PLM package is pretty decent for traditional econometric models, yet when it comes to datasets I have to deal with everyday it can be pretty inefficient. If only we had a version of within transformation that would work with multiple clustering variables… Actually, we have it in the form of felm package for R from Simen Gaure. The package transforms and estimates such models much faster than traditional methods. Unfortunately it only works for linear models. Now if only we had a version of this for conditional logit….

Example here