Using a Structured Decision Making Protocol to

Stratify Caseloads in the Child Support Program

1

ERICSASteven J. Golightly, Ph.D.

May 23, 2011

Structured Decision Making

• Decision making in which the process and criteria that must guide decision making are formally defined (Shook & Sarri, 2007)

• Can be clinical or actuarial

2

Decision Theory

• In statistical theory, the process of making choices between alternatives (Berger, 1993)

• Can be informative or descriptive

3

Risk Assessment

• Likelihood that a harmful event will occur and such an event’s likely security (Hughes & Rycus, 2007)

• Study assesses child support cases in terms of risk of non payment

4

Risk Assessment Protocols

• Child Welfare• Criminal Justice• Health Care• Credit – Risk Management

5

Dissertation

• Overachieving goal of child support• Novel approach for assessing risk in order to

determine the level of enforcement intervention necessary to collect child support

• Stratification up front • Acknowledging differences• Prioritization on a rational basis• Knox County (TN) example (PSI, 2001)

6

Research Questions

1. Can CP data be utilized to determine enforcement difficulty (e.a., risk assessment)?

2. Can child support use structured decision making?

3. Potential impact of case stratification using CP data?

7

Research Hypothesis

• Are there relationships between CP data and the child support agency’s ability to collect full payment from the NCP for at least 6 consecutive months?

8

Research Design and Method

• Non experimental design• Non programmatic analysis• Los Angeles County CSSD Archival data (FFY

2008)• Secondary data use is cost effective• Examine the relevance of various independent

variables as determinants of case success

9

Dependent Variable

• Case Success• Receipt of the ordered amount of child support for at

least 6 consecutive months

10

Independent Variables• Age• Gender• Residential zip code• Ethnicity• Marital status• Welfare status• Number of children• Relationship to each child• Ages of the children• Paternity status• Court Order

11

Purpose of Study

• Using CP data obtained at intake – Is it feasible to determine case success?

12

Study Rationale

• Reduced funding and staff• More cases• Increased need for efficiencies

Significance of the Study

• If correlation can be shown between CP data and case success, stratification could be implemented earlier in the process

13

Assumptions

• FFY 2007-08 cases were typical of cases opened in other years

• Data in applications assumed to be accurate• Benchmark of 6 months of consecutive

payments constitute case success

14

Limitations

• Time period was the beginning of the economic downturn

• Unemployment rate was only 5.1% in 2007• Given unique urban nature of Los Angeles

County, may not be possible to generalize results to other Jurisdictions

15

Delimitations

• Does not intend to provide a model for using NCP data

• Excludes consideration of reasons why NCPs did not pay child support

16

Bounds

• May be useful only in Los Angeles• Generalizing prediction tools across

Jurisdictions may be “suspect” (Farrington and Tarling, 1985)

17

Theoretical Framework

• SDM is crucial in many fields• SDM relies on gauging risk • Child support

• Low Risk• High Risk

• Decision Theory• Normative • Descriptive

• Regression Analysis• Delinquency

• Time & money18

Literature Review

• Problems associated with child support programs

• Significant change in 50 years• Family structure changes• Caseload composition• Funding/Staffing• CP data literature

19

Method

• Logistic regression• Blomberg & Long (2006)

• Importance of “success” definition

20

Research Design and Approach

• Design is the structure that holds all elements of the research project together

• Two basic categories• Experimental• Non experimental

• This study utilized a non experimental research design

• Nonparametric design• Predictive correlational study

21

Research Design and Approach (con’t)

• Archival data (FFY 2008) • Los Angeles County CSSD• ARS• Sequel Server

• SQL Server Management • Studio Software

• IBM SPSS statistics 18 software to analyze data

22

Scores and Calculations

• Correlation Coefficient Calculation• No relationship (0)• Strong relationship (1)

23

Design Justification

• Quantitative approach• Non experimental• Regression analysis• Non parametric

• Cramer’s V Test

• Predictive Correlation

24

Participants and Sample Size

25

• Custodial Parents Demographics• Caseload Composition

Study Sample

26

• FFY 2008 – Reasons for Using• 19,000 cases - Universe• Sample size of 377 = 95% confidence level and

considerable interval of 5%• Study used 1501 cases

• Random selection

Ethical Considerations

27

• Data de-identified and presented anonymously• Privacy & confidentiality• Transfer of data from SQL Software

Data Screening & Data Cleaning

28

• 1501 randomly selected cases• Cleanup to ensure No missing valves and

accurate + initiative• All cross tabulation cells had at least 5 members • Decision Points

• Age (recording 14 – 41)

• Zip Codes (first 3 digits)

• Ethnicity (truncated)

• Age of Children (parameter determination)

• Paternity Status (duplicative

Descriptive Statistics

29

• N = 1501• 1456 Females (97%)• 856 Hispanics (57%) • 375 African Americans (25%)• 163 White (11%)• 792 Never Married (53%)• 135 Married (9%)• 940 Currently/Formerly Assisted (63%)

Descriptive Statistics (con’t)

30

• Age Frequency %

• Gender

Male 45 3%

Female 1456 97%

14 – 27 44 3%

18 – 21 352 23%

22 – 25 300 20%

26 – 30 302 20%

31 – 35 227 15%

36 – 40 147 10%

41+ 129 9%

Descriptive Statistics (con’t)

31

• Residential Zip Code Frequency %900 430 29%902 207 14%903 – 907 155 10%908 79 5%909 – 916 111 7%917 191 13%918 – 934 76 5%935 – 986 128 9%Other 124 8%

Descriptive Statistics (con’t)

32

• Ethnicity Frequency %Hispanic 856 57%Black 375 25%

White 163 11%Filipino 25 2%Asian 22 1%Other 10 1%Unknown 53 3%

Descriptive Statistics (con’t)

33

• Marital Status Frequency %Never Married 792 53%Married 135 9%

Separated 144 10%Divorced 132 9%Other 32 2%Unknown 266 17%

Descriptive Statistics (con’t)

34

• Welfare Status Frequency %

• Number of Children

Current Assistance 261 17%

Former Assistance 679 45%

Never Assistance 561 38%

One 958 64%Two 356 24%

Three 131 9Four + 56 3

Descriptive Statistics (con’t)

35

• Relationship to Each Child Frequency %

• Paternity Status (Child)

Mother 1399 93%

Father 39 3%

Other Relative 18 1%

Missing 45 3%

Acknowledged 250 17%

Adjudicated 890 59%

Never at Issue 290 20%

Not Established 21 1%

Missing 50 3%

Descriptive Statistics (con’t)

36

• Court Ordered Frequency %Yes 1270 85%No 177 12%Unknown 9 -Missing 45 3%

• Regression• How the DV is numerically related to the IVs

• Correlation• The relationship of the variables

• Variables converted into nominal data• 2 types of test data & 2 types of analysis

• Nonparametric v parametric

Data Analysis

37

• Regression• How the DV is numerically related to the IVs

• Correlation• The relationship of the variables

• Variables converted into nominal data• 2 types of test data & 2 types of analysis

• Nonparametric v parametric

Data Analysis

38

• Contingency coefficient and Cramer’s v tests utilized to test for association or strength of the relationships of the variables

• Strong relationship = prediction would be feasible

• Weak relationship = prediction not reliable

Test Results

39

Test Results (con’t)

40

Nonparametric Tests for AssociationIndependent Variable Contingency Coefficient Cramer’s V

Custodial parent age .716 .726

Gender .707 .707

Residential zip code .610 .544

Ethnicity .710 .712

Marital status .185 .133

Welfare status .714 .721

Number of children .708 .709

Relationship to each child .710 .713

Ages of children .711 .715

Paternity status of children .186 .134

Court order exists .153 .110

Strengths of Association

41

Strengths of AssociationTest Result Strength

> 0.7 Very Strong

0.5 – 0.7 Strong

0.3 – 0.5 Medium

0.2 – 0.3 Weak

0.1 – 0.2 Very weak

< 0.1 Extremely weak

Results

42

Strong Association• CP Age• Gender• Ethnicity• Welfare Status• Number of children• Relationship to each child• Ages of children

Hypothesis Testing

43

• There are relationships between CP date and the child support agency’s ability to collect full payment for the NCP for at least 6 consecutive months.

• Non experimental study• Confirmed – very strong associations

between seven of the 11 independent variables and the dependent variable

Next Steps

44

• Further Analysis• FFY 2009 data

• SPSS • Deeper into sub groupings for associations• When to use?

• Establishment• Enforcement