Propensity Score Matching (PSM) in a Longitudinal Study of at-Risk Youth

Background and Purpose: The purpose of this longitudinal, quasi-experimental study was to examine whether juvenile offending is associated with later adult offending among a sample of children enrolled in a state public school system. The study answered the question: What is the predictive ability of juvenile justice involvement among youth who progressed from juvenile to later adult criminality?

Methods: Utilizing secondary administrative data from a department of education, department of juvenile justice, and department of corrections from a state in the Deep South, researchers identified a sample cohort of 7^th to 12^th grade public school children (N = 407,800) born between 1980 and 1989. First, binary logistic regression (LR) was conducted among students to examine if previous juvenile offending was associated with further adult offending, controlling for demographic and school-level variables. Subsequently, propensity score matching (PSM), a more robust approach, was used to confirm the LR model. A comparison group of students (n = 14,349) from the department of education data, matched to the study group of juvenile offenders (n = 14,349), was created using a one-to-one nearest neighbor matching procedure without replacement PSM. The matching criteria contained demographic and school-level risk factors (sex, ethnicity, economic status, four categories of discipline history, attendance, truancy status, dropout status, school transitions, highest grade completed, and English test scores).

Results: LR results indicated that the model predicting adult offending  was statistically reliable in distinguishing between students having and not having juvenile offense records, controlling for  demographic and school-level risk factors (- Log Likelihood =-67042.951; X² = 33296.33, df = 16, p<.001). Odds ratios showed that students with previous juvenile records were over 3 times as likely as students without juvenile records to have later adult correction records, at Exp(B) = 3.15. The odds ratios were 5.29, 21.09, and 2.21 for male, expulsion, and dropout, respectively. There were over twice as many students who had juvenile offending and then progressed to adult offending (n = 4,222) compared with those students who did not have juvenile offending (n = 2,020). The LR results were confirmed as expected.

Conclusions and Implications: PSM is used when a comparison group is either impossible or unethical to identify, which is common in social science research. Among social work research, PSM, a matching method and quasi-experimental design, is a valuable tool that may be utilized with at-risk youth populations. PSM was used in the current study as a confirmatory analysis following binary regression to support the findings. The current study is unique because it utilized school-level variables, dropout and attendance, and expands the scope of juvenile to adult criminality study. The study adds to the knowledge base in understanding patterns of offending among youth.