search

bcallaway11 / did

Difference in Differences with Multiple Periods, website: https://bcallaway11.github.io/did
298 stars 95 forks source link

Big differences between using universal or varying base period #142

Closed simon-lowe closed 2 years ago

simon-lowe commented 2 years ago

Hi,

First of all, thank you for a great package! I was wondering whether you could help me figure out how there could be a big difference between the exact same specification and data using the universal or the varying base period. I have a situation where I get the following: image

In blue is post-treatment, in red pre-treatment with varying base period and in green pre-treatment with universal base period.

I apologize for the crude paint representation of the corresponding graphs, but these are from confidential admin data on a secure server. I also apologize for the lack of a code or reproducible example but since I haven't figured out how this is happening, I haven't been able to create a code which reproduces this.

Thank you for your help, Simon

bcallaway11 commented 2 years ago

Hi Simon,

You may have already seen this, but please check out this post about differences between "universal" and "varying" base periods: https://bcallaway11.github.io/posts/event-study-universal-v-varying-base-period

Your picture actually looks quite similar to the figure there involving "anticpation" effects. At any rate, that's my first thought about what's going on here.

Please feel free to follow up!

Brant

simon-lowe commented 2 years ago

Hi Brant,

Thanks a lot for your super fast answer! I hadn't seen this, I stupidly only looked at the posts on the package website and on Pedro's website. Sorry about that!

It might be related to this, I'm going to think about it and try some things out.

I had thought about it being related to anticipation, but in my head this should've meant simply a translation of every point, but I think I understand why that is wrong.

Thank you again for your help and the great package! Simon

simon-lowe commented 1 year ago

Hi Brant,

Sorry to re-open this old thread. I thought I had understood this issue but today I was looking at something new and I got the following when comparing universal vs varying: image

And that I am struggling to understand how make sense of it with simply the shift. This is a case where I'm using the not-yet treated. Any thoughts?

Thanks again for your help and for the great package! Simon