Inverse Probability Weighted Selection Adjustment
Observational StudyClinical Trial
Comparing recruitment mechanisms across job vacancies generated by an intervention introduces non-experimental selection bias because newly created postings may differ systematically in observable skill, wage, and hour attributes.
Picture this
Imagine comparing exam results between students who chose to take an extra prep class and those who did not. To avoid comparing fundamentally different types of students, researchers re-weight every student's score based on their statistical likelihood of enrolling in the class, balancing the background profiles of both groups before making comparisons.
What the evidence says
Re-weighted estimations isolated that treatment vacancies experienced a 39% reduction in organic jobseeker applications (-2.670 applications, p < 0.01) and a 76% drop in firm-initiated CV database searches (-0.841 searches, p < 0.01) without confounding job selection characteristics.
- Who was studied
- N = 1,705 permanent contract vacancies posted with the Public Employment Service in France during the sanctuary period.
- How
- Inverse Probability Weighting (IPW) logistic regression using posted wage, predicted wage, working hours, experience, and qualification level variables (Hirano et al., 2003).
What to do
Apply inverse probability weighting to vacancy-level observational comparisons to isolate service efficacy from structural changes in job posting attributes.
From the source
"We thus use inverse probability weighted regressions (IPW) (Hirano et al., 2003) to try to account for the selection effects. We do not claim this fully solves the issue of potential differences in the types of vacancies posted by treatment and control firms, but we believe this makes the comparisons between treatment and control vacancies more meaningful."
Are_Active_Labor_Market_Policies_Directed_at_Firms_Effective_Evidence.pdf