Enable estimation using -fixest-

[jorpppp]

For some of our work we would like to estimate models with large numbers of fixed effects. Currently eventstudyr uses plmunder the hood, but it would be useful to be able to use fixest too.

[jorpppp]

I am tagging my coauthor @reginalpzley here since she noticed this issue. She may be able to help us add this enhancement in the future.

[zhizhongpu]

@jmshapir Hi I might be able to start working on this this week. Please assign me and allow me to open an issue branch.

[jmshapir]

@zhizhongpu thanks! Permissions elevated. Please use it wisely.

@santiagohermo if you think you may have some time to pitch in here (or even just to help @zhizhongpu get familiar with the package), please reach out to me via e-mail and I will put you in touch.

[santiagohermo]

Thanks @jmshapir!

Absolutely, happy to discuss the package and the fixest implementation with @zhizhongpu. I'll get in touch via email.

[zhizhongpu]

per meeting w/ @santiagohermo

Given that fixest is quite different from the current kernel i.e. estimatr, the next step is to add an argument to EventStudy(), default being TRUE, where the user can specify whether to use the default kernel fixest (or instead use estimatr). This means we'll keep the current code as legacy/superseded to be deprecated. The alternative is to remove the estimatr-based code in this issue.

We also discussed another issue. Since @santiagohermo prefers that we make different modifications in separate issues, I've opened #51. I'll start working on #51 first as it's smaller and can get me more familiar with the package.

fyi @jmshapir please flag if this plan does not sound good to you.

[santiagohermo]

Thanks @zhizhongpu! A pleasure to meet you, and happy to talk again in the future if helpful :)

I'm thinking that it might be wise to keep both options for a couple of reasons:

• As you pointed out in our meeting, estimatr returns an lm object which is likely to have longer support in R
• estimatr allows to replicate STATA's standard errors

We probably don't need to decide right now. We can always just add the fixest implementation and decide whether to deprecate (or not) the estimatr one later.

[zhizhongpu]

Picking this up now that #54 is closed.

[zhizhongpu]

NTS:

1. Build fixest estimation
2. Modify downstream methods (EventStudyPlot(), summary() etc) to take in fixest objects (in addition to lm_robust)
3. Consider making output object S3-compatible (as in that we switch to an OOP system)
4. Build tests