Quasi-randomization by survey date for policy analysis

In annual surveys, the interview dates are scattered over several months. When a policy takes place during the survey period, quasi-randomized data may be obtained, if individuals interviewed before and after the policy timing are similar. The policy effect can be found with a before-after difference, which is 'Quasi-Randomization by Survey date (QRS)'. QRS may be viewed as an regression discontinuity (RD) with time as the running variable. When seasonality is present, the RD-style estimator fails, but we develop a difference-in-differences style estimator, which relies on a weaker assumption analogous to parallel trends that controls for sesonality. We provide an empirical example using Korea Labor and Income Panel Study data for a weekly work-hour reduction law in 2004 from 44 to 40 h. We find that the law effect is about 2 h reduction, not 4 h as the law stipulates.