all members hav-ing an unknown perceived health status rating, H25,was omitted from the regressions for older families andfor younger, lower-income families because no familiesin; the sample in these two categories had this rating.Thus there were 43 independent variables in the initialregressions for older families, 44 independent variablesin the initial regressions for younger, lower-incomefamilies, and 45 independent variables in the initial re-gressions for younger, better-off families.

Identifying a Core Set of Variables Through StepwiseRegression

For each of the three family categories, stepwiseregression was used to select a preferred subset of inde-pendent variables from among the original 43 to 45independent variables. The stepwise regression was car-ried out using PC SAS (SAS Institute, 1985).

Stepwise regression was used for two reasons. Forone, a number of factors were operationalized by multiplevariables. For example, family health status wasoperationalized by four sets of variables: (1) family beddays due to illness, (2) family work-loss days due toillness, (3) the excellent-good-fair-poor scale of reportedhealth status, and (4) the limitations in main activityscale. Because of multicollinearity, using multiple vari-ables that operationalize the same concept in a regressionequation often yields” distorted regression coefficientsand large standard errors that indicate none of the vari-ables is significant. Stepwise regression generally selects

out a subset of the variables operationalizing, a givenconcept, thus avoiding severe multicollinearity problems.

Second, as more variables are entered into a regres-sion equation, there is a reduction in the precision withwhich the effects of any one variable can be identified.Standard errors increase, which tends to reduce thenumber of variables identified as significant. Stepwiseregression permits a trade-off between the additionalexplanatory power obtained by adding more variablesto a regression and the loss of precision in identifyingthe effects of anyone of them.

The preferred independent variable set was definedby the step of the stepwise regression that had the lowestvalue of C(p). (For C(p), see Mallows, 1973.) Thevariable entered in the step at which C(p) reached aminimum and all variables entered in preceding stepswere included in the prefemed variable set. All othervariables were excluded. However, if C(p) was stilldecreasing when all independent variables that enteredthe stepwise regression with probabilities less than 0.20had been added, then the last step before the entry proba-bility for a variable exceeded 0.20 was chosen as definingthe preferred variable set. This step was used to definethe preferred variable set in the same fashion that C(p)was otherwise used to define it. Finally, all prefemedvariable sets were required to contain the variables 130(natural logarithm of annual family income), H13 (indi-cating whether or not any family member was hos-pitalized), and D5 (natural logarithm of average familysize). If any of these three variables was missing fromthe preferred variable set developed by the stepwiseregression, the missing variable(s) was added and thevariable set that included all three of these variableswas considered the preferred variable set. Income andhospitalization were included because the literature hasfound them very important. Family size was includedin order to try to assure that effects of family size weredistinguished from effects of family structure variables,(The family structure variables used were whether afamily had children, whether it had a head and a spouseor only a head, and whether its composition was stable.)

The PC SAS stepwise regression program weightseach case (family) in the regression, but does not takeaccount of the complex sample design of NMCUES.That is, it estimates variances according to equation(11), the formula appropriate for noncomplex samples,while equation (13), which is more involved, is theappropriate formula to use in estimating variances ofNMCUES data. There appears to be no stepwise regres-sion program available that takes complex sample designinto consideration in estimating variances.

Estimating Statistical Significance Using SURREGR

Because the stepwise regression procedure in PCSAS does not take account of complex sample designin estimating variances, its estimates of statistical signifi-cance can involve large errors when it is used in analysis

.

42

of data from surveys, such as the NMCUES, whichhave complex sample designs. Therefore, the next stepsin the regression analysis procedure involved the useof a regression software program that does estimate var-iances of complex samples appropriately, using equation(13). The program used was SURREGR (Holt and Shah,1982), which runs within the SAS system. NonstepwiseSURREGR regressions were run on the preferred variablesets, and i’t is the results of these regressions that areshown in Tables II, III, and IV.

It should be noted that identical estimates of regres-sion coefficients and of R-square (the proportion of thevariance in the dependent variable explained by the inde-pendent variables) are produced by the PC SAS stepwiseregression procedure and the SURREGR regression pro-cedure. The differences of concern between the twoprocedures are in the statistical significance levels thatthey report for independent variables.

In the regressions shown in Tables II, III, and IV,the probability that the regression coefficient associatedwith any variable in the regressions was different fromzero was computed by SURREGR using the F statistic.This probability is shown in the last column in eachtable. A regression coefficient was considered significantif its probability of occurring by chance was less than0.05.

However, because there are 17 to 24 regression coef-ficients (one for each independent variable) in the pre-ferred variable sets, a simple use of a 0.05 probabilitytest would not be appropriate. Approximately one coeffi-cient meeting a simple 0.05 probability test would beexpected for every 20 regression coefficients, and thusapproximately one such coefficient would be expectedin each of Tables II, III, and IV purely by chance.

The significance test actually used was that the proba-bility associated with any one variable had to be lessthan 0.05 + n, where n is the number of independentvariables in the preferred variable set. This test is analo-

gous to the multiple t-test used in frequency tables inthis report (see Levy and Lemeshow, 1980, p. 296)and in previous reports on NMCUES family data (Dickerand Sunshine, 1987; Sunshine and Dicker, 1987a; Sun-shine and Dicker, 1987b). The actual probability corre-sponding to 0.05+ n was as follows for the three popula-tions of multiple-person families studied:

Population Probability

Older families (Table II) . . . . . . . . . . . . 0.0023Younger, lower-income families (Table Ill) . 0.0029Younger, better-off families (Table IV) . . . 0.0021

Variables with probabilities below these levels were con-sidered significant and appear in Tables G, H, and J,which report significant findings and accompany thetext discussions of findings for each of the three familypopulations.

A Further Check Using SURREGR

The preferred’variable sets should include all statisti-cal y significant independent variables, for stepwise re-gression selects the independent variables with the great-est statistical significance first, and then moves to pro-gressively less significant variables. However, in lightof possible problems arising because the PC SAS step-wise procedure does not compute variances (and hencesignificance levels) based on the NMCUES complexsample design, a check for omitted significant variableswas performed. For each of the three family populations,the full regression model, with all 43 to 45 independentvariables, was run using SURREGR. The results areshown in Tables V, VI, and VII. These results weregenerally as expected, and no statistically significantvariables were found that were not also statistically sig-nificant in the (smaller) preferred variable sets.

Table I

initial set of variables used in the stepwise regression

Variable

Variable type indicator Description of variable

Dependent variable

The Index of financially burdensome YI The ratio of annualized total out-of-pocket health expenses to annualized family income,family health expenses (continuous) transformed into its natural logarithm.’

Independent variables

Demographic and social

Head-spouse structure of family(3 categories, 2 dummy variables)z

Head-only family D1 1 = Family had a head only (no spouse) during time in survey;O= All other head and spouse combinations.

Head-spouse change3 D2 1= Family had an unstable head-spouse structure during time in survey;O= All families with stable head only or stable head-and-spouse.

See footnotes at end of table.

43

Table i-Continued

Initial set of vafibles used in Me stepwise regression

VariableVariable type indicator Description of variable

Dynamic-static nature of family(3 categories, 2 dummy variables)4

Head-spouse change3

Other change

Presence of children in family(2 categories, 1 dummy variable)

Family size (continuous)

Age of head of the family(continuous)

Age of family members(2 categories, 1 dummy variable)5

Sex of the head of the family(2categories, l dummy variable)

Race of head of the family(3 categories, 2 dummy variables)6

Black

Other

Ethnicity of the head of the family(2 categories, 1 dummy variable)

Education of head of the family(continuous)

Health related

Hospitalization of a family member(2 categories, 1 dummy variable)

Total number of hospital discharges(continuous)

Institutionalization of a familymember (2 categories, 1 dummyvariable)

Death of a family member(2 categories, 1 dummy variable)

Birth of a family member(2 categories, 1 dummy variable)

Illness in a family member(4 dummy variables)7

Cancer and other neoplasms7

Circulatory and heart disease7

Accidents, injuries, and poison-ings

D2

D3

D4

D5

D6

D7

D8

D9

D1O

D11

D12

H13

H14

H15

H16

H17

H18

H19

H20

1 = Family had an unstable head-spouse structure during time in surveyO= All families with stable head only or stable head-and-spouse.

1= Family had a stable head-spouse structure during time in survey but other familymember changed, or family did not exist full year;

tl =Other family change status.

1= Family had a member 16 years of age or younge~O= All family members 17 years of age or older.

Average family size (in persons) during time in survey, transformed into its naturallogarithm.

Age of the head of the family in years, as of January 1, 1980.

1 =All family members are 65 years of age or ovenO= Some or all family members are less than 65 years of age.

1= Female head of family;O= Male head of family.

1= Black head of family;O= Head of family of other race.

1= Other (neither black nor white) head of familfiO= Head of family either black or white.

1= Hispanic head of famil~O= Head of family of other ethnicity.

Formal education of the head of the family in years of ecfucation (18 was the highest valueused).

1 = Family had one or more members discharged from a hospital during its time in thesurvey;

O= No family members discharged from a hospital.

Annual rate of hospital discharges for all family members.

1 = Family had one or more members institutionalized during its time in the survey or, if Itdid not continue until the end of 1980, at its termination;

O= No family members were institutionalized.

1 = Family had one or more members die during its time in the survey or, if it did notcontinue until the end of 1980, at its termination;

O= No family member died.

1 = Family had one or more members who gave birth to a child during its time in thesuNe~

O= No family member gave birth toa”child. .,

1 = Family had one or more members with some type of neoplasm during its time in thesurvey

O= No family member had a neoplasm.

1 = Family had one or more members with some type of circulatory or heart disease duringits time in the survey

O= No family member had circulatory or heart disease.

1= Family had one or more members with some type of accident, injury, or poisoningduring its time in the surve~

O= No family member had an accident, injury, or poisoning.

See footnotesatendoftable.

44

Table i-Continued

Initial set of variables used in the stepwise regression

VariableVariable type indicator Description of variable

Other illnesses’ H21 1= Family members had none of the above illnesses, but one or more members had someother illness during his or her time in the survey;

O= One or more family members had one of the above illnesses or all family members hadno illness.

Perceived health status rating offamily (5 categories, 4 dummy vari-ables)a

Good

Fair

Poor

Unknown

Limitation in usual activity rating offamily (3 categories, 2 dummyvariables)g

Secondary limitation

Major limitation

Family Illness days in bed(continuous)

Family work-loss days(continuous)io

Income and insurance

Family income (continuous)

Completeness of health carecoverage (4 categories, 3 dummyvariables)ll

Partial coverage 1

Partial coverage 2

No coverage

Source of health care coverage(8 categories, 7 dummy variablesused for each population group)12

Private insuranceiz

Medicaid

Medicare

Medicare and other public

H22 1= Worst perceived health status of any family member was reported as “goodO= All family members were reported in excellent health or some family members were

reported in fair or poor health.

H23 1= Worst perceived health status of any family member was reported to be “faifiO= All family members were reported in excellent or good health or some member was

reported in poor health.

H24 1= Worst perceived health status of any family member was reported to be “poor”;O= No family member had a “poor” rating.

H25 1= Reported health status of all family members is “unknown”;O= Reported health status of at least some family members is known.

H26 1= Most severe Ii.mitation reported for any family member was either a limitation inseconday activity or a limitation in amount or kind of main activity (work, house-keeping, school, and so on);

O= No family member was reported to have a limitation or a major limitation was reportedfor one or more family members.

H27 1= Most severe limitation reported for any family member was inability to perform a usualmajor activity (work, housekeeping, school, and so on);

O= No family member was reported as unable to perform his or her usual major activity.

H28 Annual rate of total illness days spent in bed for all family members. One day was added tothe annual rate and the resulting statistic then transformed into its natural logarithm.

H29 Annual rate of total work-loss days due to illness. One day was added to the annual rateand the resulting statistic then transformed into its natural logarithm.

[30 Annualized family income in dollars transformed into its natural logarithm.

131 1= All family members covered but some or all only part yearO= All other &pes of coverage.

132 1= Some family members had coverage, but some had no coverage;O= All other types of coverage.

133 1= No family member covered;O= Partial or full coverage.

134 1= Family members only had coverage from private health insurance;O= All other sources of coverage.

135 1= Family members only had coverage from medicaid;O= All other sources of coverage.

136 1= Family members only had coverage from medicare;O= All other sources of coverage.

137 1= Family members only had coverage from medicare and other public programs;O= All other sources of coverage.

Seefootnotesatendoftable.

45

Table t-Continued

lnitid set of variables used in the stepwise regression

VariableVariable type indicator Description of variable

Medicare and private’2 138 1= Family members only had coverage from medicare and private health insurance;O= All other sources of coverage.

Other public and private 139 1= Family members had coverage from both (1) public sources other than medicare and(2) private insurance;

O=All other sources of coverage.

Other public 140 1= Family members only had coverage from public source(s) other than those listed above;O=All other sources of coverage.

Unknown 141 1= Every family member either had no coverage or had coverage from eources notidentified;

O= One or more family members had coverage from identified sources.

Geographic

Region of United States(4 categories, 3 dummyvariables)’3

North Central G42

South G43

West G44

Urban-rural location (4 categories,3 dummy variables)’4

Metropolitan suburb G45

Nonmetropolitan urban area G48

Nonurban area G47

1= Head of family resided in North Central census region;O= Head resided in other region of U.S.

1= Head of family resided in South census region;O= Head resided in other region of U.S.

1 = Head of family resided in West census region;O= Head resided in other region of U.S.

1= Head of family resided in a suburb of a metropolitan statistical area;O= Head resided in another location.

1 = Head of family resided in an urban area that was not a part of a metropolitan statisticalarea;

O= Head resided in another location.

1= Head of family resided in non-urban area;O= Head resided in another location.

‘Total out-of-pockethealthexpenses is the sum of out-of-pocket expenses for health care sewices plus family-paid premiums (see discussion in text).‘Omitted category is families with both a head and a spouse during time in suwey.3The variable is entered only once in the regression, but functions both as a measure of head-spouse structure and as a measure of dynamic-static nature of thefamily.40mitted category is static families—that is, families that had no change in membership and were in the suwey the full suwey year.;~his variable was only used in regressions for older famities (families with a member 65 years of age or over). When usad with this population, the “O”categorydesignates families with members both over and under 65 years of age.‘Omitted categoiy is families with a whtie head of family.‘Omitted category is families in which no family member reported an illness. The dummy variables for cancen heart and circulatory disease; and accidents, illnesses,and poisonings are not mutually exclusive.80mitted category is all family members reported to be in excellent health.‘Omitted category is families in which no family member reported a timitation. Family members for whom limitation status was unknown were coded as having nolimitation.‘~his variable was not used in regressions for older families (families with a member 65 years of age or over).llomi~ed Categov is families in which aII familY members had full-year coverage by pritiate health inSUranCeandlor public health care coverage Program.12For regressions for older famifies, the omi~ed category is coverage by both Medicare and private inSUranCe(138).For regressions for Y0un9er famifies, the omiffed

category is coverage from private health insurance only (134).‘sOmirted category is residence in the Northeast census re9ion-140mitfed Categow is resi,jence in the central city of a metropolitan statistical area-

NOTE Further information on the variables in ttis table may be found either in Appendix Ill, “Definition of Terms,” or in the text.

46

Table II

SURREGR regression for the index of financially burdensome family health expenses for older families

Independent variable Mean

DI Headonly . . . . . . . . . . . . . . . . .

D4 Children . . . . . . . . . . . . . . . . . .D5 Family size . . . . . . . . . . . . . . . .D6Ageof head . . . . . . . . . . . . . . .

D9 Black race . . . . . . . . . . . . . . . .Dl(J’’Other’’ race . . . . . . . . . . . . . . .

D12Education of head . . . . . . . . . . . .

H13 Hospitalization. . . . . . . . . . . . . .H14 Discharges . . . . . . . . . . . . . . .

HI 5 Institutionalization . . . . . . . . . . . .

H19 Heart disease, etc. . . . . . . . . . . .H20Accidents, etc. . . . . . . . . . . . . .H26 Secondary limitation . . . . . . . . . .

H28 Bed days . . . . . . . . . . . . . . . .

1301ncome . . . . . . . . . . . . . . . . . .132 Partial coverage 2 . . . . . . . . . . . .135 Medicaid coverage only . . . . . . . . .136 Medicare coverage only . . . . . . . . .

137 Medicare and other public coverage . .

141 Coverage source unknown . . . . . . .

G42 North Central region . . . . . . . . . .G43 South region . . . . . . . .. . . . . .

0.256

0.1080.885

68.0890.0930.0209.8400.3880.7110.0180.7240.2760.0801.8739.5870.1080.0040.0950.0530.0300.2350.346

Regressioncoefficient

–0.3840.2880.0050.007

–0.431– 0.3310.0200.2060.0760.5510.2180.1340.1560.060

– 0.853–0.169–1.119– 0.243–2.162–1.1510.1090.288

Standard errorof regression

coefficient

0.076

0.164

0.0030.1520.2400.0100.0940.0340.2170.0670.0660.0810.0250.0530.1250.3170.0790.2550.3040.0670.063

F Value

25.323.080.005.277.991.894.394.864.906.44

10.684.043.705.86

262.581.85

12.439.58

72.0414.352.66

20.83

Probabili~

0.0000

0.0836

0.97310.02480.0062

0.1735

0.03990.03090.03020.0134

0.0017

0.04830.0587

0.0182

0.0000

0.17810.00080.0028

0.0000

0.00030.10760.0000

Intercept = 4.267 Mean of dependent variable = – 2.971Number of observations = 849 Probability = 0.0000Multiple correlation coefficient squared (Rz) = 0.529 66 denominator degrees of freedomF = 54.97 with 22 degrees of freedom

NOTES: Older families are families with member(e) 65 yeare of age or over. A probability of 0.0023 was needed for statistical significance at the .05 level using an Ftest that is analogous to a multiple t test (see Appendix l). Standard error of regression blank because F test had a value of zero after rounding.

Table Ill

SURREGR regression for the index of financially burdensome faniily health expenses for younger, Iowerincome families

Standard errorRegression of regression

Independent variable Mean coefficient coefficient F Value Probability

DI Head only . . . . . . . . . . . . . . . . .D5 Family size . . . . . . . . . . . . . . . .

D6Ageof head . . . . . . . . . . . . . . .

D12Education of head . . . . . . . . . . . .

H13 Hospitalization. . . . . . . . . . . . . .H17 Birth . . . . . . . . . . . . . . . . . . .H18 Cancer . . . . . . . . . . . . . . . . .

H20Accidents, etc. . . . . . . . . . . . . .

H22 Goodhealth . . . . . . . . . . . . . . .H26 Secondary limitation . . . . . . . . . .

H29Work-lossdays . . . . . . . . . . . . .

1301ncome . . . . . . . . . . . . . . . . . .

132 Partial coverage2 . . . . . . . . . . . .

135 Medicaid coverage only . . . . . . . . .139 Other public and private coverage . . . .140 Other public coverage only . . . . . . .

141 Coverage source unknown . . . . . . .

0.422

1.258

38.126

10.802

0.334

0.059

0.082

0.467

0.374

0.095

1.190

9.104

0.145

0.117

0.230

0.007

0.280

–0.501– 0.2980.0250.0490.3600.4330.3440.1390.168

–0.2130.117

– 0.5560.218

– 1.674– 0.394– 1.972– 0.564

0.1080.1500.0040.0160.1340.2040.1770.0970.0870.2050.0320.120“0.1280.2530.1210.4900.112

21.52

3.93

47.26

9.91

7.22

4.51

3.80

2.07

3.76

1.08

13.49

21.28

2.89

43.76

10.59

16.16

25.39

0.00000.05140.00000.00240.00900.03740.05550.15510.05660.30230.00050.00000.09380.00000.00180.00010.0000

Intercept = 0.823Number of obsenrations = 920

Mean of dependent variable = –3.342Probability = 0.0000

Multiple correlation coefficient squared (R=) = 0.267 69 denominator degrees of freedomF = 21.13 with 17 degrees of freedom

NOTES: Yourrger, Iower-irrcome families are families with no member 65 years of age or over and with incomes below 200 percent of the poverty level. A probability of0.0029 was needed for statistical significance at the .05 level using an F test analogous to a multiple f test (see Appendix l).

47

Table IV

SURREGR regression for the index of financially burdensome famity heatth expenses for younger, better-off families

Standard errorRegression of regression

Independent variable Mean coefficient coefficient F Value Probability

DI Head only . . . . . . . . . . . . . . . . . 0.145 –0.136 0.079 2.94 0.0909D30therchange . . . . . . . . . . . . . . . 0.167 –0.151 0.056 7.23 0,0090D4 Children . . . . . . . . . . . . . . . . . . 0.593 0.135 0.060 5.02 0.0283D5 Family size . . . . . . . . . . . . . . . . 1.135 0.241 0.084 8.20 0.0056D6Age of head . . . . . . . . . . . . . . . 42.050 0.015 0.002 61.37 0.0000D9 Black race . . . . . . . . . . . . . . . . 0.063 –0.335 0.100 11.27 0.0013D12Education of head . . . . . . . . . . . . 12.665 0.034 0.007 21.59 0.0000H13 Hospitalization. . . . . . . . . . . . . . 0.268 0.256 0.072 12.68 0,0007H14 Discharges . . . . . . . . . . . . . . . 0.423 0.069 0.044 2.52 0,1168H15 Institutionalization . . . . . . . . . . . . 0.001 –2.017 0.700 8.31 0.0053H17 Birth . . . . . . . . . . . . . . . . . . . 0.037 0.185 0.122 2.30 0.1337H19 Heart disease, etc. . . . . . . . . . . . 0.265 0.081 0.053 2.35 0.1302H20Accidents, etc. . . . . . . . . . . . . . 0.432 0.079 0.037 4.61 0.0353H23 Fair health . . . . . . . . . . . . . . . . 0.135 0.092 0.044 4.28 0.0423H24 Poorhealth . . . . . . . . . . . . . . . 0.053 0.243 0.077 10.04 0.0023H28 Bed days . . . . . . . . . . . . . . . . 1.955 0.069 0.017 16.35 ‘0.00011301ncome . . . . . . . . . . . . . . . . . . 10.262 –0.868 0.045 369.56 0.0000131 Partial coveragel . . . . . . . . . . . . 0.141 0.074 0.050 2.22 0.1406133 No coverage . . . . . . . . . . . . . . . 0.020 – 0.529 0.110 22.96 0.0000[35 Medicaid coverage only . . . . . . . . . 0.004 –1 .023 0.488 4.40 0,0395139 Other public & private coverage . . . . . 0.135 – 0.228 0.056 16.55 0.0001140 Other public coverage only . . . . . . . 0.004 – 0.893 0.129 47.84 0.0000G43 South region . . . . . . . . . . . . . . 0.300 0.273 0.047 33.50 0.0000G45 Metropolitan suburb . . . . . . . . . . . 0.447 0.088 0.038 5.46 0.0223

48

Intercept= 3.197 Meen of dependent variable = -3.942Number of observations = 2,874 Probability = 0.0000Multiple correlation coefficient squared (R? = 0.232 69 denominator degrees of freedomF = 52.92 with 24 degreesof freedom

NOTES Younger,better-offfamiliesarefamilieswith no member65 yearsof age or overandincomes200 percent of the poverIylevelor higher,A probabilityOf0.0021was neededfor statistical significance at the .05 level using an F test analogous to a multiple t test (see Appendix l).

Table V

SURREGR regression for the index of financially burdensome family health expenses for older familie-full model

Standard errorRegression of regression

Independent variable Mean coefficient coefficient F Value Probability

DI Head only . . . . . . . . . ...’. . . . .D2 Head-spouse change. . . . . . . . . . .D30ther change . . . . . . . . . . . . . . .D4 Children . . . . . . . . . . . . . . . . . .D5 Family size . . . . . . . . . . . . . . . .D6Age of head . . . . . . . . . . . . . . .D7Allmembers 65+ . . . . . . . . . . . .D6 Female family head . . . . . . . . . . .D9 Black race . . . . . . . . . . . . . . . .Dlo’’Other’’ race . . . . . . . . . . . . . . .Dll~spanic . . . . . . . . . . . . . . . . .D12Education of head . . . . . . . . . . . .H13 Hospitalization. . . . . . . . . . . . . .H14Discflarges . . . . . . . . . . . . . . .H15 Institutionalization . . . . . . . . . . . .H16 Death . . . . . . . . . . . . . . . . . .H17BiIth . . . . . . . . . . . . . . . . . . .H16 Cancer . . . . . . . . . . . . . . . . .H19 Heart disease, etc. . . . . . . . . . . .H20Accidents, etc. . . . . . . . . . . . . .H210ther illness . . . . . . . . . . . . . . .H22 Goodhealth . . . . . . . . . . . . . . .H23Fair health . . . . . . . . . . .. b.. . .H24 Poorhealth . . . . . . . . . . . . . . .H26 Secondary limitation . . . . . . . . . .H27 Major limitation . . . . . . . . . . . . .H28 Bed days . . . . . . . . . . . . . . . .1301ncome . . . . . . . . . . . . . . . . . .131 Partial coverage-1 . . . . . . . . . . . .132 Partial coverage2 . . . . . . . . . . . .133 No coverage . . . . . . . . . . . . . . .134 Private coverage only . . . . . . . . . .135 Medicaid coverage only . . . . . . . . .136 Medicare coverage only . . . . . . . . .137 Medicare & other public coverage . . . .139 Other public & private coverage . . . . .141 Coverage source unknown . . . . . . .G42 North Central region . . . , . . . . . .G43 South region . . .. . . . . . . . . . . .G44 West region . . . . . . . . . . . . . . .G45 Metropolitan suburb . . . . . . . . . . .G46 Nonmetropolitan urban area . . . . . .G47 Nonurban area . . . . . . . . . . . . .

0.2560.0370.1050.1080.865

68.0890.3850.2210.0930.0200.0319.8400.3880.7110.0180.0480.0020.1640.7240.2760.1720.3370.2870.2420.0800.5331.8739.5870.6200.1080.0070.0310.0040.0950.0530.0120.0300.2350.3460.2100.3570.1490.190

–0.318– 0.069

0.1420.313

–0.0010.0060.045

–0.070–0.410– 0.326–0.131

0.0190.1970.0760.435

– 0.044– 0.908

0.0930.2200.1230.0330.048

L 0.0290.0060.115

–0.0250.068

– 0.8560.024

–0.144–0.222–0.181–1.117–0.256–2.152

0.053–1.140

0.1030.2750.0090.0690.1470.107

0.1110.2180.1510.171

0.0030.0730.1280.1510.2560.2430.0090.0930.0320.2430.1970.4920.0970.1030.0650.1100.0850.0800.0800.0860.0720.0250.0520.1200.1070.9060.2860.3640.0800.2520.2370.3790.0650.0630.0900.0660.1060.098

8.210.100.693.340.003.550.380.307.381.620.294.204.505.603.210.053.400.914.573.550.090.320.130.011.780.127.04

271.540.041.800.060.403.44

10.2572.670.059.062.50

19.340.011.04

u1.921.19

0.00550.75260.34920.07211.00000.06390.54180.58780.00830.20810.59430.04430.03750.02080.07780.82250.06960.34370.03620.06370.76400.57460.71780.94250.18720.72790.00990.00000.83500.18470.81250.52800.00310.00210.00000.82000.00370.11880.00000.91850.31200.17090.2789

Intercept= 4.253 Meanof dependent variable = – 2.971Number of observationa = 849 Probability = 0.0000Multiple correlation coefficient squared (Rz) = 0.537 88 denominator degrees of freedomF = 47.26 with 43 degrees of freedom

NOTE:Standarderrorof regressionblankbecauseF test had a valueof zeroafterrounding.

49

Table VI

SURREGR regression for the index of financially burdensome family health expenses for younger, lower-income familie+full model

Standard errorRegression of regression

Independent variable Mean coefficient coe-ticient F Value Probability

DI Head only . . . . . . . . . . . . . . . . .D2 Head-spouse change . . . . . . . . . . .D30ther change . . . . . . . . . . . . . . .

D4 Children . . . . . . . . . . . . . . . . . .

D5 Family size . . . . . . . . . . . . . . . .

D6Ageof head . . . . . . . . . . . . . . .D8 Female family head . . . . . . . . . . .

D9 Black race . . . . . . . . . . . . . . . .

DIO’’Other’’ race . . . . . . . . . . . . . . .

Dll Hispanic . . . . . . . . . . . . . . . . .D12Education of head . . . . . . . . . . . .

H13 Hospitalization. . . . . . . . . . . . . .

H14 Discharges . . . . . . . . . . . . . . .

H15 Institutionalization . . . . . . . . . , . .H16 Death . . . . . . . . . . . . . . . . . .H17 Birth . . . . . . . . . . . . . . . . . . .

H18 Cancer . . . . . . . . . . . . . . . . .

H19 Heart disease, etc. . . . . . . . . . . .

H20Accidents, etc. . . . . . . . . . . . . .H210ther illness . . . . . . . . . . . . . . .H22 Good health . . . . . . . . . . . . . . .

H23 Fairhealth . . . . . . . . . . . . . . . .

H24 Poorhealth . . . . . . . . . . . . . . .H26 Secondary limitation . . . . . . . . . .H27 Major limitation . . . . . . . . . . . . .

H28 Bed days . . . . . . . . . . . . . . . .

H29Work-loss days . . . . . . . . . . . . .

1301ncome . . . . . . . . . . . . . . . . . .131 Partial coveragel . . . . . . . . . . . .132 Partial coverage2 . . . . . . . . . . . .

133 No coverage . . . . . . . . . . . . . . .

135 Medicaid coverage only . . . . . . . . .136 Medicare coverage only . . . . . . . . .

137 Medicare & other public coverage . . . .138 Medicare & private coverage . . . . . .

139 Other public & private coverage . . . . .

140 Other public coverage only . . . . . . .

141 Coverage source unknown . . . . . . .

G42 North Central region . . . . . . . . . .

G43 South region . . . . . . . . . . . . . .

G44 West region . . . . . . . . . . . . . . .

G45 Metropolitan suburb . . . . . . . . . . .G46 Nonmetropolitan urban area . . . . . .

G47 Nonurban area . . . . . . . . . . . . .

0.4220.038

0.2030.807

1.258

38.126

0.426

0.201

0.021

0.10110.802

0.334

0.589

0.0070.0070.059

0.082

0.264

0.4870.3360.3740.263

0.141

0.0950.158

2.296

1.190

9.1040.2420.145

0.092

0.117

0.0150.0010.004

0.230

0.0070.2800.234

0.349

0.221

0.3280.144

0.195

– 0.2550.011

–0.1180.079

– 0.346

0.024

– 0.234

–0.143

–0.396

0.1260.0580.203

0.066

–0.674– 0.709

0.452

0.396

0.200

0.2560.1620.2340.116

0.126

–0.2370.032

0.035

0.104

–0.5400.0030.243

–0.012

– 1.598

–0.053–0-1 98–0.170

– 0.376

–1.914

– 0.554

0.102

0.230

0.039

0.1320.167

0.064

0.206

0.1550.177

0.202

0.0050.219

0.140

0.349

Q.1750.0170.230

0.110

0.735

0.7930.269

- 0.198

0.188

0.1680.2230.1100.136

0.2570,1920.143

0.048

0.034

0.118

0.174

0.2670.2650.3260.221

0.111

0.5320.172

0.114

0.110

0.138

0.1310.153

0.133

1.53

0.000.58

0.20

2.92

25.32

1,14

1.04

1.29

0.5211.81

0.78

0.36

0.84

0.802.83

3.98

1.13

2.310.534.500.73

0.241.530.05

0.54

9.54

20.870.001.94

0.00

35.76

0.040.370.59

11.42

12.93

10.38

0.804.35

0.081.021.19

0.23

0.22060.97070.4481

0.6586

0.0918

0.0000

0.2890

0.3102

0.2593

0.47190.00100.3809

0.5510

0.36220.37330.0970

0.0500

0.2917

0.13330.47080.03750.3970

0.6234

0.22100.8250

0.4669

0.0029

0.00000.98590.1684

0.9594

0.0000

0.83950.54620.4442

0.0012

0.00060.00190.3753

0,0407

0,7824

0.31620.2801

0.6357

Intercept = 0.177 Mean of dependent variable = –3.342Numbsr of observations = 920 Probability = 0.0000Multiple correlation coefficient squared (Rz) = 0.282 69 denominator degrees of freedomF = 48.03 with 44 degrees of freedom

NOTE: Standard error of regression blank becausd F test had a value of zero after rounding.

50

interview but were part of the civilian noninstitutionalizedpopulation on Janu~ 1, 1980, were classified as “non-key” persons. Data were collected for nonkey personsfor the time that they lived with a key person but,because they had a chance of selection in the initialsample, their data are not used for general person-levelanalysis. However, data for nonkey persons are usedin family analysis because nonkey persons contributedto the family’s utilization of and expenditures for healthcare during the time they were part of the family.

Persons included in the sample were grouped into“reporting units”. for data collection purposes. Reportingunits were defined as all persons related to each otherby blood, marriage, adoption, or foster care status andliving in the same dwelling unit. The combinedNMCUES sample consisted of 7,244 eligible reportingunits, of which 6,599 agreed to participate in the survey.In total, data were obtained on 17,123 key persons.The Research Triangle Institute sample yielded 8,326key persons, and the National Opinion Research Centersample yielded 8,797.

Research Triangle Institute Sample Design

A primary sampling unit (PSU) is defined as acounty, a group of contiguous counties, or parts of coun-ties with a combined minimum 1970 population sizeof 20,000. A total of 1,686 disjoint PSU’S exhaust theland area of the 50 States. and Washington, D.C. ThePSU’S are classified”as one of two types. The 16 largeststandard metropolitan statistical areas (SMSA’S) are des-ignated as self-representing PSU’S, and the remaining.1,670 PSU’S in the primary sampling frame are desig-nated as non-self-representing PSU’S.

PSU’S are grouped into strata whose members tendto be relatively alike within strata and relatively unlikebetween strata. ‘PSU’S derived from the 16 largestSMSA’S had sufficient population in 1970 to be treatedas primary strata.:The 1,659 non-self-representing PSU’Sfrom the continental United States were stratified into59 primary strata with approximately equal populations.Each of these primary strata had a 1970 population ofabout 31/3million. One supplementary primary stratumof 11 PSU’S, with a 1970 population of about 1 million,was added to the Research Triangle Institute primaryframe to include Alaska and Hawaii.

The total first-stage sample for Research TriangleInstitute consisted of 59 PSU’S, of which 16 were self-representing PSU’S. The non-self-representing PSU’Swere obtained by selecting one PSU from each of the43 non-self-representing primary strata. These PSU’Swere selected with probability proportional to 1970 popu-lation size,

In each of the 59 sample PSU’S, the entire PSUwas divided into smaller disjoint area units called second-ary sampling units (SSU’S). Each SSU consisted of oneor more enumeration districts or block groups definedby the 1970 census. Within each PSU, SSU’S were

ordered and then partitioned to form secondary strataof approximately equal size. Two secondary strata wereformed in the non-self-representing PSU drawn fromAlaska and Hawaii, and four secondary strata wereformed in each of the remaining 42 non-self-representingPSU’S. Thus, the non-self-representing PSU’S were par-titioned into a total of 170 secondary strata. In a similarmanner, the 16 self-representing PSU’S were partitionedinto 144 secondary strata.

In the second stage of selection, one SSU wasselected from each of the 144 secondary strata coveringthe self-representing PSU’S, and two SSU’S were selectedfrom each of the remaining secondary strata. All second-stage sampling was with replacement and with probabilityproportional to the SSU’S total noninstitutionalized popu-lation. The total number of sample SSU’S was 2 x 170+“144 = 484.

For the third stage of selection, each SSU was firstdivided into smaller disjoint geographic areas, and onearea within the SSU was selected with probability propor-tional to the total number of housing units in 1970.Next, one or more disjoint segments of at least 60 housingunits were formed in the selected area. One segmentwas selected from each SSU with probability proportionalto the segment housing unit count. In response to thesponsoring agencies’ request that the expected householdsample size be reduced, a systematic sample of one-sixthof the segments was deleted from the sample. Thus,the total third-stage sample was reduced to 404 segments.

For the fourth stage of selection, all of the dwellingunits within the segment were listed, and a systematicsample of dwelling units was selected. The proceduresused to determine the sampling rate for segments guaran-teed that all dwelling units had an approximately equaloverall probability of selection. All of the reporting unitswithin the selected dwelling units were included in thesample.

National Opinion Research Center Sample Design

The land area of the 50 States and Washington,D.C., was also divided into disjoint PSU’S for the Na-tional Opinion Research Center sample design. A PSUconsisted of SMSA’S, parts of SMSA’S, counties, partsof counties, or independent cities. Grouping of countiesinto a single PSU occurred when individual countieshad a 1970 population of less than 10,000. The PSU’Swere classified into two groups according to metropolitanstatus-SMSA or not SMSA. These two groups wereindividually ordered and then partitioned into zones witha 1970census population size of approximately 1 million.

A single PSU was selected within each zone witha probability proportional to its 1970 population. Itshould be noted that this procedure allowed a PSU tobe selected more than one time. For instance, an SMSAprimary sampling unit with a population of 3 millioncould be selected as many as four times. The full general-purpose sample contained 204 PSU’S. These 204 PSU’S

53

were systematically allocated to four subsamples of 51PSU’S. The final set of 76 sample PSU’S was chosenby randomly selecting two complete subsamples of51PSU’S. One subsarnple was included in its entirety, and25 of the PSU’S in the other subsample were selectedsystematically for inclusion in NMCUES.

For the second stage, each PSU selected in the firststage was partitioned into a disjoint set of SSU’S definedby block groups, enumeration districts, or a combinationof the two types of census units. Within each samplePSU, the SSU’S were ordered and then partitioned into18 zones such that each zone contained approximatelythe same number of households. One SSU had the oppor-tunity to be selected more than once, as was the casein the PSU selection. If a PSU had been hit more thanonce in the first stage, the second-stage selection processwas repeated as many times as there were fiist-stagehits. The 405 SSU’S were identified by selecting 5 SSU’Sfrom each of the 51 PSU’S in the subsample that wasincluded in its entirety and 6 SSU’S from each of the25 PSU’S in the group for which only one-half of thePSU’S were included.

The SSU’S selected in the second stage were thensubdivided into area segments with a minimum sizeof 100 housing units each. One segment was then selectedwith probability proportional to the estimated numberof housing units. The final-stage sample, in which aselection of housing units was made, was essentiallythe same as that used by the Research Triangle Institute.

Collection of Data

Field operations for NMCUES were performed by theResearch Triangle Institute and the National Opinion Re-search Center under specifications established by thesponsoring agencies. Persons in the sample dwelling unitswere interviewed at approximately 3-month intervals be-ginning in February 1980 and ending in March 1981. Thecore questionnaire was administered during each of thefive rounds of interviews to collect data on health, healthcare, health care charges, sources of payment, and healthinsurance coverage. A summary of responses was used toupdate information reported in previous rounds. Supple-ments to the core questionnaire were used during the first,third,, and fifth rounds of interviews to collect data thatwere not expected to change during the year or that wereneeded only once. Approximately 80 percent of the thirdand fourth rounds of interviews were conducted by tele-phone; all remaining interviews were conducted in per-son. The respondent for the interview was required to be ahousehold member 17 years of age or older. A proxy re-spondent not residing in the household was permittedonly if all eligible household members were unable to re-spond because of health, language, or mental condition.

Imputation

Nonresponse in panel surveys such as NMCUESoccurs when sample individuals refuse to participate inthe survey (total nonresponse), when initially partici-pating individuals drop out of the survey (attrition nonre-sponse), or when data for specific items on the question-naire are not collected (item nonresponse). In general,response rates for NMCUES were excellent. Approxi-mately 90 percent of the sample reporting units agreedto participate in the survey, and approximately 94 percentof the individuals in the participating reporting unitssupplied complete annual information. Even though theoverall response rates are quite high for NMCUES, theestimates of means and proportions may be biased ifnonrespondents have different health care experiencesthan respondents or if there is a substantial responserate differential across subgroups of the target population.Furthermore, totals will tend to be underestimated unlessallowance is made for the loss of data because ofnonresponse.

Two methods commonly used to compensate forsurvey nonresponse are data imputation and the adjust-ment of sampling weights. For NMCUES, imputationwas used to compensate for attrition and item nonre-sponse, and weight adjustment was used to compensatefor total nonresponse. The calculation of the weightadjustment factors is discussed in the section on sam-pling weights.

A specialized form of the sequential hot-deck imputa-tion method was used for attrition imputation. First,each sample person with incomplete annual data (recip-ient) was linked to a sample person with similar demo-graphic and socioeconomic characteristics who had com-plete annual data (donor). Second, the time periods forwhich the recipient had missing data were divided intotwo categories, imputed eligible days and imputed ineli-gible days. Imputed eligible days were those days forwhich the donor was eligible (that is, in scope), andimputed ineligible days were those days for which thedonor was ineligible (that is, out of scope). For therecipient’s imputed eligible days, the donor’s medicalcare experiences (such as medical provider visits, dentalvisits, or hospital stays) were imputed into the recipient’srecord. Finally, the results of the attrition imputationwere used to make the final determination of a person’srespondent status. If more than two-thirds of the person’stotal eligible days (both reported and imputed) wereimputed, then the person was considered to be a totalnonrespondent, and all data for the person were removedfrom the analytic data file.

The data collection methodology and field qualitycontrol procedures for NMCUES were designed so thatthe data would be as accurate and complete as possiblesubject to budget considerations. However, individuals

54

cannot report data that are unknown to them, or theymay choose not to report the data even if known. Thislatter situation is especially true for data relating to ex-penditures, income, and other sensitive topics. Becauseof the size and complexity of the NMCUES data base,it was not feasible, from the standpoint of cost, to replaceall missing data for all data items. The 12-month datafiles, for example, contain approximately 1,400 dataitems per person. With this in mind, the NMCUESapproach was to designate a subset of the total itemson the data base for imputation of the missing data.Thus, for 5 percent of the NMCUES data items, theresponses were edited and missing data imputed by acombination of logic and hot-deck procedures to producerevised variables for use in analysis. Items for whichimputations were made cover the following data areas.●

●

●

●

●

●

●

✘

●

●

●

●

●

●

●

Visit charges.

Source of payment codes and amounts.

Annual disability days.

Health insurance premium amount.

Length of hospital stay.

Total weeks worked in 1980.

Average hours worked per week.

Educational level.

Hispanic ethnicity.

Income.

Age and birth date.

Race.

Sex.

Health insurance coverage.

Visit dates.

These items were selected as the most important variablesfor statistical analyses.

Construction of Longitudinal Families

At the time of the initial interview, a group of personssharing a common housing unit was designated a familyif they were related to each other by blood, marriage,adoption, or a formal foster care relationship. An unmar-ried student 17–22 years of age living away from homewas also considered a part of the family, even thoughhis or her residence was in a different location. When,on subsequent interviews, this initial sampled social unitwas found to have had changes in membership, it becamenecessary to find a decision rule (or set of decisionrules) for deciding when a family continued, when itended, and when anew family began.

The decision rule chosen was initially referred toas a principal-predecessor-principal-successor rule(Dicker and Casady, 1982; Whitmore, Cox, and Folsom,1982; Moser et al., 1983). The term came from the

understanding that, at any given point in time, a familymay have several predecessor families from which itsmembers came and several successor families into whichits members would go. The decisionmaking problem,therefore, was to objectively select only one predecessorfamily (the principal predecessor) and only one successorfamily (the principal successor) as representing the familythrough successive stages in time. If no principal succes-sor family could be found, the initial family had ended.If no principal predecessor family could be found, thecurrent family (at the time of the interview) was a newfamily. Later discussions in the literature referred tothe above rule under a different name. It came to becalled a “reciprocal, majority population rule” (McMilJIen, 1984; Dicker, 1984) because the principal-predeces-sor-principal-successor rule came to be understood asa rule that linked families on the basis of cross-familymajorities. Thus, if two families (as defined above) existat different but adjacent points in time, they are thesame family if and only. if a majority of the eligiblemembers of the first family are found in the secondfamily and a majority of the eligible members of thesecond family are also found in the first family. Thereciprocity of the comparison is crucial. A unidirectionalmajority-either from the first family to the second fam-ily or from the second to the first—is not sufficientfor the two families to be defined as the same.

Several aspects of the rule as applied in this surveyneed further elaboration. First, the rule was applied toall families in the longitudinal universe (not only tothose in the initial sample) that had cross-membershipconnections with initially sampled families. Second, onlypersons eligible over time to be in both families beingcompared were counted when calculating cross-familymajorities. For example, persons in family 1 who diedor otherwise left the universe were not eligible for mem-bership in family 2 and were not counted. Likewise,persons who entered family 2 from outside the universeduring the interval between interviews, such as a newbornbaby or a soldier returning to civilian status, could nothave been in family 1 (that is, were not eligible forinclusion in that family) and also were not counted.Third, the reciprocal majority population rule, as statedabove, links only two families adjacent in time. How-ever, transitivity between linkages is implied in the rule.This means that given three families (families A, B,and C) existing at three different points in time, iffamily A is the same as family B and family B isthe same as family C, then family A is also the sameas family C. A longitudinal family, therefore, is eitherone or a series of point-interval families linked by thereciprocal majority population rule. Fourth, the finalsample of families was limited to initially sampledfamilies and all other families derived from these familiesthat had at least one initially sampled person (a keyindividual) in them on their beginning date. Thus, thecollection of families examined for family constructionpurposes was divided into key families (a family with

55

a key individual), which were in the sample and givena positive sampling weight, and nonkey families (a familywithout a key individual), which were not in the sampleand given a sampling weight of zero. One reason fornot including nonkey families in the sample is that verylittle data for them were available. Moreover, assump-tions were often required to construct these families.(For more details on this methodology, see Dicker andCasady, 1982, and Whitmore, Cox, and Folsom, 1982.)

The dynamic sample of longitudinal families derivedfrom this process tended to have characteristics that aregenerally sociologically believed to define the beginningand ending of families. For example, an even mergerof two individuals through marriage always produceda new family. Similarly, an even split in a two-personfamily as the result of divorce or separation alwaysended the family. On the other hand, an uneven splitin a larger family would not necessarily end such afamily, In most cases, the original family continuedas the larger part”of the split. For example, if an adultchild left a family of three persons or more to set upa separate household, “in most cases the original familycontinued as the same but smaller family. Such an out-come appears to be in agreement with the sociologicalconsensus that the loss of a single family member, otherthan the head or spouse, does not usually end the originalfamily. The majority of uneven spli~ arise from thistype of situation.

By the same reciprocal majority rule, however, aseparation of husband and wife in a situation wherechildren remained with one of the spouses in most casescontinued the old family, now reconstituted as a single-spouse family with children. This result may not appearto be the sociologically prefened one. However, a moredetailed review of the class of events of which thisis a special case suggests that this result is in line bothwith the results based on sampling criteria for othermembers of the class and with sociological expectationsof what the result should be for those class members.For example, given a head-spouse family with children,the loss of a head or spouse because of death or in-stitutionalization is rarely thought of sociologically asan event ending the family. Rather, the social consensusappears to be that the original family continues, althoughin a recognizably changed state. The same may be saidfor the situation in which a head or spouse enters themilitary or goes overseas and is absent from the familyfor long periods. The family is not defined as endedbut as continuing with an absent spouse. In this survey,all of the above events are defined as out-of-scope sam-pling events that cannot affect the identity of the familyover time. Therefore, families would not end becauseof their occurrence. Only when the separating head orspouse remains within the noninstitutionalized U.S.population (the universe of inference) does the dilemmaarise from sampling and sociological considerations as

to whether the original family has ended. This inscopeevent, however, is similar in its effect on family function-ing as the four previously mentioned out-of-scope events.In all of these situations, the family loses a significantrole player. As a consequence, important family roleobligations go unfulfilled (or only partially fulfilled).It seemed appropriate, therefore, to treat all of theseevents in the same manner (as a functionally equivalenthappening) for the purpose of constructing longitudinalfamilies. Given the lack of a sociological consensusfor treating the above class of events, the reciprocalmajority population nde produces an appropriate, if notconsensual, decision. When the separating head or spouseor adult child remains within the universe, the reciprocalmajority population rule must also be applied to findout if he or she has formed a new family. The decisionwill depend on whether the person joins a previouslyexisting family in the universe and the size of the familyjoined.

An uneven merger of two preexisting families alsopresents some decisionmaking problems from a sociolog-ical perspective. Suoh mergers occur when one or morerelated persons join another set of related persons orwhen a marriage occurs and one or more of the marriagepartners bring children from a previous marriage (oranother related person) with them. The first type ofsituation presents few problems. Most of these casesinvolve the entering or reentering of continuing familiesby elderly parents, adult children, or other relatives.Usually these new family members constitute the smallerof the two merging families. The larger of the twofamilies entering the merger generally has reciprocalmajority linkages to the newly merged family, (Thesmaller family never has.) The two reciprocally linkedfamilies are considered one continuing family. Occasion-ally, an uneven merger may produce a totally new familyif the merged family cannot be linked to any preexistingfamily. The above result appears to be in line withthe general sociological consensus that a family’s identityis not changed by the addition or return of elderly parents,adult children, etc. Of course, if the additional familymembers come from out of scope (that is, if they arenewborn children, come out of an institution, or returnfrom the military or from overseas), they do not affectthe identity of the family. These instances probably repre-sent the majority of uneven mergers. However, thereis less sociological consensus as to what the mergedfamily represents when an uneven merger results froma marriage. The reciprocal majority population rule treatsthis situation in the same manner as the preceding one.For situations in which a single spouse enters an alreadyexisting larger family, the result appears appropriate.Where both spouses bring large families into the mar-riage, the result may be questionable. However, theselatter situations represent a very small number of cases.

56

Construction and Use of Family Weights

Initial Family Weights

The target population of the household survey (HHS)was civilian noninstitutionalized families existing in theUnited States at any time during 1980. The universeof families existing on any specific day during 1980was potentially different from that existing on any otherday of the year. Conceptually, one could have conducteda census of the eligible population of the United Stateson January 1, 1980. By following this initial universeof families throughout the year, every unique longitudinalfamily unit could be identified and labeled. These lon-gitudinal family units are defined by a beginning date,an ending date, and a set of persons who qualify aseligible (civilian and noninstitutionalized) family mem-bers. In addition to all family units that can be linkedto the initial January 1 family universe, there are personsand families who were ineligible on January 1, 1980,but subsequently returned to the civilian nonin-stitutionalized population without merging with familiescontaining individuals who were eligible on January 1.Such individuals and families were eligible for the samplebut did not have a chance of entering it. Poststratificationweight adjustments partially compensated for thisundercoverage.

The family weights for longitudinal families in thehousehold sample were developed from the samplingweights for the initially sampled families, which werecalled originating base reporting units (OBRU’S). Foreach HHS longitudinal family, the key family membersall belonged to the same OBRU. Hence, the initial familyweight for the jtk key HHS longitudinal family was com-puted as follows.

where n~) is the number of key individuals in familyj on its beginning date, g(j) is the total number ofmembers of family j on its beginning date, and wO~]is the OBRU initial sampling weight for the key membersof family j. Thus, the initial family weight is the OBRUsampling weight adjusted for person-level multiplicity.Essentially, this formula means that the sampling weightof a family beginning on January 1, 1980, is the sameas the household sampling weight, regardless of whenthe family ended or family membership changed in thesubsequent 12 months. However, if a family began onsome day after January 1, 1980, the household samplingweight was adjusted to take into account the fact thatthe new family may have had multiple chances of gettinginto the sample. However, as previously pointed out,positive sampling weights were developed only for keylongitudinal families. Further details of the methodology‘for HHS longitudinal sampling weightsby Whitmore, Cox, and Folsom (1982).

are provided

Adjustment for Undercoverage and Nonresponse

Poststratification adjustment of the initial HHSfamily weights to the family counts based on theMarch Supplement to the 1980 Current Population Sur-vey (CPS) was used to reduce the variance .of es-timators and the bias from undercoverage. Thesecounts, however, were from estimates based on an up-dating of the 1970 census. Therefore, NMCUES fam-ily counts and estimates may not agree with familycounts and estimates based on the 1980 census. Thepoststratification adjustments and a weighting class ad-justment were also used to reduce the bias from nonre-sponse of longitudinal families.

A key HHS longitudinal family was classified asresponding if it satisfied the following three require-ments.

1.

2.

3.

At least one key family member was classified as arespondent; that is, at least one key family memberresponded for at least one-third of his or her eligi-ble days in the survey.

The total number of responding (known eligible)days during the family’s existence summed over allfamily members is at least one-third of the totalnumber of eligible days during the family’s exist-ence summed over all members of the family.

The family contained no students who were listedonly on the parents’ round 1 secondary reportingunit roster and for whom no other data collectioninstrument was ever received.

This definition of a responding family was felt to beconsistent with the definition of person-level responseand was used to create the HHS family response indi-cator variable. Only about 0.1 percent of all longitudi-nal families were declared to be nonresponding be-cause of condition 3. Imputation of a full year of datafor these students was problematic. Hence, inclusionof condition 3 in the definition of a responding familywas felt to be cost effective.

The initial multiplicity-adjusted family weight wascomputed for all longitudinal families from the initialOBRU weight. A poststratification adjustment wasthen made for nonresponse of families linked to nonre-sponding OBRU’S, producing an adjusted weight. Aweighting class adjustment was performed for nonre-sponding longitudinal families generated by respondingOBRU’S. This adjusted weight was then truncated toproduce a new family weight. The final adjustmentwas a poststratification and smoothing to the MarchCurrent Population Survey family counts to producethe final HHS longitudinal family weight, FWEIGHT.An alternative family weight, AWEIGHT, which wasadjusted for each family’s eligible days, was also cGm-puted from FWEIGHT to facilitate analytic tabulations.AWEIGHT, a time-adjusted familyFWEIGHT times the proportion of

weight, is equal to1980 for which the

57

family existed. (Computationally, it equals FWEIGHTtimes the family’s survey eligibility days divided by366, the total number of days in 1980.) The time-adjusted family weights, AWEIGHT, sum to the aver-age daily number of HHS-eligible longitudinal familiesin the United States in 1980.

Estimators

This family weighting scheme produces the adjustedfamily weight, FWEIGHT, which can be used directlyfor estimation of annual health care utilization and expend-iture. For example, if Y(j) represents the total expenditureof thejz~ HHS longitudinal family for a particular medicalservice in 1980, then

ZFWEIGHTQ)Y~>

estimates the total expenditure of all civilian nonin-stitutionalized families in the United States for this medi-cal service in 1980, where the summation extends overall longitudinal families in the NMCUES HHS sample.

Rates of utilization and expenditure are, however,of more interest than population totals. The rates ofannual utilization and expenditure per family for a givenfamily domain, say domain d, are defined at the popula-tion level by

R(d) = [ $, X~(j>Y(j>]/[ ,.l X~(j]PE~]],

wherej =

x&) =

Y(j)=

PE~> =

19---> J indexes the population of all keylongitudinal families that ever existed in 1980(that is, all longitudinal families that had achance for selection as key NMCUESfamilies);

1 if familyj belongs to domain d,Ootherwise;

total utilization or expenditure for family jduring the portion of 1980 that @ily j waseligible for NMCUES; and -

proportion of 1980 that family j was eligiblefor NMCUES, or (FAMEND – FAMBEG+ 1)/366, where FAMEND = family endingdate (days of 1980 numbered 1 through 366)and FAMBEG = family beginning date.

The family aggregates, YQj, can be viewed as sumsof associated person-level visit counts or expendituresfor key and nonkey individuals belonging to family jduring the time period in which they were membersof the family. The denominator of R(d) is the averagedaily number of families of type d that existed during1980. The bracketed portion of the numerator of R(d)is simply the total number of health care visits or thetotal expenditures of a specified type experienced byNMCUES eligible persons while they belonged tofamilies of typed.

58

Unbiased estimators for the numerator and de-nominator of R(d) lead to the ratio estimator r(d), forwhich the equation is:

r(d) = [XFWEIGHT~>X~)Y~)] /[ZFWEIGHT(j)X.(j]PE~~] ,

where the summation extends over all longitudinalfamilies in the sample. Of course, it is necessary tocompute X&) and PE~~ only for responding familiesbecause FWEIGHT is zero for all other families. Twoalternative formulations of this estimator that may bemore convenient for some computations are:

r(d) = [XAWEIGHT~)X&~Y~~/PE~J] /[XAWEIGHT(j]Xfi)],

and

r(d) = [XFWEIGHT~)X~(j)Y~’] /[XAWEIGHT~)X.(j]] ,

where the summations extend over all longitudinalfamilies and AWEIGHT~], as previously noted, is thefinal time-adjusted weight for familyfi that is,

AWEIGHT~] = FWEIGHT~) PE~).

Throughout this report, all estimates are based onthe first of these two alternative formulations. All countsof expenditures for health care employ as the measureof expenditure used

XAWIGHT~JX~~)Y(j)/PE~~ ,

and all counts of families employ as the number offamilies in question

XAWEIGHT(j)X~(J) .

To be more specific, the statistics presented in thedetailed tables of this report are estimated as follows,

The number of families with given characteristic(s)is estimated as

ZAWEIGHT~]X&),

whereX~~J = 1 if family j has the characteristic(s) inquestion and Ootherwise.

Note that this estimator estimates the number of familyyears experienced by families with the given characteris-tic(s) or, equivalently, the average number of familieswith the given characteristic(s) that would have beenfound at a randomly chosen point in time in 1980. Itis, in general, less than the cumulative total of distinctlongitudinal families with the given characteristic(s) thatever existed at any time in 1980Ysome of which existedfor only part of the year,

The mean for use or expenditure is always the meanrate per family year and is estimated as

The percent of families with a given characteristicis estimated as

where XU~] =1 if family j has the given characteristicand Ootherwise.

Note that this estimator has as its denominator the esti-mated number of family years experienced by all familiesin a domain defined by a set of family characteristicsand has as its numerator the estimated number of familyyears experienced by families in the domain that alsohave the utilization characteristic in question. In otherwords, the estimator involves a ratio of family years.

Special Requirements for Imputationof Family Data

As noted in the previous section, estimation of utili-zation and expenditure rates requires family aggregatedata, say Y~], where the aggregates can be obtainedas sums of associated person-level visit counts or expend-itures. To compute the family aggregate Y~], it is neces-sary to sum over all members of family j, both keyand nonkey. Moreover, computation of annual utilizationand expenditure statistics requires a full year of datafor every member of each responding family. Hence,in the attrition imputation, a weighted sequential hot-deckprocedure was used to produce complete data for indi-viduals who did not respond for the full year. In theattrition task (Cox and Sweetland, 1982), each individualwas first classified as either having complete data orhaving incomplete data, based on whether the individualhad responded for all 366 days in 1980. The data recordsfor individuals who had not responded for the full yearwere completed by attrition imputation, including impu-tation of eligibility status (eligible or ineligible) for eachday in 1980. The,major importance of the attrition taskis that it provided a full year of data for every individualfrom which family aggregates, Y~), can be computed.The concept of a key responding family was definedin such a way, however, that minimal use of data fromthe attrition task is required. Of course, missing itemdata can also lead to missing values for the family aggre-gate, Y~). Hence, item imputation procedures (Coxet al., 1982) were performed in addition to attritionimputation to assure the availability of complete datafor important analytical variables for every eligible dayfor each family member.

Reliability of Estimates

Standard Errors

The estimates presented in this report are based ona sample of the target population rather than on theentire population. Thus, the values of the estimates maybe different from values that would be obtained froma complete census. The difference between a sampleestimate and the population value is referred to as thesampling error, and the expected magnitude of the sam-pling error is measured by the standard error. Estimatedstandard errors for the estimates in Tables A–F are gener-ally next to each estimate.

The SESUDAAN (Shah, 1981) standard error esti-mation software package was used to produce the esti-mates of standard errors. SESUDAAN is a Taylor Seriesprocedure, developed and released by the ResearchTriangle Institute. It runs within the Statistical AnalysisSystem (SAS Institute, Inc., 1982).

In addition to sampling errors, the estimates pre-sented in this report are subject to nonsampling errors,such as biased interviewing and reporting, undercover-age, and nonresponse. The standard error does not pro-vide an estimate of nonsampling errors. However, asdiscussed in preceding sections, every effort was madeto minimize these errors.

Confidence Intervals

The estimates in this report are subject to samplingerror. The true values are unknown. But the samplingerror can ‘be used to determine a range of values suchthat the true value will, be within that range with aknown probability. This range is called a confidenceinterval.

Suppose that @isan unbiased estimator for the param-eter 0, an~ S~ is a consistent estimator for the standarderror of, 0. Under appropriate central limit theoremassumptions regarding d, the statistic Z = (~ – 0)/S~has an approximate standard normal distribution for largesamples. Thus, an approximate (1 – a) X 100 percentconfidence intefial for f3is given by

where zd2 and Z1–d2 are the appropriate values froma standard normal table.

As an example, Table A shows that, of all multiple-person families in the civilian noninstitutionalized popu-Iatiofi of the United States, an estimated 17.5 percenthad total charges for health care of $3,000 or morein 1980. The estimated standard error of 17.5 percentis .60 (Table A). As Z.ox = —1.96 and 2.975 =1.96, a 95-percent confidence interval for the percentof all multiple-person families with such charges in 1980is 17.5 & (1.96 X .60), or the interval 16.3–18.7percent. Approximately 95 percent of the confidence

59

intervals constructed in this manner will contain thetrue percent of families with total charges for healthcare of $3,000 or more in 1980.

Confidence intervals for the difference of two param-eters can be constructed in a similar manner. Suppose01 and f32are the values of the parameter of intere~tin tw~ mutually exclusive population subgroups. If 61and 62 are ~nbia:ed es~mators of 61 and 62, respec-tively, then d = O, – 132is unbiased ford = 131– 192and

var(~– Var(b,)+ Var(dz)– 2 cov(o~ ,42).

Unfortunately, the estimation of Var(~) presents aproblem because it is not possible for the National Centerfor Health Statistics to provide the reader with covarianceestimates for all possible pairs of subdomains of potentialinterest. However, if it is reasonable to assume thatCov(dl ,~2) = O, the standard error of d can be estimatedby

SJ= -.

Then, under appropriate central limit :heorem assump-tions regarding d, the statistic Z~= (d – d)/SJ has anapproximate standard normal distribution for large sam-ples, and the interval

is an approximate (1 —a) X 100 percent confidenceinterval for the difference d.

For example, suppose we wanted to construct a 95-percent confidence interval for the difference betweenthe percent of older families with total charges for healthcare of $3,000 or more (f?l)and the percent of younger,lower-income families with the same amount of totalcharg:s (Oz). It can be seen in Table A that 81 = 26.9and 02 = 15.5, so

2=8,–82

= 26.9 – 15.5

Then, as a= .05, it follows that 2.12= -1.96 andZI -a12 = 1.96, so the 95-percent confidence interval forthe difference of interest is (15.48, 7.32).

T~e :eader should be aware that the assumption thatCOV(O,,02)= O is frequently not true for complexsample surveys. ,This warning is especial]y germane forsample designs, such as the NMCUES design, that relyon cluster sampling :t ~ne or more stages of sampleselection. If COV(O1,6Z) is positive, the confi-dence interval will tend to be too large, and the confi-dence I:vel will be understated. More seriously, ifCOV(O1,02) is negative, the confidence interval willtend to be too small, and the confidence level will beoverstated.

Hypothesis Testing

The statistics Z and Zdcan be used to test hypotheses,For example, the size a critical region for the compositehypothesis

Ho: d>do

versus

HA: d<do

is given by

As an example, suppose that before any data werecollected one had a reason to believe that the percentof younger, lower-income families with total chargesfor health care of $5,000 or more (01) was less thanthe percent of older families with the same amount oftotal charges (Oz). Letting d = 191- Oz, this can berestated as a formal hypothesis as

Ho:d?O

versus

HA:d<O.= 11.4.

From Table A, it can be seen that S~l = 1.7and S~2= 1.2, so

Note that what is believed to be the true state of natureis reflected by the one-sided alternative.

It can be seen from Table A that

d = 8.0 – 17.8= –9.8

=~2.89+ 1.44

=m

= 2.08.

60

and

= 1.75,

so that Z~O= – 5.60. As there are three categories forfamily socioeconomic status in Table A, a multiple ttest based on the Bonferroni inequality (Levy andLemeshow, 1980) will be used to assess the significanceof the comparison. Comparing three categories, two ata time, and not taking sign into account gives threepossible comparisons. Use of the table in Levy andLemeshow (1980, p. 296) gives a one-tail critical valueof – 2.13, Therefore, HO is rejected in favor of HAas zdo~ Za.

A: . discussed earlier, the assumption thatCov(@l,62) = O must be carefully evaluated. If infact the covariance is positive, the size of the test willbe smaller than a, and if the covariance is negative,the size of the test will be larger than a. Readers whowant to conduct more sophisticated analysis of theNMCUES data are advised to consult with a statisticianknowledgeable in the analysis of data from complexsample surveys.

Appendix IllDefinition of Terms

Accidents, injuries, and poisonings—This categoryincludes injuries; wounds; burns; poisonings; toxic ef-fects; complications of medical and surgical care; andearly complications, late effects, and impairments dueto the previous causes. Conditions were reported bythe family respondent and recoded according to the Inter-national Classification of Diseases, 1975 revision.

Age offamily head—Age is as of January 1,1980.Ambulatory physician visit—A visit by a patient to

a physician’s office, clinic, or similar place is an ambula-tory physician visit. Visits are counted whether a physi-cian or only a member of the physician’s staff is seen.House calls and visits to school or workplace clinicsare also included. Family visits are the sum of all visitsby family members during the time they were in thefamily.

Bed days—Bed days are days spent in bed by afamily member because of illness or injury. Family beddays are the sum of all bed days of family membersduring the time they were in the survey, prorated tothe time they were in the family.

Cancer and other neoplasms—This category in-cludes malignant neoplasms of all sites and tissues, ‘be-nign neoplasms, carcinoma in situ, and other and un-specified neoplasms. Conditions were reported by thefamily respondent and recoded according to the Intern-ationalClassification of Diseases, 1975 revision.

Circulatory and heart disease—This category in-cludes rheumatic fever, rheumatic heart disease, hyper-tensive disease, ischemic heart disease, diseases of thepulmonary circulation, other forms of heart disease, cere-brovascular disease, and other diseases of the circulatorysystem. Conditions were reported by the family respond-ent and recoded according to the International Classifica-tion of Diseases, 1975 revision.

Civilian noninstitutionalized family—This refers tofamilies in which all members are members of the civiliannoninstitutionalized population. Families whose headsare members of the military are defined as not beingcivilian families and are excluded in their entirety fromthis report, although they were included in the sampleand the weighting. In the sample, there were 49 suchfamilies (about 0.7 percent). Family members other thanthe head who were in the military were excluded fromthe survey even if they resided with the family.

Dental visit—A visit to a dentist’s office is a dental

visit. A dentist or a member of the dentist’s office staffmay have provided services. Family visits are the sumof all visits by family members during the time theywere in the family.

Education of family head—The years of school com-pleted by family heads 17 years of age and over constitutethe education of family heads. Only years completedin regular schools, where persons are given a formaleducation, are incIuded. A “regular” school is one thatadvances a person toward an elementary or high schooldiploma or a college, university, or professional schooldegree. Thus, education in vocational, trade, or businessschools outside the regular school system was not countedin determining the highest grade of school completed.

EthniciQ of family head—The ethnicity of familyheads 17 years of age and over is as reported by thefamily respondent. The ethnicity of family heads under17 was imputed. Ethnicity is classified as (1) Hispanic,which includes Puerto Rican, Cuban, Mexican,Mexicano, Mexican American, Chicano, other LatinAmerican, and other Spanish or(2) non-Hispanic.

Family—A family is a group of people who sharea common housing unit and are related to each otherby blood, marriage, adoption, or a foster care relation-ship. An unmarried student 17–22 years of age livingaway from home is also considered part of a familyeven though his or her residence was in a differentlocation. The group of people who compose the familymay change composition over time, causing the familyto take on one or a combination of the following time-related states: existing over time without change in mem-bership; existing over time with change in membership;going out of existence before the end of the survey;coming into existence after the beginning of the survey;or existing for the whole survey. For more detail, seeAppendix II.

Family dynamics—A family is considered unchang-ing, or static, if it existed for the whole of 1980 andits membership was unchanged. Families that hadchanges in membership and/or did not exist for the wholeof 1980 are considered changing, or dynamic, families.

Family illness days in bed-See bed days.Family income in 1980-For each person in the

family, data were collected on 12 categories of in-come. These included income from employment forpersons 14 years of age and over; income from various

62

government programs; income from pensions; alimonyor child support; interest income; and net rental income.When information was missing, income was imputed.The total income of persons who were members of morethan one family was allocated to each family they werein, in proportion to the amount of time they were inthat family. Person-level incomes in each family weresummed to create a family-level total. If a family didnot exist for an entire year, the family income wasadjusted to an annual basis by dividing actual incomeby the proportion of the year the family existed.

Family-paid premiums—The amount paid for pre-miums by a family and not reimbursed. Much of thecost of premiums in the United States is paid by employ-ers or by public funds rather than by families.

Family size—The time-weighted average number ofpersons in a family determines the size. Family sizewas computed by (1) summing the number of days inthe family for each person who was ever a family memberand (2) dividing this sum by the number of days thefamily was in existence. For example, if a family existedfor 200 days and had two persons who were membersthroughout its existence and one person who was a mem-ber for 80 days, the family size is 2.4.

Family structure—Family structure refers to the pres-ence or absence of family head, spouse, and childrenunder 17, and whether these persons were present forthe family’s entire duration or part of its duration.

Family work-loss days-See work-loss day.Family years—Family years refers to the length of

time that a family, or a collection of families, existedas a unit of analysis in (were eligible for) the survey,as measured in units of a year or fractions of suchunits.

For an individual longitudinal family in theNMCUES sample, the number of family years equalsthe number of days the family was eligible for theNMCUES sample divided’by 366, the number of daysin 1980 (the NMCUES sample period). For such a familyweighted to represent a group of families in the NMCUESuniverse, the number of family years is AWEIQHT~),which is equal to FWEIGHTQ), the basic adjustedweight, times PE~), the proportion of the year the familywas eligible for the sample. For a group of samplefamilies, the associated number of family years is thesum of. the AWEIGHT’s. For further details and fullerdefinitions of variables, see the section on estimatorsin Appendix II.

Group quarters—This is a structure occupied byfive or more unrelated people who lived or ate together,or for whom there was neither direct access from theoutside or through a common hall nor complete kitchenfacilities. Only noninstitutional group quarters were in-cluded in the NMCUES sample frame. Each unrelatedperson in a group-quarter household was considered aseparate one-person family, unless he (or she) was astudentawayfromhome, (Seedefinitionof family.)

Head of family-A person was designated as the

family head by the respondent at the time of the firstinterview. If no head was designated or this informationwas missing, a family head was imputed. Among familiesin which the person designated as head changed overtime, the characteristics of the person who was designatedhead the longest were used for all head-of-familyvariables.

Health care coverage—Health care coverage refersto the situation in which a private insurance plan orpublic health care coverage program (Medicare,Medicaid, and so forth) can be used to pay all or apart of a family’s or person’s health care costs. Forthis report, completeness of health care coverage wascoded into four categories: (1) “full coverage,” meaningall family members had health care coverage duringtheir entire survey eligibility period; (2) “partial coverage1,“ meaning all family members had health care coverageat some time, but some or all had coverage for onlypart of their survey eligibility period; (3) “partial cover-age 2,” meaning some family members had health carecoverage for at least part of their survey eligibility period,but some never had coverage; and (4) “no coverage,”meaning no family members had health care coverageat anytime during their survey eligibility period.

For this report, a family was coded as having aparticular source of health care coverage (such as privateinsurance, Medicare, Medicaid, or a particular combina-tion of coverages) on the basis of the known coverageof family members. Only when the source of coveragewas unknown, or not assignable, for all family memberswas the family coded as having source of coverage un-known. However, the coding categories for individualsupon which the family health care coding in this reportwas based do not identify the coverage source(s) forindividuals with part-year or no coverage. Thus, mostfamilies coded as having source of coverage unknownare families with no members having full-year coverage.Also, the coding categories for individuals upon whichthe family health care coverage coding in this reportwas based are different from the categories used in aprevious family report (Dicker, 1983a) that dealt withonly a part of the survey year. As a result, there maybe differences in coverage estimates between the reports.

In this report, coverage by CHAMPUS is classifiedas coverage by private insurance.

Health care services-NMCUES includes informa-tion on eight types of health care services. These are(1) inpatient hospital care, (2) inpatient physician care,(3) ambulatory physician visits, (4) hospital emergencyroom and outpatient visits, (5) dental visits, (6) prescrip-tion acquisitions, (7)<ervices of other independent medi-cal providers such as chiropractors, speech therapists,faith healers, and psychologists (unless such providersare working as part of a physician:s staff, in whichcase their services are counted in physicians’ care) and(8) acquisition of other health care supplies such aseyeglasses, orthopedic items, hearing aids, ambulanceservices, and diabetic items. Excluded from the data

63

.- .in this report are nonprescription medicines, nursinghome care, and care in other long-term care institutions.

Hospital admission—Hospital admission is the for-mal acceptance by a hospital of a patient who is providedroom, board, and regular nursing care in a unit of thehospital, including patients admitted for childbirth. Apatient admitted to the hospital and discharged on thesame day is included as a hospital admission. A hospitalstay resulting from an emergency department visit isalso included. Family hospital admissions are the sumof all admissions of family members during the timethey were in the family.

Hospital dischafge—A hospital discharge is the for-mal release by a hospital of a patient who was providedroom, board, and regular nursing care in a unit of thehospital. A patient admitted to the hospital and dis-charged on the same day is included as a hospital dis-charge. A hospital stay resulting from an emergencyroom visit and subsequent admission of the patient isalso included. Family hospital discharges are the sumof all discharges by family members during the timethey were in the family.

Hospital emergency room—The emergency room isa facility within a hospital organized to provide medicalservices to people needing immediate medical or surgicalintervention. People receiving care in the emergencyroom may be admitted to a hospital.

Hospital emergency room visit—This is a face-to-face encounter between a patient (not necessarily ambula-tory) and a medical person in the hospital emergencyroom. Encounters by patients transported to theemergency room by police or by emergency medicalservice are included. The visit may result in a hospitaladmission. Family emergency room visits are the sumof all emergency room visits by family members dutingthe time they were in the family.

Hospital outpatient departrnent-This is a hospital-based ambulatory care facility organized to provide non-emergency medical services. Persons receiving servicesdo not receive inpatient nursing care. Examples of outpa-tient departments or clinics are pediatrics, obstetrics andgynecology, eye, and psychiatric.

Hospital outpatient department visit—This is a face-to-face encounter between an ambulatory patient anda medical person in a hospital outpatient department.The patient comes to a hospital-based ambulatory carefacili~ to receive services and departs on the same day.If more than one department or clinic was visited ona single trip, each department or clinic visited wascounted as a separate visit. Family outpatient departmentvisits are the sum of all hospital outpatient departmentvisits by family members during the&me they werein the family.

Household—This” refers to occupants of a housingunit or group quarters included in the sample. A house-h~ld can be one person, a family of related people,a number of unrelated people, or a combination of relatedand unrelated people. Therefore, a household can contain

more than one family. (See definition of family,)Housing unit—A housing unit is a group of rooms

or a single room occupied or intended for occupancyas separate living quarters. This means that (1) the occu-pants do not live and eat with any other persons inthe structure, and (2) there is either direct access fromthe outside or through a common hall or complete kitchenfacilities for the use of the occupants only.

Inpatient hospital care—This is health care providedto a patient by a hospital during the period from thepatient’s admission to the patient’s discharge. This in-cludes admissions for deliveries of babies.

Inpatient physician care—This care is provided toa patient by a physician (or a physician’s staff) duringthe period from the patiept’s admission to a hospitalto the time of the patient’s discharge from the hospital,Such care was only recorded in NMCUES if a physician’sbill, separate from the hospital bill, was rendered forsuch care. Otherwise, a patient was not regarded ashaving received inpatient physician care.

Institution—An institution is a place providing room,board, and certain other services for residents or patients,Correctional institutions, military barracks, and orphan-ages were always considered institutions in NMCUES,Places that provide long term health care were also iden-tified as institutions if they provide either nursing orpersonal care services. Certain other facilities licensed,registered, or certified by a State agency or affiliatedwith a Federal, State, or local government agency werealso defined as institutions. People residing in institutionswere not included in the household sample.

Key person-See the discussion under “SampleDesign” in Appendix II.

Limitation in major activi~—Four categories weredeveloped for classifying limitation in major activity:

1. Cannot perform usual major activity (such as work-ing, going to school, or keeping house).

2. Can perform usual major activity but limited in kindor amount.

3. Can perform usual major activity but limited in kindor amount of other activity.

4. Not limited.

People 6 years of age and over were classified intoany of the categories; children 1–5 years of age wereclassified into categories 1, 2, and 4; and children under1 year of age were classified into categories 1 and 4.In this report, categories 2 and 3 are combined intothe category “secondary limitation.” The NMCUES clas-sified persons with unknown limitations as not limited,

Longitudinal familyA longitudinal family is a fam-ily identified as the same family over a time period,It may or may not have had changes in family member-ship during the time period. (See the definition of family.)

Marital status—Marital status for each person 17years of age and over is as indicated by the householdrespondent.

64

Metropolitan area-see urban-rural location.Metropolitan center city-See urban-rural location.Metropolitan suburb-See urban-rural location.Multiple-person family--A family with an average

size of 1.5 members or more is a multiple-person family.National household component4ne component of

NMCUES,this consists of multiple household interviewswith an area probability sample of people in the nonin-stitutionalized population of the United States in 1980,

Nonkey person— See the discussion under “SampleDesign” in Appendix II,

Nonmetropolitan urban urea-See urban-rurallocation,

Nonurban area-See urban-rural location.Number of families—This refers to the average

number of families with a given set of characteristicsthat would have been found at a randomly chosen pointin time in 1980, This is equal to the number of familyyears experienced during 1980 by families with the givencharacteristics, It is, in general, Ipss than the cumulativetotal number of distinct longitudinal families with thegiven characteristics that ever existed at any time in1980, some of which existed for only part of the year.

One-person family-A family with average size lessthan 1.5 is a one-person family. More than one individualmay be involved.

Out-of-pocket expenses—This is the amount paidby a family and not reimbursed by either insurance orother health care payment programs.

Out-of-pocket expenses for health care services—This is the amount paid out-of-pocket by a family forall eight of the types of health care services coveredby NMCUES. (See definition of health care services,above. ) This does not include family-paid premiums.

Perceived health status—This is the family respond-ent’s rating on a 4-point scale of the health of a familymember compared with the health of other persons ofthe same age, as reported at the time of the first interview.The categories are “excellent,” “good,” “fair,” and“poor.” When a family consisted of only one member,this was a self-rating.

Point-interval family~A point-interval family is afamily with exactly the same family membership overa time period. A change in family membership endsone point-interval family and begins another.

Poverty status of family—Tkpoverty status in 1980was calculated by dividing the ~mily’s income in 1980by the appropriate 1980 poverty level threshold and con-verting it to a percent. For example, a,family with incomebetween two and three times the poverty level thresholdthat corresponds to its size and other characteristics wouldbe classified in the 200-299 percent category, The pov-erty level thresholds, as used by the U.S. Bureau ofthe Census, are determined by the age and sex of thefamily head and the average number of persons in thefamily. In 1980, average poverty level thresholds by%mily size (weighted for the mix of families by sexand age of head) were: 1-person, $4, 190; 2-person,

$5,368; 3-person, $6,565; 4-person, $8,414; 5-person,$9,966; 6-person, $1 1,269,7-person, $12,761; 8-person,$14,199; 9-person and larger, $16,896.

Premium—The premium is the amount paid for pri-vate health insurance or other health care coverage,

Prescription acquisition—This describes the obtain-ing of a medication by a family member requiring aprescription from a doctor or dentist, Both initial fillingsof prescriptions and refills are counted as acquisitions,Family prescription acquisitions are the sum of all acqui-sitions by family members during the time they werein the family,

Principal respondent—This is the member of thereporting unit who provided most of the informationfor the people in the reporting unit,

Proxy respondent—As used in this sukey, a proxyrespondent was a person who provided information forpeople in the reporting unit but who was not a memberof the reporting unit, A proxy respondent was usedonly when no member of the reporting unit could supplythe information because of physical or mental incapacity.

Race of family head—The race of the family headis as reported by the family respondent or imputed.Race is classified as “white,” “black,” or “other.” The“other” race category includes American Indians, Alas-kan Natives, Asians, Pacific Islanders, and people notidentified by race, The category “all other” includesthe categories “black” and “other.”

Rate perfamily year—Amount of care used or dollarsexpended by a family or group of families is dividedby the number of family years experienced by thesefamilies while eligible for the NMCUES sample, Alldata on the use of care or on health expenditure inthis report are presented in terms of a rate per familyyear. For a given family, the rate per family year equalsY~]lPE~J,

where Y(~~= use of care or expenditure during family’speriod of eligibility for NMCUES sample,and

PE~) = proportion of year family was eligible forthe NMCUES sample.

The section on estimators in Appendix II presents moredetails of calculations.

REF. DATE—The reference date was the date ofthe previous intemjew in most cases. For the first inter-view, however, it was January 1, 1980. For new persons,it was the date they joined the reporting unit. For thefinal interview, it spanned the time between the next-to-last interview and December 31,1980.

Region—The 50 States and the District of Columbiaare categorized into four regions by the U.S. Bureauof the Census. This classification by region was usedin this report. The regions and their constituent partsare as follows. NORTHEAST:Maine, New Hampshire,Vermont, Massachusetts, Rhode Island, Connecticut,

65

New York, New Jersey, Pennsylvania; NORTH CEN-TRAL: Michigan, Wisconsin, Ohio, Indiana, Illinois,Minnesota, Iowa, Missouri, North Dakota, SouthDakota, Nebraska, Kansas; SO~ Delaware, Mary-land, District of Columbia, Virginia, West Virginia,North Carolina, South Carolina, Georgia, Florida, Ken-tucky, Tennessee, Alabama, Mississippi, Arkansas,Louisiana, Oklahoma, Texas; WEST: Montana, Idaho,Wyoming, Colorado, New Mexico, Arizona, Utah,Nevada, Washington, Oregon, California, Alaska,Hawaii.

Reporting unit (RU)-A reporting unit is the basicunit for collecting data in the household componentsof NMCUES at the time of each interview. A reportingunit consisted of all~related people residing in the samehousing unit or group quarters during the reference periodcovered by a particular interview. One person couldgive information for all members of the reporting unit.

Secondary reporting unit—Unmarried students17–22 years of age usually living in a sampled householdbut away from home as fi.dl-time students were consid-ered secondary reporting units. Also, in a householdwith multiple families, the reporting unit with the largestnumber of individuals was usually designated the primaryreporting unit, and all other families were designatedsecondary reporting units.

Sex-Sex was recorded by the interviewer in theinitial NMCUES interview.

Spouse—The spouse is the person designated bythe respondent as the spouse of the head of the family.

Total charges-fiis is the full amount billed (eitheractual or imputed) to a family for all eight types ofhealth care services covered by NMCUES, whether theseamounts are paid out-of-pocket by the family, paid byhealth care coverage, or go unpaid.

Total out-of-pocket expenses—This is the amountpaid by a family for out-of-pocket expenses for healthcare services ph.Isthe amount of family-paid premiums.

Urban-rural Location—Households were identifiedas located either within or outside of a Standard Metropol-itan Statistical Area (SMSA). The definitions and titlesof SMSA’s are established by the U.S. Office of Manage-ment and Budget with the advice of the Federal Commit-tee on Standard Metropolitan Statistical Areas. House-holds located inside SMSA’S are further classified asbeing located within the SMSA’S central city (called“central city”) or not (“other”). Households located out-side of SMSA’S are classified as “urban” if they arelocated in (1) places of 2,500 inhabitants or more thatare incorporated as cities, villages, boroughs (exceptAlaska), and towns (except in New England, New York,and Wisconsin), but excluding persons living in the ruralportions of extended cities; (2) unincorporated placesof 2,500 inhabitants or more; or (3) other territo~, incor-porated or unincorporated, included in urbanized areas.Otherwise, households located outside of SMSA’S areclassified as “rural.”

Work-loss day—A work-loss day is a day on whicha person did not work at his or her job or businessbecause of a specific illness or injury. The number ofdays lost from work is determined only for persons17 years of age and over who reported that at any timeduring the survey period they either worked at or hada job or business. Family work-loss days are the sumof all work-loss days of family members during thetime they were in the survey, prorated to the time theywere in the family.

66

AuailableOnly~om theGouemment~”ntin.g Oflce...

CertifiedLfieWles

Vil SM-atica of theUlliJllia$mstatae,1sss mVdum8 U, S~ 6

“$ C.,?A”TM!*I “s “f.,!. ..0 . . . . . SE.V!CCS .9.,,. . ..!!. *,!<. . N.!!,..! *,!,! (4>!. ..!.!.>,.,,,,..

Lif Cta DICS:.

I,-

.

The NATIONAL CENTER ~R HEAL~ current and trend data by age, sex, and raceSTA~TICS has just released for sale the from 1900 to 1985, survivors to specified ages,1985 life expectancy tables. The W tables, and median length of life. The booldet isce~ed as official documents, are uW by availableordy from the Governmentlawyen as legal evidence, insmce analysts, Prindng Wce at a mst of $1.50

~h~

actuaries, pension planners, demographers, and per mpy. Use the handy otier firmother researchers. The publication includes below to purchase your mpy today. i[il

——.———— ———— ———— —————— ———— .-— — ——. . ———— ———— ———— ——

Publication Order Fom Mail to: Superintendent of Documenl

Order processing code: *6247Government Printing Office

Washington, D.C. 20402

❑ -. please send me mpi= of Vital Statistics of the United States, 19S5, Lfie Tables

GPO Stwk Numbar-017-OZ-01027-5 Price $1.50

The total wst of my oi’der is $ Foreign orders please add an dditional 2WA

Prices imlude rwlar domestic postage and handling and are good thrcugh December, 198S. After that date, please call Order and Information Desk at

(202) 7=3238 to verify prices.

Please ~pe or PrintPlease choose method of paymenknu

(Companyor personalname)•1

(Additionaladdre~attention line) •1

(Streetaddress)

Check payableto the Superintendentof Documents

GPO DepositMount ~-n

VIM, MasterCardor Choicekcount

II

(Oily,State,ZIP Code) (Signature)

I )(Da~me phone including areacode) (Creditcard expirationdate)

Thank pu for your order! Thank you for your order! Thank you’ for your order! Thank you for your order! Thank you for your ord~

Our warehouses here at the GovernmentPrinting Office contain more than 16,000different Government publications. Nowwe’ve put together a catalog of nearly1,000 of the most popular books in ourinventory. Books like Infant Care,National Park Guide and Map, TheSwace Shuttle at Work. Federal Benefits. , .

Talents, and The Back-Yard Mechanic.Books on subjects ranging fromagriculture, business, children,and diet to science, space exploration,transportation, and vacations. Find outwhat the Government’s books are allabout. For your free copy of ournew bestseller catalog, write—

for Veterans and Dependents,Merchandising Your Job 937Bestsell

New CatalogPost Office Box 37000Washington, D.C. 20013

ers

Department of Health and Human Services

Otis R. Bowen, M. D., Secretary

Health Care Financing Administration

William L. Roper, M. D., Administrator

OffIce of Research and Demonstrations

Judith B. Willis, Director

OffIce of Research

J. Michael Fitzmaurice, Ph. D., Director

Division of Program Studies

Carl Josephson, Director

Survey Studies Branch

Herbert A. Silverman, Ph. D., Chief

Public Health Service

Robert E. Windom, M .D.,Assistant Secretary for Health

Centers for Disease ControlJames O. Mason, M. D., Dr. P. H., Director

National Center for Health Statistics

Manning Feinleib, M. D., Dr. P.H., Director

OffIce of Vital and Health Statistics Systems

Peter L. Hurley, Associate Director

Division of Health Interview Statistics

Owen T. Thomberry, Jr., Ph. D., Director

Utilization and Expenditure Statistics Branch

Robert A. Wright, Chief

U.S. DEPARTMENT OF HEALTH ANDHUMAN SERVICES

Public Health ServiceCenters for Disease Contro!National Center for Health Statistics3700 East-VVest HighwayHyattsville, Maryland 20782

OFFICIAL BUSINESSPENALW FOR PRIVATE USE, $300

,.

BULK RATEPOSTA::g&N;:E.S PAID

PERMIT NO. ~