estimating the labor market signaling value of the ged · estimating the labor market signaling...

ESTIMATING THE LABOR MARKET SIGNALINGVALUE OF THE GED

JOHN H TYLER

RICHARD J MURNANE

JOHN B WILLETT

This paper tests the labor market signaling hypothesis for the GeneralEducational Development (GED) equivalency credential Using a unique data setcontaining GED test scores and Social Security Administration (SSA) earningsdata we exploit variation in GED status generated by differential state GEDpassing standards to identify the signaling value of the GED net of human capitaleffects Our results indicate that the GED signal increases the earnings of youngwhite dropouts by 10 to 19 percent We nd no statistically signicant effects forminority dropouts

The positive correlation between education and earnings isone of the consistent ndings of the human capital literature Inthe early 1970s however Arrow 1973 and Spence 1973 formu-lated an alternative information-based explanation for the educa-tion-earnings relationship Over the last 25 years many econo-mists have conducted empirical work aimed at exploring thesignaling hypothesis It has proved difficult however to distin-guish between human capital and signaling explanations of theobserved relationship between education and earnings1

We are especially grateful to Jeffrey Kling Joshua Angrist Lawrence Katzand Caroline Minter Hoxby for helpful comments We owe thanks to JanetBaldwin of the GED Testing Service of the American Council on Education LindaHeadley Walker of the State Education Department of New York Leslie Averna ofthe Connecticut Department of Education and John Sojat of the Florida Depart-ment of Education for their assistance in providing data on GED testers and topersonnel of the Office of Research and Statistics at the Social Security Administra-tionrsquos Baltimore Headquarters particularly Peter Wheeler and Russell Hudsonfor providing and assisting with Social Security earnings data The support ofthese individuals and organizations does not imply their endorsement of thecontents of this paper

We gratefully acknowledge the support provided by the National Center forthe Study of Adult Learning and Literacy (NCSALL) and the Rockefeller RussellSage Spencer and Smith Richardson Foundations Work supported by NCSALLwas supported under the Educational Research and Development Centers Pro-gram Award Number R309B60002 as administered by the Office of EducationalResearch and ImprovementNational Institute on Postsecondary Education Librar-ies and Lifelong Learning United States Department of Education The contentsdo not necessarily represent the positions or policies of the National Institute onPostsecondary Education Libraries and Lifelong Learning the Office of Educa-tional Research and Improvement or the United States Department of Educationand you should not assume endorsement by the Federal Government

1 For recent attempts see Lang and Kropp 1986 Jaeger and 994and Park 1994

r 2000 by the President and Fellows of Harvard College and the Massachusetts Institute ofTechnologyThe Quarterly Journal of Economics May 2000

431

Ideal data for identifying the returns to a signal wouldcontain exogenous variation in signaling status among individu-als with similar levels of human capital In this paper weapproximate that ideal in estimating the returns to a particularlabor market signal the General Educational Development (GED)credential and in so doing we provide a test of the signalinghypothesis for this credential

The GED was introduced in 1942 to provide a way forveterans without a high school diploma to earn a secondary schoolcredential The GED program has evolved markedly so that todaythe credential is the primary lsquolsquosecond chancersquorsquo route to high schoolcertication for school dropouts in the United States Each yearmore than one million young people drop out of school andeventually about one-third of them will acquire a GED

Using interstate variation in GED passing standards we areable to compare individuals who all chose to try to obtain a GEDand who have the same GED test scores but who differ in GEDstatus because of differences in the stringency of the passingstandards in their state of residence We will argue that thislsquolsquonatural experimentrsquorsquo research design allows us to net out theeffect of human capital on earnings leaving clear estimates of thesignaling value of the GED credential in the labor market

Our estimates indicate that for young white dropouts who areon the margin of passing the GED exams the signaling value ofthe GED increases annual earnings by 10 and 19 percent We donot nd statistically discernible returns to the GED signal foryoung minority dropouts a puzzle we discuss later in the paper2

In the next two sections we discuss the GED credential andour identication strategy We then present our data methodsand results The paper closes with an extensive discussion of thepotential threats to our identication strategy an interpretationof our results and a summary

I THE GED IN THE LABOR MARKET AND IN THE LITERATURE

The ve tests in the GED battery cover mathematics writingsocial studies science and lsquolsquointerpreting literature and the artsrsquorsquo

2 Because of data limitations we are only able to identify individuals in ourdata as being members of one of two racialethnic groups (1) white non-Hispanicsand (2) everyone else The latter group includes African-Americans Hispanicsand other minority group members For expository clarity we refer to the twogroups as lsquolsquowhitesrsquorsquo and lsquolsquominoritiesrsquorsquo in this paper

QUARTERLY JOURNAL OF ECONOMICS432

The writing component of the test includes an essay and the restof the battery consists of multiple choice questions In the yearsrelevant for this paper the GED Testing Service (GEDTS) estab-lished a minimum passing standard that required a minimumscore of at least 40 or a mean score of at least 45 over the ve testsin the battery However the GED program is jointly supervised bythe GEDTS and the state education agencies and each state isfree to set standards higher than the GEDTS minimum Moststates do so creating the sources of variation critical to our study

In 1996 a record 758500 dropouts attempted the GED examsand half a million were awarded the certicate Table I illustratesthe trends behind that record number Based on October CPSdata the rst row of Table I shows that the high school completionrate for 18ndash24 year-olds trended upward slightly from 1975through 1992 and has been relatively steady through 1996Beginning in 1988 the October CPS allows determination of

TABLE IHIGH SCHOOL COMPLETION RATES OF 18ndash24 YEAR-OLDS FROM 1975ndash1996 AND BY

METHOD OF COMPLETION FROM 1988ndash1996

Completionmethod

Yearab

1975 1980 1985 1988 1989 1990 1991 1992 1993 1994 1995 1996

Total (percent)Completed 836 839 854 845 847 856 849 864 862 858 853 862Diploma 803 805 806 807 812 812 788 775 765Alternativec 42 42 49 42 52 49 70 77 98

White non-HispanicCompleted 872 875 882 887 890 896 894 907 901 907 898 915Diploma 844 851 848 852 857 855 842 826 810Alternative 42 39 48 42 50 47 64 72 105

Black non-HispanicCompleted 702 752 810 809 819 832 825 820 819 833 845 830Diploma 761 769 779 773 759 761 752 754 730Alternative 48 50 53 52 61 58 81 90 100

HispanicCompleted 622 571 666 582 594 591 565 621 644 618 628 619Diploma 544 548 548 534 566 582 542 540 552Alternative 38 47 42 31 55 61 76 88 67

Source Tables 13 and A25 in Dropout Rates in the United States 1997 U S Department of EducationOffice of Educational Research and Improvement

a The numbers for the years 1992ndash1997 reect new wording of the educational attainment items in theCPS

b The numbers for the years 1994ndash1997 reect changes in the CPS due to newly instituted computer-assisted interviewing and the change in the population controls to the 1990 Census-based estimates withadjustment for the undercount in the 1990 Census

c Completed high school by means of an equivalency test such as a GED exam

LABOR MARKET SIGNALING VALUE OF THE GED 433

whether high school lsquolsquocompletionrsquorsquo culminated in a regular diplomaor an alternative certication and the GED is by far the mostcommon alternative form of certication Cameron and Heckman1993 The second and third rows of Table I show that the steadyschool completion rate over the 1990s masks opposite movingtrends completion via a regular high school diploma trendeddownward while completion via GED certication trended up-ward The last three panels of the table show that these trendswere true for all racialethnic groups

In general GED holders are a relatively advantaged groupcompared with uncredentialed dropouts On average they com-plete more years of schooling before dropping out they havehigher levels of measured cognitive skills and their parents havemore education Even after controlling for these relative advan-tages however there are reasons to believe that GED holdersmight fare better in the labor market than dropouts without thecredential

First for dropouts with low levels of basic cognitive skills orwith English language deciencies raising skills to levels re-quired to pass the GED battery may require considerable workresulting in increased human capital While Cameron and Heck-man 1993 report a median study time for the GED of only abouttwenty hours the wording in the survey question on which theirestimates are based could lead to an underestimation of the totalGED preparation time For example it is not clear whetherdropouts who required participation in English as a SecondLanguage (ESL) or Adult Basic Education (ABE) courses prior toGED preparation would count the time spent in those classes asGED preparation on the survey question Also there is a longright-hand tail to this distribution so that the least skilledGED-holders may spend considerable time preparing to pass thetests Thus the amount of human capital associated with acquir-ing a GED especially for low skilled dropouts is still unclear3

Returns to a GED can also be explained by a signalinghypothesis Assuming a distribution of underlying productivity inthe pool of dropouts the more productive dropouts would like tosignal their higher productivity to employers and receive a higher

3 In addition to human capital returns related to test preparation there maybe an additional human capital payoff from increased access to postsecondaryeducation and training associated with GED acquisition In a later section we willshow that any human capital effects resulting from this route are likely to be verysmall for the dropouts who drive our results


wage offer The GED credential may serve as this signal in thelabor market If the costs of acquiring the GED and the wageoffers based on GED status are such that the expected net benetsof GED acquisition are negative for the less productive dropoutsand positive for the more productive dropouts then we couldobserve a separating equilibrium where GED acquisition ispositively correlated with productivity and wages4

Since the returns to a GED can be explained by either humancapital or signaling theory the empirical task is to estimate thesignaling value of the credential net of any human capital effectsassociated with GED acquisition A rst step is establishing thecorrelations between GED acquisition and labor market outcomes

Table II presents estimates from log earnings regressionsusing High School and Beyond data on individuals who weresophomores in 1980 The dependent variable is the log of average1990ndash1991 earnings years when most of the sample would havebeen 25ndash26 years of age

Across all of the models regular high school graduatesconsistently show large advantages over dropouts regardless ofraceethnicity In the models in columns 1 and 2 that do not controlfor sophomore test scores GED-holders have earnings that areabout 16 percent higher than uncredentialed dropoutsmdashwellestimated for white dropouts and less well estimated for minoritydropouts Adding test score controls in the third and fourthcolumns however reduces the estimated GED effect for bothwhites and minorities so that neither estimate is statisticallysignicant

Since the positive but statistically insignicant estimates ofthe lsquolsquoGED effectrsquorsquo pick up both human capital and labor marketsignaling effect on earnings one interpretation of these estimatesis that any signaling effect of the GED is very small However asecond interpretation we favor is that these baseline equationsare misspecied because they do not allow the effect of the GED onearnings to vary according to the skill level of the dropout This isimportant in interpreting our main results since our estimatesare based only on very low skilled GED-holders We return to thissubject later

4 The actual pecuniary costs associated with GED acquisition are generallylow ranging from zero dollars in some states to around $50 in other states Thusthe binding costs for low productivity dropouts are more likely associated withopportunity costs tied to long preparation time in upgrading skills or in the psychiccosts associated with the school-like preparation for and taking of the seven andthree-quarter hour battery of tests


In a widely cited study Cameron and Heckman 1993 useddata from the National Longitudinal Survey of Youth (NLSY) toshow that male conventional high school graduates consistentlyoutperform male GED-holders in the labor market Their workhowever is less denitive on the question of how GED-holdersfare relative to uncredentialed dropouts They nd that maleGED-holders have wages that are between 3 and 6 percent higherthan uncredentialed male dropouts but the results are notstatistically signicant They conclude that lsquolsquo GED recipients liebetween uncredentialed dropouts and regular high schoolgraduates in their economic standing but are much closer todropoutsrsquorsquo Cameron and Heckman 1993

TABLE IIEARNINGS REGRESSIONS FOR INDIVIDUALS IN THE HIGH SCHOOL AND BEYOND

SURVEY (DEPENDENT VARIABLE IS THE LOG OF THE AVERAGE OF 1990ndash1991ANNUAL EARNINGS AND STANDARD ERRORS ARE IN PARENTHESES)

Whitesa Minoritiesb Whites Minorities

Intercept 9518 9320 9455 9199(0057) (0079) (0059) (0081)

Female 2 0397 2 0306 2 040 2 0298(0019) (0030) (0022) (0032)

GED 0162 0164 0094 0083(0072) (0109) (0072) (0109)

High school graduate 0536 0581 0380 0400(0057) (0073) (0059) (0075)

Math test score 0012 0008(0002) (0002)

Reading test score 00002 0002(0003) (0005)

Writing test score 0002 0008(0003) (0005)

Science test score 2 0008 0003(0004) (0005)

Vocabulary test score 0007 0009(0003) (0004)

Region dummiesc Yes Yes Yes YesR2 011 0087 0133 0131Nd 5403 2810 5403 2810

a White non-Hispanicb Everyone not white non-Hispanicc The regions are South West North Central and Northeastd Included in the sample were all dropouts (those with and without a GED as of the 1992 survey) and

individuals with at least a high school degree who were interviewed in the 1992 HSB survey who were not inthe military in 1992 and not in college in both 1990 and 1991 who did not have zero earnings in both 1990 and1991 and who had nonmissing values for their 1980 sophomore test scores


While informative their comparisons between uncreden-tialed dropouts and GED-holders are potentially problematic fortwo reasons First data limitations required that all uncreden-tialed dropouts in the datamdashboth those who chose not to acquire aGED and those who attempted to acquire a GED but failed theexamsmdashform the comparison group in their study It is question-able whether either of these dropout groups provides a satisfac-tory counterfactual

Constraints imposed by model specication however arepotentially the most important and interesting reason to reexam-ine the returns to the GED Similar to the specications used inTable II all of the Cameron and Heckman models constrain anyGED effect on wages or employment to be the same for alldropouts regardless of the skills they possess Models that do notallow for a differential effect of the GED by skill may missimportant elements of the way the GED works in the labormarket This may be an especially important consideration regard-ing a signaling hypothesis For example it may be that the GED isa more important signal for dropouts who leave school with lowskills than for those who leave school with higher skill levels

As a result of our identication strategy our estimatesidentify the signaling effect of the GED only for those dropouts onthe margin of passing the GED examsmdashthat is for the very leastskilled GED holders If the returns to a GED are larger for this lowskilled population than for more highly skilled dropouts then ourresults are potentially consistent with those of Cameron andHeckman

II ESTIMATING THE SIGNALING VALUE OF THE GED USING

DIFFERENTIAL STATE GED PASSING STANDARDS

In this paper we take advantage of interstate variation inGED passing standards to generate variation in GED statusamong individuals with the same observed levels of humancapital Our identication strategy is simple Given two sets ofstates one with a lower GED passing standard than the other wecompare individuals with marginally passing GED test scores inthe lower passing standard states with individuals who have thesame GED test scores in the higher standard states Howeverbecause of the more stringent GED passing standard our compari-son individuals in the higher passing standard states will nothave the credential Once we have accounted for differences in


state labor markets we can estimate the impact of the GED onearnings by comparing the mean earnings of the two groups Sincethese individuals have the same GED test scores any humancapital effect on earnings captured by the GED test scores isremoved from the estimates There may be other forms ofunobservable human capital differences between the two groupsthat would bias our estimates We address the most obvious andproblematic of these in a later section in the paper and concludethat they do not pose serious problems for the main estimates inthis paper Thus we interpret the estimates from this methodol-ogy as the signaling effects of the GED on the earnings of dropoutswho would choose to obtain the GED and are at the margin ofpassing5

Our data contain information on 1990 GED candidates from45 states All of the states in our sample dene their GED passingstandards by a combination of the minimum test score and themean score over the ve tests In our research we are able to usethree of the possible seven different passing standards thatexisted across the nation in 1990 They are in ascending order ofdifficulty

c a minimum score of at least 40 or a mean score of at least45

c a minimum score of at least 35 and a mean score of at least45 and

c a minimum score of at least 40 and a mean score of at least45

The nature of our data in combination with the standards abovedictate that we use minimum-mean score combinations to con-struct different GED score groups The score groups were con-structed so that based on score group and state in which the GEDwas attempted we can be assured of GED status Thus weconstructed ten GED lsquolsquoscore groupsrsquorsquo based on the intersection ofthe minimum and mean score ranges represented in the rows andcolumns of Table III

In Table III individuals in score groups 1 and 2 would haveGED scores below the passing standard in any state in 1990Individuals in score group 5 or higher would have scores above thepassing standard of any state in our data in 1990 It is only inscore groups 3 and 4mdashthe lsquolsquoaffected score groupsrsquorsquomdashthat we nd

5 Thus our results are not estimates of the effect of the GED on the earningsof a random dropout


variation in GED status by the state where the GED exams weretaken in 1990 In each of these two scoring ranges individuals instates with the lower passing standardmdashthe lsquolsquotreatment statesrsquorsquomdashhave a GED while individuals in lsquolsquocomparison statesrsquorsquo do not dueto higher passing standards

Thus we have these three natural experiments to test thesignaling hypothesis as it pertains to the GED We denote theseexperiments by the affected score group in each experiment

mdashExperiment 4 where variation in GED status by state is inscore group 4 the treatment states are those states that award aGED in score groups 4 and higher and the comparison states arethose that award a GED in score groups 5 and higher


mdashExperiment 3 where variation in GED status by state isin score group 3 the treatment states are those states that awarda GED in score groups 3 and higher and the comparison states arethose that award a GED in score groups 4 and higher

Since the GED passing standards are set by the individualstate education departments and since these agencies are embed-ded in the state political system the interstate variation in pass-ing standards on which we rely is not transparently exogogenousThe fact that GED passing standards in general and the ones weexploit in particular have been relatively stable over time

TABLE IIIGED SCORE GROUPS FORMED BY COMBINING MINIMUM AND MEAN SCORES

(OUTLINED CELLS 5 VARIATION IN GED-STATUS BY STATE DARK SHADING 5 ALL

POSSESS GED NO SHADING 5 NONE HAVE GED)

Minimum score

Mean score

45 5 45

20ndash34 Score group 135ndash39 Score group 2 Score group 440ndash44 Score group 3 Score group 545ndash46 Score group 647ndash48 Score group 749ndash50 Score group 851ndash52 Score group 9531 Score group 10


however suggests that states do not set the GED passing stan-dard based on current state economic conditions6 We assume thatthe variation in passing standards we use is exogenous condi-tional upon state xed effects

III RESEARCH DESIGN

A Data

To exploit the sources of GED signaling status describedabove we needed information on demographic characteristicsGED test-scores state in which tested and earnings for a sampleof dropouts The rst three types of information came fromadministrative records of the GED Testing Service and the stateeducation agencies of New York Florida and Connecticut7 In thedata from New York Florida and Connecticut we have informa-tion on the universe of dropouts aged 16ndash21 who last attemptedthe GED tests in 1990 For data obtained from the GED TestingService we have from each of 42 states a sample of dropouts whowere aged 16ndash21 when they attempted the GED battery in 1990

Using Social Security numbers supplied by GED candidatesat the time of testing Social Security Administration (SSA)programmers attached Social Security-taxable earnings data toour GED test-le data providing valid matches for about nine outof ten dropouts8 The resulting micro-level data set containeddemographic and GED test score information collected when thedropouts in the sample attempted the GED in 1990 together withtheir Social Security earnings records for each year from 1988 to1995 The SSA does not release individual earnings to research-ers As a result the SSA used the micro-level data to construct anaggregated data set satisfying SSA condentiality requirementsand our research design9 The nal aggregated data set consists of990 cells dened by the state where the GED was attempted

6 For example out of the 42 states used in our study only Louisiana in 1982California in 1984 New York in 1985 Arkansas in 1987 and Washington in 1988changed their GED passing standard between 1980 and 1990

7 We augment the GED Testing Service data with data from New YorkFlorida and Connecticut because these states were missing from the GED TestingService data

8 Providing SSNs on the GED test form is voluntary but inspection of theFlorida and Connecticut samples indicates that about 98 percent of examinees 21and younger (our target group) provide this information

9 SSA condentiality regulations require that no information be released incells where there are fewer than three observations Other SSA limitations arediscussed in the Data Appendix


GED score group gender and whether the GED candidate waswhite or minority10 Each cell contained average FICA earningsfor a particular year for the individuals in the cell cell frequencyand the standard deviation of earnings

Table IV presents summary statistics for the aggregateddata In the table we possess the universe of young dropouts whoattempted the GED in 1990 in New York Florida and Connecti-cut and so relatively speaking the number of individuals contrib-uted by those states is large (column 1) The analytic sample isabout 70 percent white (column 6) and 55 percent male (column7) Columns 8 and 9 present mean earnings ve years afterdropouts attempted the GED battery and these earnings aregenerally low11 While not reported here the variances of 1995earnings for any particular subset of our sample are generallylarge relative to the mean as is typical with annual earningsColumn 2 shows the minimum score group in which individualswould have a GED in a given state reemphasizing the threedifferent passing standards present in our data Columns 3 4 and5 of Table IV designate whether a state falls into the treatment-group or comparison-group in each experiment that we usedemonstrating how some states contribute to the treatment groupin one experiment and to the comparison group in another

We use New York and Florida to form the comparison group intwo of our three experiments We single out these states becauseNew York and Florida data contain the accumulated best scores ofdropouts who have either attained a GED or who have lsquolsquostoppedoutrsquorsquo as of 199012 The 1990 GEDTS data however contain only aone-year lsquolsquosnapshotrsquorsquo of GED-attempters and some unknownportion of individuals with failing scores in the 1990 GEDTS datawill have subsequently retaken and passed the battery in years1991ndash199513 Therefore any comparison group constructed from

10 To avoid small cells in the affected score groups that would be censored bythe SSA we could not stratify more nely on raceethnicity and on age

11 Earnings in New York are low compared with those in other states Inadditional data we nd that the positive earnings of dropouts in New York are thehighest of any state indicating that in our data many young dropouts in NewYork had zero Social Security taxable earnings in 1995

12 By this we mean that we are certain that nonpassers in the New York andFlorida data did not return after 1990 to retake the exams and potentially receive aGED in any of the years prior to our measurement of earnings In these statesanyone who took any portion of the tests after 1990 would not be included in thedata regardless of their passing status

13 For example using the Connecticut and Florida individual-level datawe nd that about 70 percent passed the GED on the rst try in 1988 Out of thosewho did not pass about 40 to 50 percent retested and acquired a GED within two


nonpassing GED candidates in the GEDTS data will be contami-nated with individuals who subsequently obtained a GED Thisoccurs in Experiment 3 and the results from this experiment

years The result is that the overall pass rate went from 70 percent on the rst tryin 1988 to an eventual pass rate of about 85 percent by the end of 1990

TABLE IVSUMMARY STATISTICS AND TREATMENTCOMPARISON-GROUP DESIGNATION BY STATE

FOR YOUNG DROPOUTS WHO ATTEMPTED THE GED IN 1990

Stategroupa

Col 1N

Col 2Minimum

scoregroup

for GEDb

Cols 3 4 amp 5Treatment or

comparison in

Cols 6 amp 7 of samplewho are

Cols 8 amp 9Mean 1995earnings

Exp3 Exp3 Exp4 White Male No GED GED

AZ 3457 4 mdash C T 65 57 7176 9336CT 1692 4 mdash mdash mdash 71 51 8779 10413FL 17905 5 C mdash C 74 56 7767 9694KY 1949 4 mdash C T 83 58 6797 8149NY 19134 5 C mdash C 53 56 6777 8764TN 1668 4 mdash C T 94 50 8479 9079TX 5027 3 T T mdash 60 51 7755 9286VA 1657 4 mdash C T 78 59 7905 9662Miscc 2852 4 mdash C T 79 54 7559 8813Atlantic states 3237 4 mdash C T 75 55 8184 10327New England 1801 4 mdash C T 91 52 8273 9727North Central 2313 4 mdash C T 85 55 9031 10364Low pass

statesd 4138 3 T T mdash 76 55 7899 8732Miscellaneous

statese 4202 4 mdash C T 80 55 7750 9032WA and CA 4713 5 mdash C mdash 64 57 8017 9351High pass

statesf 4255 5 mdash C mdash 67 55 6258 8669Totalg 80000 mdash mdash 69 55 mdash mdash

a States AZ through VA had enough observations to serve as their own lsquolsquostate grouprsquorsquo Other states had tobe grouped into either geographic or other logical groupings due to small numbers of observations See theData Appendix for a listing of the states in each group not otherwise detailed here

b There were seven different GED passing standards in 1990 that ranged from the low standardrepresented by score group 3 to that of Wisconsin which required a mean of 50 and a minimum of 40

c Alaska Hawaii Idaho Montana Alabama Iowa Kansas and Nevadad This group is composed of Mississippi Louisiana and Nebraska and together with Texas they

comprise the group of states with the lowest passing standard a mean of 45 or a minimum of 40e This group is composed of Colorado Illinois Pennsylvania and Minnesota the only states where there

were statistically signicant differences between individuals with and without imputed writing scores (Seeabove and the Data Appendix for a discussion of imputed writing scores in the GEDTS data)

f This group is composed of Arkansas Delaware Maryland Oklahoma South Dakota and Utah andalong with Washington and California these states have the same higher standards as do New York andFlorida

g Note that with about 36000 observations combined New York and Florida heavily inuence thestatistics in this row


provide a conservative lower bound estimate of the signalingeffect of the GED

B Statistical Analyses

The key treatment-control comparison is captured in thefollowing model describing the annual earnings of dropouts in anunobserved individual-level world

(1) Yi 5 b 0 1 b 81STi 1 b 82SGi 1 a (Ti p ASGi) 1 b 3Femalei 1 ui

where

Y 5 individual irsquos annual earnings in some year after theGED was attempted

ST 5 vector of state dummiesSG 5 vector of dummies representing the GED score group

containing individual iT 5 1 if individual i is in a lower standard treatment state 0

otherwiseASG 5 1 if individual i is in the affected score group and 0 if

individual i is in a higher score group where the lsquolsquoaf-fected score grouprsquorsquo is either score group 3 or score group4 depending on the experiment14

Female 5 1 if individual i is a female 0 if a maleu 5 error term with zero expectation and orthogonal to

predictors

For reasons we explain later we stratify our data into whiteand minority dropouts and we would like to t equation (1)separately for each of these two raceethnicity groups Howeverwe cannot t the hypothesized model in equation (1) because wedo not have individual-level data on earnings Fortunately basedon equation (1) the parameter of interest a representing theeffect of the GED on the earnings of dropouts whose GED-status isaffected by state-of-residence is

(2) a 5 EYi Ti 5 1ASGi 5 1 2 EYi Ti 5 0ASGi 5 1

2 (EYi Ti 5 1ASGi 5 0 2 EYi Ti 5 0ASGi 5 0)

14 We omit those who score below the affected score group range because inthe GEDTS data these score groups are contaminated with individuals who havea GED and whom we cannot identify This contamination downwardly biasesestimates of a Estimates using these lower score groups are available from theauthors upon request These lsquolsquocontaminatedrsquorsquo estimates are consistently lower thanour preferred estimates regardless of the experiment and the racial group


suggesting that a method-of-moments estimator of a can beconstructed from sample means This difference-in-differencesestimator is

a ˆ 5 (YT 2 YC) 2 (YTHi 2 YCHi)

where

Y 5 mean earnings of any group

T indexes the treatment groupmdashindividuals in the affected scoregroup who are in a low-passing-standard state in a givenexperiment and thus have a GED

C indexes the comparison groupmdashindividuals in the affected scoregroup who are in a high-passing-standard state and thus donot have a GED

THi indexes individuals in treatment states that are in a scoringgroup higher than the affected score group and thus have aGED

CHi indexes individuals in comparison states that are in the samehigh scoring group as the THi group and thus also have aGED

We use all individuals in score groups 5 through 10 to formthe high comparison groups THI and CHi in the difference-in-differences estimation This avoids arbitrarily deciding whichscore group to use in this capacity and maximizes sample size inthe high comparison groups15 Finally notice that the seconddifference in the difference-in-differences estimator YTHi-YCHiremoves state xed effects from the estimate

We can estimate a and conduct hypothesis tests usingSSA-provided aggregate within-group statistics on earnings16 Weidentify a using two assumptions The rst is that the differentialstate GED passing standards generate exogenous variation inGED status This is embodied in the assumed independence of theinteraction and error terms The second assumption is thatequation (1) contains no higher order interactions Later weexamine empirical evidence regarding violations of these assump-

15 Additional analyses verify that our ndings are not qualitatively differentwhen we use any one of the score groups as the high comparison group as opposedto using everyone simultaneously

16 The standard errors for the estimator are obtained in the standardmanner with no assumptions of equality of variances across groups


tions conrming that any resulting biases have little impact onour estimates

In computing our difference-in-differences estimates weweight the calculation of within-group earnings to account for thedifferent gender distributions within treatment and comparisongroups and across states within these groups In other analysesnot reported here we conrm that our ndings are not sensitive tothe choice of weights

IV RESULTS

A Difference-in-Differences Estimates of the Impact of the GEDon the Earnings of Young Dropouts

Table V presents separate estimates of the effect of the GEDon 1995 earnings for white and minority dropouts from threedifferent experiments involving young dropouts who attemptedthe GED in 199017 We focus initially on estimates generated byexperiments 3 and 4 In these experiments acquisition of a GEDis associated with about a $1500 increase in annual earnings foryoung white dropouts an increase of approximately 19 percentThe point estimates for whites from the two different experiments($1473 and $1531) are remarkably similar given that the statesmaking up the treatment group in each experiment are different

Both experiments 3 and 4 use New York and Florida ascomparison-group states Experiment 3 like experiment 3 usesTexas Louisiana Mississippi and Nebraska as treatment statesHowever experiment 3 excludes New York and Florida andinstead uses states from the GEDTS data in the comparisongroups including states where the GED is awarded in score group4 As explained above when GEDTS data are used to constructthe comparison group the group will be contaminated by anunknown number of individuals who obtained a GED after 1990Because of this the 1995 earnings of the comparison group will beinated (assuming a positive treatment effect of the GED onearnings) Thus the $907 difference-in-differences estimate inexperiment 3 represents a downwardly biased conservativeestimate of the impact of the GED on earnings

17 Preliminary analyses indicated that results for white males and femaleswere similar and the results for minority males and females were similar Thus toincrease statistical power we aggregated across the genders retaining thewhite-minority distinction


The results for nonwhite dropouts differ sharply from theresults for white dropouts The three experiments yield nostatistically signicant evidence that acquisition of a GED resultsin higher earnings for minority dropouts We return to theminority results later Based on the results from experiments 4 3and 3 our estimates are robust to the use of different treatmentand comparison groups

B Timing of GED Treatment Effects

To avoid underestimating the impact of the GED by measur-ing earnings too close to receipt of the credential we have

TABLE VDIFFERENCE-IN-DIFFERENCES ESTIMATES OF THE IMPACT OF THE GED ON 1995

EARNINGS OF DROPOUTS WHO TESTED IN 1990 (STANDARD ERRORS ARE

IN PARENTHESES)

Experiment 4 Experiment 3 Experiment 3

State passingstandard is Low-High

standardcontrast


standardcontrast


standardcontrastLow High Low High Low High

Panel A WhitesTest score is

Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)

Difference-in-differencesfor whites

1473(678)

1531(529)

907(481)

Panel B MinoritiesTest score is

Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)

Difference-in-differencesfor minorities

2 1357(906)

231(548)

2 67(518)

5 signicant at the 001 level 5 signicant at the 005 level 5 signicant at the 010 levelExperiment 4 Test Score Low score group 5 4 Test Score High score groups 5 5ndash10Passing Standard Low 35 minimum score and 45 mean score Passing Standard High 40 minimum score

and 45 mean scoreLow Passing Standard states All states except for TX LA MS NE FL NY CA WA and CT High

Passing Standard states NY and FLExperiment 3 Test Score Low score group 5 3 Test Score High score groups 5 5ndash10Passing Standard Low 40 minimum score or 45 mean score Passing Standard High 40 minimum score

and 45 mean scoreLow Passing Standard states TX LA MS and NE High Passing Standard states NY and FLExperiment 3 Test Score Low score group 5 3 Test Score High score groups 5 5ndash10Passing Standard Low 40 minimum score or 45 mean score Passing Standard High 35 minimum score

and 45 mean scoreLow Passing Standard states TX LA MS and NE High Passing Standard states all states except TX

LA MS NE NY FL and CT


concentrated on using the 1995 earnings of young dropouts whoattempted the GED in 1990 as our outcome measure To exploreboth pretreatment earnings differences as well as the possibilitythat benets accruing to GED-acquisition do indeed take time todevelop we construct equivalent difference-in-differences esti-mates using earnings in the years 1988ndash1989 and 1991ndash1995These comparisons continue to use 1990 test-takers as theanalytic sample and we concentrate on white dropouts

Figures IndashIII show year-by-year difference-in-differences esti-mates for experiments 4 3 and 3 Posttreatment earnings inFigures I and II (years 1ndash5 after the GED attempt) suggest that ittakes time for the GED to pay off for young white dropouts FigureI shows that in the rst two years after GED-acquisitionGED-holders actually earn less than uncredentialed dropoutswith the same GED scores Over time GED-holders in thetreatment group gain on their uncredentialed counterparts in thecomparison group so that by the fth year after GED-acquisitionthey are earning $1473 more per year And while it appears fromFigure II that treatment individuals in experiment 3 can makethe GED pay off sooner than those in experiment 4 the trends arethe same Estimates from experiment 3 are not well suited forstudying the timing of GED effects because each year after 1990the comparison group becomes increasingly contaminated byretesters who return to obtain a GED We consider pretreatmentearnings in a later section

V THREATS TO IDENTIFICATION IN THE NATURAL

EXPERIMENT ESTIMATES

There are four general sources of potential bias in ourprimary results in Table V First to the extent that individualsuse the GED to access postsecondary education or training ourfailure to control for these human capital effects would lead tooverestimates of the signaling value of the GED Also there arethree different types of endogeneity bias that could affect ourestimates We explore these potential biases in this section

A Postsecondary Education and Training Effects

Most academic postsecondary programs and many trainingprograms as well as the loan and grant programs that can helppay for them require a school leaving certicate For mostdropouts who desire admittance to these programs the GEDserves this purpose It follows that acquisition of a GED could lead


FIGURES IndashIIIPretreatment and rst through fth year Difference-in-Differences Estimates forYoung White Dropouts ( 5 Signicant at the 001 Level 5 Signicant at the

005 Level 5 Signicant at the 010 Level)


to higher average levels of human capital through increasedaccess to postsecondary education and training programs Sincewe do not account for this potential human capital effect interpre-tation of our estimates as returns to a signal could overstate thecase Analysis of other data indicates however that these types ofhuman capital effects do not seriously threaten the signalinginterpretation of our estimates

In HSB data we nd that the least skilled GED-holders (thosein the lowest quartile of the sophomore math test) accumulated onaverage less than ve postsecondary credits by age 28 Based onan academic year being equal to 30 credits this is one-sixth of anacademic year Other authors Kane and Rouse 1995 MurnaneWillett and Boudett 1999 estimate that a year of postsecondaryacademic work has a return of about 5 percent relative to highschool graduation This suggests that less than one percentagepoint (16 3 005) of any estimate based on low skilled GED-holders might be due to postsecondary education human capitalaccumulation rather than labor market signaling

Regarding postsecondary training Murnane Willett andBoudett 1997 use the NLSY to show that both male and femaleGED-holders are more likely to obtain off-the-job training (ietraining in a government-sponsored program or a proprietaryschool) than are comparable uncredentialed dropouts18 Howeverin later work Murnane Willett and Boudett 1999 they also ndno effect of this off-the-job training on the wages of male GED-holders and Boudett 1998 nds no effect of off-the-job trainingon the wages of female GED-holders Taken together theseanalyses suggest that neither postsecondary education nor off-the-job training carries substantial earnings effects for low skilleddropouts who are on the margin of passing the GED exams

B Endogeneity Bias Associated with Individual Behavior

Interpreting the results in Table V as the causal impact of theGED signal on earnings rests on an assumption that our treat-ment and comparison groups are balanced on unobservablecharacteristics that might inuence their earnings If the differ-ent passing standards inuence individual behavior in systematicways then this assumption may be violated There are threedecisions that might be inuenced by differential GED passing

18 Since the GED signal may help dropouts obtain jobs that offer trainingany returns to post-GED on-the-job training are returns to the signaling value ofthe GED


standards across states the decision to attempt the test thedecision to migrate to another state and the decision about howmuch effort to exert on the test Each of these potential decisionshas different implications for our identication strategy

First when faced with higher standards some individuals onthe margin who would otherwise test may elect not to attempt theGED battery Assuming that this behavior is negatively correlatedwith productivity-enhancing traits such as persistence self-condence and motivation then this type of selection wouldresult in an overestimate of the mean earnings of potentialGED-holders in comparison group states The net effect would bea downward bias in the estimated effect of the GED on earnings

Endogenous migration decisions are a problem if more moti-vated dropouts tend to migrate to lower standard states in orderto increase their chances of passing the exams or because of otherfactors that are also related to earnings We examine this possibil-ity by examining the migration patterns of dropouts in the NLSY

Rows 1 2 and 3 in Panel A of Table VI represent the mobilitypatterns of GED-holders who were in low medium and high

TABLE VISTATE MIGRATION PATTERNS BETWEEN LOW MEDIUM AND HIGH GED PASSING

STANDARD STATES FOR GED-HOLDERS AND UNCREDENTIALED DROPOUTS IN THE

NLSY (WEIGHTED TABULATIONS ROW PERCENTAGES SUM TO 100)

Panel AGED-holdersa

Passing standard in stateof residence at age 14

GED passing standard in state ofresidence in year of GED acquisition

Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

Panel BUncredentialed dropoutsPassing standard in state

of residence at age 14

GED passing standard instate of residence at age 21

Low Medium HighLow 867 66 68Medium 25 919 56High 15 90 895

a To match the age group used in this paper the percentages in Panel A are based on the 512GED-holders in the NLSY who were aged 16ndash21 when they obtained their GED and who obtained thecredential in 1979 or later The migration patterns of all GED-holders in the NLSY are very similar to thisyounger age group We excluded those dropouts who obtained their GED prior to 1979 because we cannotdetermine the state of residence at GED acquisition for earlier years Finally there are ve states (SC NMND WI and NJ) with standards that do not t our low medium and high categories


standard states (respectively) at age fourteen The percentagesacross the rows sum to 100 so that 89 percent of the individualswho started out in low standard states were still in a low standardstate when they obtained their GED 85 percent had moved to amedium standard state and 26 had moved to a high standardstate by the time of GED acquisition

Systematic migration from high to low standard states wouldbe indicated by relatively little migration out of the upperleft-hand cell and relatively more migration to the left out of boththe middle cell of the table and the lower right-hand cell of thetable Instead we see that about 10 percent of GED-holdersregardless of the passing standards in their state of residence atage fourteen obtain their GED in a state with a higher or lowerstandard Furthermore among those GED-holders who do movebetween age fourteen and GED acquisition there is no evidence ofsystematic movement to lower standard states

As a further check on the mobility patterns of GED-holdersPanel B of Table VI presents the migration probabilities (as of age21) of the uncredentialed dropouts in the NLSY The results aresimilar to those of Panel A again suggesting that there are nosystematic links between GED state passing standards and themigration decisions of young dropouts19

The third source of bias that could result from individualresponses to state passing standards relates to test-taking effortThe concern is that individuals in our treatment group may havehigher unobserved skills than individuals in the comparisongroup This situation could occur if some of the more highly skilleddropouts in the lower standard states exerted just enough effort tonarrowly pass the exams In this case our treatment group wouldbe contaminated with individuals whose skills were higher thantheir GED scores indicated We can explore this potential sourceof bias with our data

To do this we assume that some individuals in lower stan-dard states whose true skills are higher than what is required tolsquolsquojust passrsquorsquo exert just enough effort to pass the GED battery As aresult some of these higher ability individuals are in score groups3 or 4 in our data Figure IV displays the sorting that would result

19 A separate problem would arise if GED candidates did not migrate butsimply crossed state lines to attempt the test in a lower standard state We notethat all states have residency requirements to attempt the GED Furthermore thechief state GED officer in Texas informed us that proof of residency was oftenrequested in testing sites situated close to the borders with other states Erwin1999


from this assumption The top row of Figure IV depicts the tenobserved GED score groups The next two rows depict hypotheti-cal distributions of true underlying skills across the ten GEDscore groups used in this study when some higher ability individu-als exert just enough effort to pass The middle row illustrates ahypothetical distribution of true skills for low standard states(where the GED cutoff is at score group 3) and the bottom rowillustrates this situation for medium standard states (where GEDcutoff is at score group 4) As an example according to the middlerow of Figure IV observed score group 3 in low standard states notonly has individuals whose true underlying skills place them inthis score group (lsquolsquotrue 3srsquorsquo) but also individuals with higher skills(lsquolsquotrue 5s and 6srsquorsquo) who exerted just enough effort to pass This typeof lsquolsquoability contaminationrsquorsquo of score group 3 in the low standardstates would cause us to overestimate the effect of the GED onearnings

We can adjust our estimates to account for this type ofdifferential effort by assuming the following

(1) the observed proportion of individuals in the affectedscore group in the higher standard states represents whatthe true skill distribution in the affected score group inthe lower standard states would be in the absence ofdifferential effort across the states20 (eg for experiment

20 As an anonymous reviewer pointed out this assumption would likely nothold if there were differential participation rates across lower and higher standardstates However we assume that fewer individuals attempt in the higher standardstates If more were to attempt then presumably the percentage who lsquolsquojust failrsquorsquo inthese higher standard states would go up and based on the algorithm we describe

FIGURE IVScore-Group Sorting around Low and Medium GED Passing Standards (Shaded

Cells 5 Score Groups Where the GED is Awarded)


3 the correct proportion of true 3s in the low standardstates is given by the proportion of observed 3s in thehigher standard states) and

(2) the mean earnings of higher skilled individuals who scoreless than their true abilities are the same as the observedmean earnings of similarly skilled individuals who sortthemselves into their true score groups

To illustrate how we use these assumptions to generateadjusted estimates consider experiment 3 using just Texas a lowstandard state and Arizona a medium standard state Individu-als in Texas in score group 3 have a GED while those in Arizona inscore group 3 just missed getting a GED Say that we see 4percent of the distribution of GED attempters in Texas in scoregroup 3 but only 1 percent of the distribution of attempters inArizona in that score group We are worried that the largerpercentage in Texas is the result of higher ability individualsexerting just enough effort to pass in that state placing them inscore group 3 As a result the observed earnings in score group 3in Texas are a weighted average of score group 4 types and higherability types and are biased upward from what they would be ifeveryone scored according to their underlying ability

To lsquolsquopurgersquorsquo the estimated earnings in score group 3 of theearnings of the higher ability types we will assume based onwhat we observe in Arizona that out of the 4 percent in scoregroup 3 in Texas only one-fourth are true score group 3 typeswhile the remaining three-fourths are actually say true lsquolsquo5srsquorsquo Ifwe assign to the three-fourths who we assumed are 5s theobserved mean earnings for score group 5 in Texas we can thenestimate the earnings of the one-fourth whom we believe to betrue 3s These are our adjusted earnings estimates

Adjusted estimates based on these assumptions and algo-rithm are in Table VII We assume that the treatment group ineach of the experiments is primarily contaminated by true 5s ortrue 6s and so we employ three different assumptions in Table VIIregarding the mix of 5s and 6s First we assume no contamina-tion which simply redisplays our original results from Table V forease of comparison Next we assume a 50ndash50 mix of 5s and 6s andthen to get a very conservative lower-bound estimate we assumeonly 6s contaminate the treatment group

our adjusted estimates would be closer to our unadjusted estimates Thus ifanything we believe that differential participation gives us conservative adjustedestimates


TABLE VIIADJUSTED ESTIMATES IN THE PRESENCE OF DIFFERENTIAL TESTING EFFORT ACROSS

LOW AND HIGH STANDARD STATES FOR YOUNG WHITE DROPOUTS (STANDARD

ERRORS ARE IN PARENTHESES)

Adjusted numbersof observations

Adjusted estimates underdifferent skill composition

assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e

Treatment groupcontamination

assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451

no contaminationf 1473(678)

the contaminationis 50 true 5sand 50 true 6s

1166(1416)

all contaminationis from true 6s 612

(1526)Experiment

30031 6137 471 190 281

no contamination 1531(529)

contamination is50 true 5s and50 true 6s

1715(1138)

all contaminationis from true 6s

1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

5 signicant at the 001 level 5 signicant at the 005 levela pc 5 the observed proportion of individuals from high standard states in the affected score groupb Ntreat 5 the total number of observations in the treatment states across all score groupsc NT 5 the number of observations in the affected score group in the lower standard statesd The assumed number of treatment group individuals with the correct true scoree The assumed number of treatment group individuals with true scores that would place them in score

group 5 or 6 these are the lsquolsquocontaminatingrsquorsquo individualsf We include our original estimates in this table (the estimates we obtain under an assumption of no

contamination) for ease of comparisong The mean earnings of score group 5 for the treatment states in this experiment are actually slightly

lower than the mean earnings of the treatment group here and so a calculation using the earnings of true 5swould result in a higher adjusted estimate


The simple story from Table VII is that while all of theadjusted estimates are noisy the point estimates in Table VII arenot far from our original difference-in-differences estimates dis-played in Table V It is only when we assume that all of the 451lsquolsquocontaminantsrsquorsquo in the affected score group in experiment 4 aretrue 6s that our estimates are driven substantially down (to $612)Under a more reasonable assumption that the contamination isequally spread between 5s and 6s the original estimate falls 20percent to $1166 Meanwhile in experiments 3 and 3 thestronger contaminating assumption yields adjusted estimatesthat are only 10 and 3 percent below the original estimates

In summary differential effort across low and high standardstates may lend an upward bias to our estimates The evidencehowever is that the magnitude of this bias especially in experi-ments 3 and 3 is not so severe as to discredit our originalestimates

C Endogeneity Bias Associated with State Behavior

A second area of potential bias consists of differential statepolicies between low and high standard states We examine twodifferent types of policies that would bias our results stateminimum wages and state course graduation requirements

If states with lower GED standards tended to have higherminimum wages then our estimates could be biased upward Thereasoning is as follows It could be that higher scoring GED-holders are not generally affected by minimum wage laws (as aresult of their relatively higher skills) while lower scoringGED-holders are If high minimum wage states are states thatsystematically have relatively low GED passing standards thenour treatment group individuals may have higher earnings rela-tive to the comparison group for reasons not associated with theeffect of the GED on earnings

An examination of the data suggests that this is not a seriousconcern however simply because there is little variation in 1995state minimum wages Most states in the United States in 1995had a minimum wage of $425 Only nine states had higherminimum wages ranging up to $530 in Alaska However none ofthe four states in our lowest passing standard category had aminimum wage higher than $425 and states with higher mini-mum wages are relatively equally divided between low and highpassing standard states

One could argue that it is the lsquolsquoeffective minimumrsquorsquo wage


rather than the absolute minimum that is critical for ouranalyses Lee 1999 shows that the same absolute minimumwage can differentially compress wages at the bottom end of thewage distribution in low versus high wage states If lower GEDpassing standard states are also systematically lower wage statesthen this type of wage compression might cause the wages of lowscoring GED-holders (those at the margin of passing) and higherscoring GED-holders to be closer in low standard states biasingour estimates upward This type of wage effect is essentially astate-skill interaction among GED-holders that is not included inour model In the subsection on lsquolsquoEmployer Behaviorrsquorsquo whichfollows we present evidence suggesting that missing state-skillinteractions are not important factors in explaining our results

Another potential policy difference that could affect ourestimates concerns high school graduation requirements Highergraduation requirements could be associated with higher dropoutrates As a result the underlying ability of the pool of dropoutscould be different in states with different high school graduationrequirements In particular if states with low GED passingstandards have systematically higher graduation requirementsfor high school then unobserved ability and motivation in the poolof dropouts in these states may be greater than in other states Inthis case our estimates would have an upward bias

Our examination of course graduation requirements in 1990shows however that lower GED passing standard states hadlower course graduation requirements rather than higher onesThe mean number of Carnegie units21 required for graduation byGED passing standard was 184 in the lower standard states 193in the medium standard states and 20 in the higher standardstates Thus to the extent that there is a systematic relationshipbetween course graduation requirements and GED passing stan-dards it does not work in a direction that would lead to an upwardbias in our results

D Endogeneity Bias Associated with Employer Behavior

Finally it could be that employers behave differently in lowversus high GED passing standard states in ways that would bias

21 A Carnegie unit is a unit for measuring the amount of high schoolacademic work by the number of classroom hours spent in one subject Developedin 1899 by a committee of the National Education Association one Carnegie unit ina subject equals 120 classroom hours lasting 40 to 60 minutes each and meetingfour to ve times a week 36 to 40 weeks during the school year Most high schoolsrequire sixteen Carnegie units per subject for graduation including requiredcourses such as English mathematicsAmerican history and science


our estimates These types of biases show up as omitted interac-tions in our model In particular omitted statendashscore or statendashGED interactions are of concern Statendashscore interactions thatfavor high-scoring dropouts in high-standard states would biasour estimates upward while interactions that favor high-scoringdropouts in low-standard states would bias our estimates down-ward StatendashGED interactions that favor GED-holders in highstandard states would bias our estimates upward while suchinteractions that favor GED-holders in low standard states wouldbias our estimates downward

It is possible that employers in states where the GED-passingstandard is relatively high give more credence to and place moreemphasis on the credential than do employers in low standardstates In this case ceteris paribus GED-holders would earn moreon average in high standard states than similar GED-holders inlower standard states and this type of mechanism would cause anupward bias in our estimates

Ex ante we cannot sign the bias associated with possiblestatendashscore interactions We can however bring ex post evidenceto bear on the question tting models where we let the effect ofGED test scores as captured by a variable constructed to approxi-mate a continuous GED score have different slopes in lowmedium and high standard states22 Table VIII shows estimatesfrom a model tted over cells representing score groups above theGED passing standard in each state This model t separately forwhites and minorities allows for xed state effects a linear effectof GED test score and then different slopes for GED test score ineach of the three different groups of states dened by lowmedium and high GED passing standards

The nonsignicant coefficient estimates on the passing stan-dard by score interactions suggest that statendashskill interactionsare not a serious problem for our estimates Furthermore thepoint estimates imprecise as they are indicate that any statendashskill interactions would tend generally to favor more highlyskilled individuals in higher standard states

These results combined with the most plausible statendashGEDinteraction effects suggest that any bias from either or both of thetwo effects would bias our estimates upward To assess howserious any total bias from these effects might be we use the

22 We discuss the construction of this variable and the weighted leastsquares tting of the model in the Statistical Appendix


results from experiments 3 and 4 in Table V We expect anyupward bias due to omitted interactions to be greater in experi-ment 3 than in experiment 4 The reason is that experiment 3 usestreatment states with the lowest standards in the United Statesin comparisons with the high standard states of New York andFlorida while experiment 4 uses treatment states with mediumstandards in comparisons with New York and Florida Thus thespread between the experiment 3 and experiment 4 estimatesgives us some measure of how our results are being affected by thetotal bias attributable to omitted interactions23 That spread

23 This result rests not only on our stated assumption regarding thedirection of the bias of the statendashtest-score interactions but also on the assump-tions that (1) any statendashtest-score interaction is monotonically increasing in testscores and (2) that at least within the narrow scoring range (score groups 3 and 4)represented by the two experiments the treatment effect is the same We thinkboth of these assumptions are reasonable given that we are looking at a relativelywell-dened and narrowly dened sample namely high school dropouts who scoreabove a certain range on the GED battery of exams

TABLE VIIIWEIGHTED LEAST SQUARES REGRESSIONS TESTING FOR STATE-SKILL INTERACTIONSa

(DEPENDENT VARIABLE IS THE CELL MEAN OF 1995 ANNUAL EARNINGS AND

STANDARD ERRORS ARE IN PARENTHESES)b

Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)

GED score medium states dummy 548 2 148(402) (870)

GED score high states dummy 215 991(450) (860)

State dummies Yes YesF-test of H0

GED score medium states andGED score high states interactions are jointly zero 0347 0273

Number of cells 266 192Number of individual level observations 28193 7778

5 signicant at the 001 level 5 signicant at the 005 level 5 signicant at the 010 levela Low medium and high refer to states with low medium and high GED passing standards Low

standard states are the omitted groupb Construction of the weights and our use of the cell mean data is described in the Statistical Appendixc The lsquolsquoGED scorersquorsquo variable is the within-state average of the GED mean score (the mean of the ve tests

in the GED battery) in each GED score group The models are t using only GED score groups 5ndash10 and usingall available states


however is less than 4 percent ($1473 versus $1531) Thereforewe conclude that omitted interaction bias cannot go far inexplaining our estimates

In summary as with almost any quasi-experimental settingthere are scenarios under which our identifying assumptionswould not hold We have examined the most plausible violations ofour identifying assumptions and have found our estimates forwhite dropouts to be robust to various specication tests

E Pretreatment Contrasts

As a nal specication check we examine pretreatmentdifference-in-differences contrasts In general a necessary butnot sufficient condition for the validity of inferences drawn froman experiment is that there be no pretreatment differencesbetween treatment and comparison outcomes Figures I and IIIshow no statistically discernible pretreatment differences in themean earnings of the treatment and comparison groups inexperiments 4 and 3 In particular point estimates of pretreat-ment earnings comparisons in experiment 3 ($110 and $58) arealmost zero In contrast Figure II shows that the pretreatmentcomparisons in experiment 3 are negative and statistically signi-cant We might therefore expect the treatment group to fare worsethan the comparison group in the absence of treatment suggest-ing that estimates of the treatment effect may be downwardlybiased with these groups

VI RESULTS FOR MINORITY DROPOUTS

Unlike our ndings for whites we nd no statistically signi-cant GED-effect for minority group members We discuss twopotential explanations for these differential minority results

Using a separate data set from Florida that allows us todetermine the testing center at which individuals took the GEDexams we nd that about 17 percent of the 16ndash21 year-oldminority males in Florida who obtained their GED in 1995 (theearliest year available in these data) did so while incarceratedThe comparable gure for young white males is 4 percent and forwhite females and minority females the gures are 03 and 1percent respectively Thus one explanation for our minorityresults may lie in the relatively high proportion of young minoritymales who obtain their GED while incarcerated


There are two complementary reasons why individuals in ourdata who obtained their GED while incarcerated might havedepressed earnings relative to individuals who obtained it in amore conventional setting First the stigma of incarceration maydepress the postprison earnings of dropouts and eliminate anypositive signaling value of the GED credential Second manydropouts who obtained a GED while incarcerated may have zeroearnings ve years later because they are still in prison With thedata used in this project we have no way of sorting out theimportance between these two complementary explanations forthe difference between the signaling value of the GED for whiteand minority dropouts

A second explanation for the minority results hinges on thefollowing (a) different behavioral assumptions behind why indi-viduals acquire the credential (b) different white-minority distri-butions associated with those behavioral assumptions and (c)employers engaging in statistical discrimination when faced withlack of information24 Under this explanation one set of dropoutsacquires a GED because they either place a value on the creden-tial or they perceive that employers place a value on the creden-tial Another set of dropouts acquires the credential incidentallybecause it is lsquolsquoquasi-compulsoryrsquorsquo in a program from which they areseeking benets Examples of such programs are Aid to Familieswith Dependent Children (AFDC) Job Training Partnership Act(JTPA) programs the Job Corps and many community levelyouth employment programs directed toward young dropouts

The GED may be a signal of productive attributes for thegroup that actively pursues the signal and a signal of lsquolsquogovernmentprogram participationrsquorsquo for those who acquired it in an incidentalfashion Employers may discount the value of the lsquolsquoincidentalrsquorsquoGED relative to the lsquolsquoactively pursuedrsquorsquo GED If the distributions oflsquolsquoactively pursuedrsquorsquo and incidental GED-holders differ by raceethnicity then we would expect the signaling returns to thecredential to differ across the white and minority groups

Exploring this lsquolsquogovernment-program-GEDrsquorsquo hypothesis is dif-cult because the only estimates we can generate for adultdropouts come from information on JTPA classroom programswhile the only estimates for dropouts younger than 21 come fromthe relatively small yearly cohorts of the Job Corps There is no

24 We are indebted to Caroline Hoxby for the ideas in this part of thediscussion


systematic data collection on the large numbers of dropout youthswho received GEDs while in community-based government-funded employment programs Bos 1999

Nevertheless we can get some rough estimates of the percent-ages of GEDs that are obtained through government sponsoredprograms by whiteminority status To estimate GED numbersobtained through JTPA participation we start with the fact that146952 individuals participated in JTPA classroom programs in1990 Committee on Ways and Means U S House of Representa-tives 1993 Next we assume that white-minority proportions inthe classroom component of JTPA programs are the same asoverall JTPA proportions 48 percent minority and 52 percentwhite We next assume that for both whites and minorities 42percent of those in classroom programs obtained their GED in199025 Based on these assumptions we estimate that in 1990about 29600 minority and 32000 white dropouts obtained theirGED through JTPA participation

Meanwhile 70 percent of the participants in Job Corps areminority group members Again we assume that 42 percent of the49200 new 1990 enrollees in the Job Corps who were dropoutsobtain their GEDs and that 70 percent of the newly mintedGED-holders are minority and 30 percent are white Based onthese estimates we estimate that an additional 14500 minoritydropouts obtained their GED while participating in Job Corpsactivities as compared with 6200 white dropouts Thus weestimate that a total of 44100 minority dropouts and 38200 whitedropouts obtained their GEDs through these two programs forwhich we can generate estimates

Finally we use weighted tabulations of the GED-holders inthe NLSY to estimate that 73 percent of the 439117 new GEDcredentials issued in 1990 GED Testing Service 1991 went towhite dropouts and 27 percent to minority dropouts Based onthese assumptions 37 percent of the GEDs awarded to minoritydropouts and 12 percent of the GEDs awarded to white dropoutsin 1990 may have been associated with either JTPA classroomtraining or Job Corps participation If GEDs awarded throughgovernment-sponsored summer youth employment programs godisproportionately to young minority dropouts then an evenhigher percentage of GEDs awarded to minorities could be

25 We base the 42 percent on GED attainment gures in the JOBSTARTDemonstration Cave et al 1993 At a 70 percent pass rate this would mean thatabout 60 percent of program participants attempted the GED


through government program participation These estimates ofGEDs obtained through government-sponsored training at leastgo in the right direction and are potentially of sufficient magni-tude to support the hypothesis outlined above26

Support for this government-program-GED hypothesis basedon evaluations of random assignment social experiments is mixedEvaluations of the JOBSTART Demonstration a JTPA-fundedprogram directed at 17ndash21 year-old economically disadvantagedschool dropouts found that program participation increased GEDreceipt by a statistically signicant 134 percent points Cave etal 1993 However the evaluation also indicated that the treat-ment group did not have subsequently higher earnings than thecontrol group a nding that ts with the government-program-GED hypothesis

VII SUMMARY AND CONCLUSIONS

Using differential state GED passing standards as an identi-cation strategy we nd that the signaling value of the GEDincreased the 1995 earnings of young white dropouts on themargin of passing the exams by 10 to 19 percent We nd nostatistically signicant evidence that the credential impacted the1995 earnings of young minority dropouts in the same scoringrange Our results are robust across natural experiments that usedifferent treatment and comparison groups Finally sensitivityanalyses indicate that possible violations of the identifying assump-tions used to obtain our estimates can explain little of thoseestimates

At rst glance our results may appear inconsistent withCameron and Heckmanrsquos 1993 nding of no statistical differenceat age 25 between the average earnings of GED-holders and thoseof observationally similar uncredentialed dropouts However acloser look shows that the differences can be explained bydifferences in samples and methodologies In particular ourestimates are based only on the lowest skilled GED-holders those

26 We note that if employers can actually ascertain who did and did notobtain their GED on their own initiative then the lsquolsquorealrsquorsquo value of the GED forwhites would be $1500(1 2 012) 5 $1705 and that based on this we should get anestimate for minorities of $1705 3 037 5 $631 While this estimate is within thecondence intervals of our estimates for minorities in Table V it is relatively farfrom the point estimates for minorities in that table We also note however that ifthe lsquolsquoGED-in-prisonrsquorsquo effect we discussed earlier is important in depressing theearnings of minority males then we would expect a pooled (male and female)minority estimate to be well below the $631 estimate


on the margin of passing the exams In contrast Cameron andHeckman base their nding on a sample of dropouts varying ininitial skill level and they adopt the assumption that the impactof the GED on earnings is independent of initial skill level If GEDeffects are concentrated among the least skilled and are negligiblefor the more-skilled dropouts then our results pointing to largeGED effects for the dropouts on the margin of passing the examsare potentially consistent with those of Cameron and Heckman

The ndings of two recent papers support this explanationUsing data from the NLSY Murnane Willett and Boudett 1999t models in which the effect of GED acquisition on subsequentearnings is allowed to vary across initial skill levels They ndthat three years after obtaining the credential the annualearnings of male GED-holders who left school with very lowcognitive skill levels are about 10 percent higher than theearnings of low scoring uncredentialed dropouts They nd noeffect of the GED on the earnings of dropouts who left school withstronger skills Murnane Willett and Tyler 1999 report thesame patterns in HSB data

The ndings of these papers support our primary estimatesindicating substantial GED effects for low-skilled dropouts Thesepapers also show that if the effect of the GED is constrained to beindependent of initial skill level (the assumption made by Cam-eron and Heckman 1993) the average difference between theearnings of GED-holders and observationally similar permanentdropouts is smallmdasha result of the negligible effect of the GED onthe earnings of higher skilled dropouts

In closing we note that our estimates represent the privatereturns to the labor market signaling component of the GED As aresult the ndings in this paper potentially raise important socialwelfare questions If the GED signal only serves to redistributeearnings raising the earnings of dropouts who acquire a GED atthe expense of dropouts who do not acquire the signal then thegross social returns to the credential are zero and the net socialreturns are negative (since there are costs associated with obtain-ing the signal) If the GED signal leads to better matches betweendropouts and jobs however then the social returns can equal oroutweigh the private returns Stiglitz 1975 We have no empiricalmethod of determining whether or not employers use the GED tobetter match dropouts to jobs and so the impact of the GED ontotal social welfare remains an open question

If there is a job-matching component to the credential


however then the social returns to the GED are dependent uponthe passing standards To see this note that a standard either solow as to allow everyone to pass or so high as to allow no one topass would provide no job-matching information Furthermore itcould be that maximization of total social welfare requires state-specic passing standards since the underlying distributions ofthe skills of dropouts and the demand for skills by employers mayvary from state to state Thus the decentralized manner in whichGED passing standards are currently established may approxi-mate the regime under which total social welfare would be thegreatest Less clear however is whether the current set of statepassing standardsmdashwhere states as diverse as New York andIowa have the same GED passing standardmdashis the set thatmaximizes any social returns to a GED

DATA APPENDIX

A Data Sources

As discussed in the text the underlying individual-level datafor this project come from four different sources the statedepartments of education in Connecticut Florida and New Yorkand the GED Testing Service (GEDTS) Data from the three statedepartments of education represent the universe of 1990 GEDattempters who as of 1995 had not reattempted the test battery

The data from the GED Testing Service were originallycollected by the GEDTS as part of an ongoing several-year projectto study responses to proposed new GED test items and as aresult of idiosyncratic administrative decisions these data con-tain considerably fewer observations in every state than thenumber of test-takers indicated in GEDTS reports on the 1990tests GED Testing Service 1991 Based upon our review of theprocesses that produced the GEDTS data we have no compellingreason to believe that within-state sample selection bias is aproblem with these data Also about 50 percent of the observa-tions in the data forwarded to us had missing writing scoresInquiries to programmers at GEDTS indicated that the writingscores were missing as a result of administrative record-keepingdecisions at GEDTS To retain observations and prevent sampleselection bias that might otherwise occur if we eliminated recordswith missing writing scores we used a quartic function of the fourtest scores available to predict the writing score and we tseparate models for males and females whites and minorities


B SSN Validation

Table IV in the body of the paper shows that SSA program-mers successfully matched and validated a high percentage ofrecords Except in the case of New York which is explained abovematches were validated using SSA data on sex race and year ofbirth A proposed match was assumed to be valid if the GED andSSA sex codes matched and the GED and SSA years of birthdiffered by no more than two years in absolute value and if theGED and SSA race codes did not disagree on white versus blackdesignation

C Social Security Earnings Coverage and Taxable Maximum

As detailed by Angrist

OASDI and HI (Medicare) are contributory programs in which coveredworkers pay a tax and report earnings to the SSA via employers or theInternal Revenue Service The SSA keeps track of taxable earnings on theSummary Earnings Record (SER) Under the Federal Insurance Contribu-tions Act (FICA) of the Self-employment ContributionsAct (SECA) earningsare taxed for Social Security purposes up to the Social Security Taxablemaximum and are therefore recorded on the SER only up to this maximumThe taxable maximum has been raised every year By 1980 over 85 percentof covered male workers had earnings below the maximum A separate andhigher taxable maximum was instituted for HI in 1991All taxable earningswhether reported for the purposes of OASDI or HI and whether reported aspart of a mandatory or voluntary coverage provision should appear on theSER

About 95 percent of all jobs in the United States were covered by theOASDI program as of 1991 Exceptions fall into ve major categories (1)Federal civilian employees hired before 1984 (2) railroad workers (3) someemployees of State and local governments already covered under a retire-ment system (4) household and farm workers with low earnings and (5)persons with no wage and salary earnings and very low earnings fromself-employmentMembers of the uniformed services have been covered since1956 and receive noncontributory wage credits to improve their insuredstatus Recent important changes include the 1983 coverage of most federalemployees and employees of nonprot organizations HI coverage of manystate and local employees in 1986 OASDI coverage of State and localemployees without an employer retirement plan in 1990 and coverage ofreserve soldiers in 1987 Angrist 1997

Taxable maximum values for the years 1991ndash1995 are1991mdash$534001992mdash$555001993mdash$576001994mdash$606001995mdash$61200


Given the generally low annual earnings of dropouts we did notanticipate that earnings values top-coded at the taxable maxi-mum would be a problem in our data and this turned out to be thecase In the GEDTS Connecticut and Florida data27 the percentof young dropouts with any top-coded values are 01 percent 02percent and 02 percent respectively

G SSA Condentiality Edit

The SSA condentiality regulations that affected the datareleased to us censored cells that had fewer than three individualsor cells where the standard deviation of earnings was zero Thiscensoring masked the 1995 mean earnings in 164 of the 2138 totalcells released to us on individuals who took the GED battery in1990

STATISTICAL APPENDIX

To t the models used to obtain the estimates in Table VIIIwe must rst assign a metric to the GED score groups displayed inTable V We have chosen to use the within-state average of theGED mean score in each score group as this metric

Given this scaling of the GED score groups the estimates inTable VIII are based upon regressions where we are interested inyic the true earnings of the ith person in the cth cell in our dataHowever as a result of the nature of the data released to us by theSocial Security Administration we only have a sample estimate ofthe mean earnings microc in the cth cell We assume that yic 5 microc 1 wicwhere wic N(0 h 2)

We would like to t models in which the dependent variable isthe cell population mean earnings microc and the explanatory vari-ables of primary interest are associated with individuals whomake up the cth cell

We do not know the population cell mean earnings microc but wepossess the sample estimates for each cell microc provided by theSocial Security Administration We assume that

microc 5 microc 1 uc uc N(0 j 2)

In our estimation we must account for the fact that the microcrsquos areestimated with varying precision

27 We do not have top-coding information on New York


We represent the models we want to t as

microc 5 xc d 1 n c

where x is a vector of the independent variables and d is a vector ofregression parameters to be estimated However because the microc

are unknown we must instead t the following model

(A1) microc 5 xc d 1 ( n c 1 uc) 5 xc d 1 e c

Assuming that u and n are independent that E n n 8 5 s 2I andthat

Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n

with nc the cell frequency in the cth cell we have

E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20

middot middot middot 0

middotmiddot middotmiddotmiddotmiddot

s 2 1 v C2

s 2

where C is the total number of cellsEfficient estimation of d in equation (A1) requires a consistent

estimate of V which in turn requires estimates of each v c2 and of

s 2 Estimates of the v c2 are obtained by using the cell variances

(supplied by the SSA) to estimate the h c2 in combination with the

cell frequencies nc (also supplied by the SSA) Following Ha-nushek 1974 we use a two-step estimator for s 2 We then use ourestimate of V to form a weighted least squares (WLS) t of themodel in equation (5)

BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES

Angrist Joshua D lsquolsquoEstimating the Labor Market Impact of Voluntary MilitaryService Using Social Security Data on Military Applicantsrsquorsquo EconometricaLXVI (1997) 249ndash288

Arrow Kenneth J lsquolsquoHigher Education as a Filterrsquorsquo Journal of Public Economics II(1973) 193ndash216

Bos Hans Personal communicationApril 1999


Boudett Kathryn Parker lsquolsquoIn Search of a Second Chance The Consequences ofGED Certication Education and Training for Young Women without HighSchool Diplomasrsquorsquo PhD thesis Harvard University 1998

Cameron Stephen V and James J Heckman lsquolsquoThe Nonequivalence of HighSchool Equivalentsrsquorsquo Journal of Labor Economics XI (1993) 1ndash47

Cave George Hans Bos Fred Doolittle and Cyril Toussaint JOBSTART FinalReport on a Program for School Dropouts (Manpower Demonstration Re-search Corporation 1993)

Committee on Ways and Means U S House of Representatives lsquolsquoOverview ofEntitlement Programs 1993 Green Bookrsquorsquo (Washington DC U S Govern-ment Printing Office 1993)

Erwin Catherine Personal communication September 1999GED Testing Service GED 1990 Statistical Report (Washington DC American

Council on Education 1991)Hanushek Eric A lsquolsquoEfficient Estimators for Regressing Regression Coefficientsrsquorsquo

American Economic Review LXIV (1974) 66ndash67Jaeger David and Marianne Page lsquolsquoDegrees Matter New Evidence on Sheepskin

Effects in the Returns to Educationrsquorsquo University of Michigan PopulationStudies Center Research Report No 94-307 1994

Kane Thomas J and Cecilia Rouse lsquolsquoLabor Market Returns to Two-Year andFour-Year Collegersquorsquo American Economic Review LXXXV (1995) 600ndash614

Lang Kevin and David Kropp lsquolsquoHuman Capital versus Sorting The Effects ofCompulsory Attendance Lawsrsquorsquo Quarterly Journal of Economics CI (1986)609ndash624

Lee David S lsquolsquoWage Inequality in the United States during the 1980s RisingDispersion or Falling Minimum Wagersquorsquo Quarterly Journal of EconomicsCXIV (1999) 977ndash1024

Murnane Richard J John B Willett and Katherine P Boudett lsquolsquoDoes Acquisitionof a GED Lead to More Training Post-Secondary Education and MilitaryService for School Dropoutsrsquorsquo Industrial and Labor Relations Review LI(1997) 100ndash116

Murnane Richard J John B Willett and Katherine P Boudett lsquolsquoDo MaleDropouts Benet from Obtaining a GED Post-Secondary Education andTrainingrsquorsquo Evaluation Review XXIII (1999) 475ndash502

Murnane Richard J John B Willett and John H Tyler lsquolsquoWho Benets fromObtaining a GED Evidence from High School and Beyondrsquorsquo Review ofEconomics and Statistics LXXXII (2000) 23ndash37

Park Jin Heum lsquolsquoEstimation of Sheepskin Effects and Returns to Schooling Usingthe Old and the New CPS Measures of Educational Attainmentrsquorsquo PrincetonUniversity Industrial Relations Section Working Paper No 338 August 1994

Spence Michael lsquolsquoJob Market Signalingrsquorsquo Quarterly Journal of EconomicsLXXXVII (1973) 355ndash374

Stiglitz Joseph lsquolsquoThe Theory of Screening Education and the Distribution ofIncomersquorsquo American Economic Review LXVI (1975) 283ndash300


Ideal data for identifying the returns to a signal wouldcontain exogenous variation in signaling status among individu-als with similar levels of human capital In this paper weapproximate that ideal in estimating the returns to a particularlabor market signal the General Educational Development (GED)credential and in so doing we provide a test of the signalinghypothesis for this credential

The GED was introduced in 1942 to provide a way forveterans without a high school diploma to earn a secondary schoolcredential The GED program has evolved markedly so that todaythe credential is the primary lsquolsquosecond chancersquorsquo route to high schoolcertication for school dropouts in the United States Each yearmore than one million young people drop out of school andeventually about one-third of them will acquire a GED

Using interstate variation in GED passing standards we areable to compare individuals who all chose to try to obtain a GEDand who have the same GED test scores but who differ in GEDstatus because of differences in the stringency of the passingstandards in their state of residence We will argue that thislsquolsquonatural experimentrsquorsquo research design allows us to net out theeffect of human capital on earnings leaving clear estimates of thesignaling value of the GED credential in the labor market

Our estimates indicate that for young white dropouts who areon the margin of passing the GED exams the signaling value ofthe GED increases annual earnings by 10 and 19 percent We donot nd statistically discernible returns to the GED signal foryoung minority dropouts a puzzle we discuss later in the paper2

In the next two sections we discuss the GED credential andour identication strategy We then present our data methodsand results The paper closes with an extensive discussion of thepotential threats to our identication strategy an interpretationof our results and a summary

I THE GED IN THE LABOR MARKET AND IN THE LITERATURE

The ve tests in the GED battery cover mathematics writingsocial studies science and lsquolsquointerpreting literature and the artsrsquorsquo

2 Because of data limitations we are only able to identify individuals in ourdata as being members of one of two racialethnic groups (1) white non-Hispanicsand (2) everyone else The latter group includes African-Americans Hispanicsand other minority group members For expository clarity we refer to the twogroups as lsquolsquowhitesrsquorsquo and lsquolsquominoritiesrsquorsquo in this paper






Completionmethod

Yearab

1975 1980 1985 1988 1989 1990 1991 1992 1993 1994 1995 1996



























Intercept 9518 9320 9455 9199(0057) (0079) (0059) (0081)

Female 2 0397 2 0306 2 040 2 0298(0019) (0030) (0022) (0032)

GED 0162 0164 0094 0083(0072) (0109) (0072) (0109)


Math test score 0012 0008(0002) (0002)


































Minimum score

Mean score

45 5 45




III RESEARCH DESIGN

A Data




















Stategroupa

Col 1N

Col 2Minimum

scoregroup

for GEDb


comparison in




















where







predictors








where













IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES



























Completionmethod

Yearab

1975 1980 1985 1988 1989 1990 1991 1992 1993 1994 1995 1996



























Intercept 9518 9320 9455 9199(0057) (0079) (0059) (0081)

Female 2 0397 2 0306 2 040 2 0298(0019) (0030) (0022) (0032)

GED 0162 0164 0094 0083(0072) (0109) (0072) (0109)


Math test score 0012 0008(0002) (0002)


































Minimum score

Mean score

45 5 45




III RESEARCH DESIGN

A Data




















Stategroupa

Col 1N

Col 2Minimum

scoregroup

for GEDb


comparison in




















where







predictors








where













IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES








































Intercept 9518 9320 9455 9199(0057) (0079) (0059) (0081)

Female 2 0397 2 0306 2 040 2 0298(0019) (0030) (0022) (0032)

GED 0162 0164 0094 0083(0072) (0109) (0072) (0109)


Math test score 0012 0008(0002) (0002)


































Minimum score

Mean score

45 5 45




III RESEARCH DESIGN

A Data




















Stategroupa

Col 1N

Col 2Minimum

scoregroup

for GEDb


comparison in




















where







predictors








where













IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES
















































Minimum score

Mean score

45 5 45




III RESEARCH DESIGN

A Data




















Stategroupa

Col 1N

Col 2Minimum

scoregroup

for GEDb


comparison in




















where







predictors








where













IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES
























III RESEARCH DESIGN

A Data




















Stategroupa

Col 1N

Col 2Minimum

scoregroup

for GEDb


comparison in




















where







predictors








where













IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES



































Stategroupa

Col 1N

Col 2Minimum

scoregroup

for GEDb


comparison in




















where







predictors








where













IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES



























where







predictors








where













IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES

























where













IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES

























IV RESULTS











IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES




























IN PARENTHESES)



standardcontrast


standardcontrast




Low 9628 7849 1779 9362 7843 1509 9362 8616 746(361) (565) (670) (400) (312) (507) (400) (219) (456)

High 9981 9676 305 9143 9165 2 23 9143 9304 2 162(80) (65) (103) (135) (63) (149) (135) (135) (150)


1473(678)

1531(529)

907(481)


Low 6436 8687 2 2252 7005 7367 2 363 7005 6858 147(549) (690) (882) (347) (347) (495) (347) (290) (452)

High 7560 8454 2 894 7782 8375 2 593 7782 7568 214(184) (96) (207) (214) (93) (233) (214) (133) (252)


2 1357(906)

231(548)

2 67(518)

































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES
















































Panel AGED-holdersa



Low Medium High

Low 890 85 26Medium 34 909 56High 21 81 899

































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES

















































assumptions

pca Ntreat

b NTc pc(Ntreat) d

NT 2pc(Ntreat) e


assumptionof

Adjustedestimate

Experiment4

0011 18332 653 202 451



1166(1416)


(1526)Experiment

30031 6137 471 190 281



1715(1138)


1374(1219)

Experiment3g

0062 6137 471 380 91



936(568)


881(573)

































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES


















































Whites Minorities

Intercept 10385 5178(1765) (3443)

Female dummy 2 4316 2 1429(138) (269)

GED scorec 272 720(340) (690)












































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES

























































DATA APPENDIX

A Data Sources




B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES























B SSN Validation





















microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES




































microc 5 xc d 1 n c





Eumun 5v c

2 m 5 n where v c

2 5h c

2

nc

0 m THORN n


E e e 8 5 s 2V 5 s 2

s 2 1 v 12

s 20



s 2 1 v C2

s 2






BROWN UNIVERSITY

HARVARD UNIVERSITY

HARVARD UNIVERSITY

REFERENCES























estimating the labor market signaling value of the ged · estimating the labor market signaling...

Documents