Can csdid be applied to a dynamic non-compliance context? #238
Description
Hello, I have a question whether csdid can be applied to a dynamic non-compliance context? I conducted a randomized field experiment to test the effect of an advertisement. Although units (users) were randomly assigned to treatment and control group on day T, the actual exposure to the treatment (i.e., the ad) occurred on different days depending on when each unit encountered the advertisement. This leads me to wonder: should I be using staggered adoption methods to estimate the ATT? I'm uncertain because the exact timing of exposure appears to be endogenous. I am also confused with the traditional LATE estimate, because the first treatment time was different for users. How can I identify the treatment effect in this case?
Thanks for your attention and look forward to replies.