My understanding is that if E(u)=0 is violated, then the distribution of the parameter is not normal, therefore t-statistic is invalid. But why is that the distribution of the parameter would not be normal under such violation? Thanks for helping out!