Direct implementation of unbalanced panel #90
Description
Currently, with unbalanced panel, we play a trick and run it through the repeated cross sections code and then fix the standard errors at the end. But it would be nice to have a specific implementation for unbalanced panel:
- This would likely be much faster
- There are issues related to time-varying covariates in this case, see: use with unbalanced panels #76
- Perhaps the sorts of underlying missing-at-random type assumptions that rationalize unbalanced panel could be weakened if we provide a direct implementation.
This one is probably a lot of work though!