I'm having a debate with colleagues over the decision of whether or not to include factors in a household GLM we know are not currently implementable (due to IT constraints). My view is that these factors should be brought into the modelling software and any interesting results noted but that they should not be ultimately included in the models since they will not be implementable. By excluding them, this will allow other implementable factors that are correlated with these factors to pick up some of the effects of these non-implementable factors.
