Read on Twitter
Are you a health or education policy researcher? Health economist? Clinical epidemiologist? Do you analyze two-group/two-time-period (quasi-experimental) studies? If so, then this methods thread is for you! 1/
Inspired by confusing differences in how various fields describe study designs, @laura_tastic and I dug deep into two methods: comparative interrupted time series (CITS) to difference-in-differences (DID). Working paper here: https://arxiv.org/abs/2006.11346 2/
tl;dr – As Death Cab for Cutie says, “There are different names for the same thing.” At their most flexible, DID & CITS are the same thing! But researchers often constrain (& thus differentiate) them w/ more assumptions (parallel trends in DID, linearity in CITS). 3/
Re-analysis of 3 published papers to understand the tradeoffs of CITS versus DID. One example violates parallel trends, the second violates linearity, and the third shows subtler violations of both. What to do? Our answer: use the most flexible model specification. 4/
Still want to use the constrained versions of CITS or DID? Great. Make sure you describe and justify your assumed data-generating mechanism and related causal assumptions. 5/
We hope applied papers will give more detail on counterfactual assumptions & model specifications. Then readers can assess the plausibility of your conclusions, whether you speak econ, epi, stats, or something else entirely. 6/
PS – A special thanks to @hpdslab members for their feedback and @thedavidpowell well and @PaculaRosalie for data and code for a re-analysis. Data and code for reanalysis #2 can be found here: https://doi.org/10.7910/DVN/47TMEO
