In case the model is rejected, try to find a better model by incorporating covariates that describe the differences in change between sites adequately. But what if such covariates cannot be found? See the question "which proportion of missing values in the data is allowed?"
