I want to analyze the effects of a health care reform introduced in one region in 2009 on various outcomes using difference-in-difference (DID) analysis. I have access to repeated cross-sectional patient-level data, which is however only available from 2008-2012, hence, I cannot investigate the common trend assumption before the reform. Therefore, I want to match/weigh exposure and control groups before the DID analysis using propensity score matching or entropy balancing.
The exposure group consists of patients from the region with the reform (N ≈ 1000 per year) and the control group consists of patients from the other regions in the country (N ≈ 4000 per year).
My question/problem is how and when should I do the matching/weighing of patients when I have repeated cross-sections (I do not follow patients, there are new patients every year), i.e. in which dimensions (cross-sectional and longitudinal)?
I have seen a couple of alternatives, but can’t figure out which one is the best/right one:
E = exposure group; C = control group; T = whole study period; T0 = time before reform (2008); T1 = time after reform (2009-2012)
- Match E with C during T
- Match E with C during T0
- Match E with C during T1
- Match E in T1 with E in T0
- Match E with C in T1
- Match the matched E with C in T0
Are there more alternatives? What is the best way to go and why? Could it also be so that one alternative is best for propensity score matching and another alternative is best for entropy balancing?
Thanks in advance for any input!