The evaluation used a quasi-experimental design, employing propensity score matching to construct a comparison group of Unemployment Insurance claimants who were statistically similar to program participants in terms of demographics, age, educational attainment, previous occupation and earnings, and recent earnings losses.
Multivariate statistical models were used to estimate the net effects of the program on three main outcomes:
- Percentage of time employed,
- Annual earnings, and
- Annual receipt of unemployment benefits.
- The program is most effective when employees return to their employer of record.
- Earnings decreased for the first 6 years but increased in years 7–11.
- Participants saw a lower percentage of time employed over the first 3 follow-up years, but participants had 2.7% more time employed than the comparison group in year 5, and 8.7% more time employed in year 11.
How it was helpful to an Evaluation and Research Hub (Eval Hub) PLC member: “I tried to replicate this study for [our state], because we were having several troubles with our training programs here. And what is really interesting in this study is the combination of the methodologies used. It's like a multivariate regression model and also an experimental model, which is very, very interesting to me.”This is a PLC-recommended resource – an evaluation report, tool, or technical assistance product that a Peer Learning Cohort member identified as being particularly helpful to his or her research and evaluation efforts. See a list of all PLC member-approved resources.