Reunification for Siblings in Out-of-Home Care Using a Statistical Technique for Examining Non-independent Observations

1
Reunification for Siblings in Out-of-Home Care
Using a Statistical Technique for Examining
Non-independent Observations

• Presented by
• Joseph Magruder, MSW
• Center for Social Services Research
• University of California, Berkeley
• Paper Authored by
• Daniel Webster, PhD
• Aron Shlonsky, PhD
• Terry Shaw, MSW
• M. Alan Brookhart, PhD

National Association for Welfare Research and
Statistics 45th Annual Workshop Madison,
Wisconsin August 31, 2005 The Performance
Indicators Project at CSSR is supported by the
California Department of Social Services and the
Stuart Foundation
2
Background

• Increasing use of administrative data, often
  containing entire populations
• Evidence that siblings comprise more than half of
  all children in out-of-home care
• Sampling bias posed by failing to account for
  interdependence of siblings

3
Background continued . . .

• Little past research has used sibling information
  in either multivariate or longitudinal analysis
• Shlonsky, Webster, Needell (2003)
• Guo Wells (2003)
• Special issue of Children and Youth Services
  Review (2005)
• Present study builds on this past work
• Incorporates method that accounts for
  non-independence of siblings when using a
  generalized linear model such as logistic
  regression
• Includes sibling constructs as covariates

4
Study Sample

• California Childrens Services Archive Data
  System
• Jan. 1, 2000 Dec. 31, 2000 Entry Cohort in care
  5 days or more into Child Welfare Supervised Care
  (with at least one sibling who entered care
  during that time)
• Primary placements shelter, guardian, missing
  excluded final study sample accounted for 56 of
  all first entries for the calendar year
  (N15,517)
• Children were followed from initial placement for
  12 months (or discharge from care if that
  occurred sooner)

5
Analysis

• Multivariate model of reunification versus not by
  12 months after entry to care
• Proc GENMOD (using generalized estimating
  equations--GEE) instead of Proc LOGISTIC
• Same parameterslarger Standard Errors (wider
  Confidence Intervals)
• Requires a clustering variable (e.g., family id)
• Conventional logit for comparison

6
Results

• Estimates for most factors (age, removal reason,
  placement type, and placement moves) consistent
  with past reunification studies
• Strong relationship between sibling group
  covariates and likelihood of reunification
• Sibs placed initially with at least one other
  sibling in same home had about one third greater
  odds of reunifying (OR1.30) than sibs not placed
  together
• Sibs placed initially all together in same home
  had almost twice the odds of reunifying (OR1.82)
• Surprisingly, large sibling groups (6 or more
  children) no less likely to reunify
• Estimates from conventional logit run as
  comparison demonstrate need to control for
  autocorrelation
• Some categories change from significant to highly
  significant, or non-significant to significant

7
CORRECTED (GEE)estimate Three Siblings 0.00995
odds ratio 1.01SE 0.071 95 CI (0.88,
1.16)UNCORRECTED (Logit)estimate Three
Siblings 0.00995odds ratio 1.01SE 0.047
95 CI (0.92, 1.11) Lack of correction may
over-assign statistical significance, etc.
Model Estimates Example
8
CORRECTED (GEE)estimate Six Siblings -.2006
odds ratio 0.82SE 0.214 95 CI (0.54,
1.24)UNCORRECTED (Logit)estimate Six Siblings
-.2006odds ratio 0.82SE 0.119 95 CI
(0.61, 0.96) Lack of correction may mis-assign
statistical significance, etc.
Model Estimates Example
9
Odd Ratios from GEE and Comparison Logit
Number of Siblings in Group
GEE Logit Two
1.00 1.00 Three 1.01 1.01 Four
1.06 1.06 Five 1.07 1.07 Six
0.82ns 0.82 Seven or more
0.82ns 0.82
Siblings Placed Together in First Placement
GEE Logit Not together
1.00 1.00 All placed together 1.82 1.82
Some placed together 1.30 1.30
ns Not significant Statistically
significant Highly statistically significant
10
Odd Ratios from GEE and Comparison LogitEthnic
Group
GEE Logit White
1.00 1.00 Black 0.96 0.96 Hispanic
0.81 0.81 Asian/PI 1.56
1.56 Native Amer. 0.58 0.58
Statistically significant Highly
statistically significant
11
Discussion

• Limitations of present study
• Shortcomings of administrative data
• Single entry cohort for one calendar year
• Children continue to reunify after twelve months
  
• Incorporating sibling-specific information gives
  new insights to longitudinal outcomes such as
  reunification
• When using logistic regression, issues about
  statistical significance due to lack of
  independence can be corrected using Proc Genmod
  (GEE) in SAS

12
Comments, Questions, or ReprintsPlease Contact
Daniel Webster, PhD Center for Social Services
Research School of Social Welfare 16 Haviland
Hall University of California Berkeley, CA
94720 510.290.6779 dwebster_at_berkeley.edu Paper
citation Webster, D., Shlonsky, A., Shaw, T.,
Brookhart, A. (2005). The ties that bind II
Reunification for siblings in out-of-home care
using a statistical technique for examining
non-independent observations. Children and Youth
Services Review, 27(7), 765-782.