Dominik Liebl @dliebl.com · Jan 21 1 / 2
Event-study plots in DiD are incredibly persuasive—but, honestly, not always honest. Why? If parallel trends or no anticipation fail, DiD estimates are biased. Testing against a zero-effect null then becomes misleading. 🚨New 📄: arxiv.org/abs/2512.06804 #CausalInference #EconSky #StatsSky #rstats
9 likes 1 replies
?
Replies
Honest causal inference accounts for this bias by testing effects against an honest reference band (zero ± bias). Our contribution: We make event-study plots honest using simultaneous confidence bands (SCBs). 🤓 We use infimum- and supremum-based SCB!