We show that the usual inference methods used in Di fferences-in-Di fferences (DID) might not perform well with few treated groups and heteroskedastic errors. One important example is when there is variation in the number ...