The standard assumption about the error term " is that it is independent and identically

distributed (i.i.d.) from case to case. That is, var "=#2I. Furthermore, we also assume that the errors are normally distributed in order to carry out the usual statistical inference. We have seen that these assumptions can often be violated and we must then consider alternatives. When the errors are not i.i.d., we consider the use of generalized least squares (GLS). When the errors are independent, but not identically distributed, we can use weighted least squares (WLS), which is a special case of GLS. Sometimes, we have a good idea how large the error should be, but the residuals may be much larger than we expect. This is evidence of a lack of fit. When the errors are not normally distributed, we can use robust regression.