[Clyde Schechter]

I'm not able to advise you on that. Certainly aggregating to a broader time frame would reduce the sparsity of positive outcomes and make estimation better (particular in a linear probability model). But it also has some drawbacks. If the influence of the event/intervention has only a short duration, then data aggregated over a quarter will dilute out the effect, so you may no longer be able to detect it in your data. So it really depends on what you can reasonable expect about the duration of effect of the event/intervention you are studying. You haven't said what that event is, and, frankly, even if you did, none of this is in my area and I probably still would not know what to say.

[Fanetti Mazakura]

Thank you very much, Clyde!

Your approach to graphing looks reasonable to me, and will probably be far more understandable than any of the numbers we are talking about here. Of course, what you are plotting is not coming from your model, it is tracking the actual data and is not adjusted for the covariates in your model.

egen mean0 = mean(cond(treated == 0, sold, .)), by(month)
egen mean1 = mean(cond(treated == 1, sold, .)), by(month)
gen diff = mean0 - mean1
sort month
line mean* diff month , legend(order(2 "treated" 1 "untreated" 3 "difference"))

Do you have any suggestions on how to improve the graphs? I don't know, maybe add confidence intervals or something else? Or change the approach to take into account the model - plot the residuals from the regression?

[Clyde Schechter]

So I might do something like this:

margins treated, at(month = (list of values of month that span the interesting range of values))
marginsplot
margins, dydx(treated) at(month = (list of values of month that span the interesting range of values))
marginsplot

Note that -marginsplot- accepts pretty much all options available in -graph twoway-, so you can modify the appearance of these graphs to your taste. That also means that you can put the graphs in -name()-d windows, and then -graph combine- them if you wish. Note, by the way, that -marginsplot- can only be run immediately following -margins-, so if you need to experiment to get the graph appearance to your liking, you have to re-run -margins- each time. I don't know why that is, but that is how it is.

The resulting plots will show model adjustments. I do not see the point of plotting the residuals of the regression here.

[Fanetti Mazakura]

Thank you, Clyde!

I keep asking questions and I am sorry if it's getting tedious. Perhaps I should start a new thread.

Going back to the model, more specifically the amount of trade variable that is present only if a trade occurs, I am worried that there might be selection bias. Given that I have a limited dependent variable, one thing that comes in mind is to use the Tobit model. However, I am not sure whether it would work in a Difference-in-Differences setting. But before digging more thoroughly into it I decided to ask you for your opinion.

Do you think the Tobit model is a good idea? If you do, would you try to help me with the implementation of it, and perhaps interpreting the results. Furthermore, would I be able to include fixed effects? ( I haven't really worked with the Tobit model. )

Best regards!

[Fanetti Mazakura]

Since I got a bit off-topic, I started a new thread on the Tobit Difference-in-Differences model:

https://www.statalist.org/forums/for...ferences-model

I would appreciate any advice shared - either here or in the new post.

Cheers!