The problem is that if a record is missing observations, then comparing it to those that are not is like comparing apples and oranges. For example, if you are predicting exclusive and have trained your model and found that both childage and primi are very influential in predicting exclusive. Then if you have one observation that has records for both childage and primi and one record that is missing a value for primi, then it would be inconsistent to use the same model to predict both records. This has to do with the fact that when you are calculating model coefficients, you are finding the impact of any independent variable on the dependent variable while holding all other independent variables constant. If you don't have those other variables (due to missing values), then really what you have is a completely different model e.g. y=x_1+x_2 vs. y=x_1. So the 'n' or number of observations for your model should be the same for all variables used in your model. I am not sure what you have seen in other publications, perhaps they were running multiple experiments with different cohorts and then comparing them? By default, SAS will drop any observations with missing values that are used in the model for this reason. I hope that helps.
Best,
Daniel
... View more