did_multiplegt_dyn with binary outcome

[marie carmel]

I was thinking of the did_multiplegt_dyn and had question pertains to the outcome variable.

Is it possible to use both a binary outcome and a scaled outcome with this command? If so, how would the Average Treatment Effect (ATE) be interpreted in the case of a binary outcome? Specifically, would it follow the interpretation of a Linear Probability Model (LPM)?


Marie

[Jeff Wooldridge]

Do you have a staggered intervention with exit from treatment? If there/s no exit, why not just use xthdidregress and choose the twfe, ra, or aipw option? These all use linear models and the estimated are interpreted as effects on probabilities -- just like an LPM.

Or, you can use the logit method that I proposed in my 2023 Econometrics Journal paper. The Stata files are available on my shared Dropbox, pinned on Bluesky and X. This is directly comparable to xthdidregress twfe but it uses logit instead of linear.

[marie carmel]

Dear Professor Wooldridge,

Thank you for your response. I believe it would be helpful to provide more details about my study.

I am examining the effect of successful terrorist attacks on support for democracy, which is measured as a binary variable at the individual level. My dataset is a repeated cross-section, and the treatment variable—terrorist attacks—is a binary indicator at the village-year level. Importantly, my study includes units that exit from treatment.
I hope this additional context clarifies my research setup. Please let me know if further details are needed, or if you have any recommendations regarding suitable econometric approaches for this type of data structure.

Best regards,
Marie

[marie carmel]

To be more specific, Units can exist and re-enter treatment in my case. One village can face a successful terrorist attack in 2012, but not in 2013, and face a a successful terrorist in 2014.