I was testing the relevance of an IV and the F value changed importantly when including sample weights (about 5 when the model is weighted and about 12 when it is not). However, considering that an IV is a LATE estimation, should we include sample weights? I have not seen any reference o paper discussing it.