This is harder than one might imagine. One (mostly) can not just plot
the logits, because much of the data is at the extremes where p=0 or
. Often the best we can do is the plots that we did
earlier, perhaps adding some sort of smooth line to the data. (Next
example, we will try that). We will be able to do better once we have
a model and can consider residuals---but be careful which residuals we
think about.