reghdfe warning about singleton observations in clustered bootstrap

[Scarlet XUQ]

Thank you for your attention .I am estimating a firm-year panel model with firm and year fixed effects and standard errors clustered at the firm level. I also use a clustered bootstrap with 500 replications to estimate the confidence interval for an indirect effect.

In each replication, I resample firms by their panel identifier, generate a new bootstrap identifier, and run reghdfe with firm and year fixed effects. When I include keepsingletons, Stata repeatedly reports:
WARNING: Singleton observations not dropped; statistical significance is biased
My questions are:

1. Should keepsingletons be removed from both the main regressions and bootstrap regressions?
2. Should singleton observations be dropped consistently across both mediation equations?
3. Is using the newly generated bootstrap identifier in the absorbed fixed effects appropriate?
4. Does retaining singleton observations affect the validity of the bootstrap confidence interval for the indirect effect?

[Andrew Musau]

Welcome to Statalist. Please note our strong preference for the use of full real names (see FAQ Advice #6). If you overlooked this when registering, click on “Contact us” at the bottom right-hand corner of the page. Also note that reghdfe is community-contributed and is from https://github.com/sergiocorreia/reghdfe. (FAQ Advice #12).

The potential bias from retaining singleton observations is discussed in https://scorreia.com/research/singletons.pdf. Official Stata panel-data commands generally retain singleton observations by default, so there is no universal practice regarding how they should be handled. A conservative approach in this case would be to exclude them, as reghdfe does by default.