🚀 go-pugleaf

I need some  help on the following problem:

I have a 10% non-response rate on a census of about 10,000
subjects. I do have some background information on the non-
respondents. Using this information in a linear model with
dependent variable=Responded/Did-Not-Respond (using PROC CATMOD),
I found that 3 variables have a significant effect.
What to do next?

1 - Forget about the non respondents in spite of the fact
that they seem to be a different bunch;

2 - Use the 3 background variables to predict the responses
of the non-respondents (the variables I try to predict are
ordinal);

3 - If 2), how to do that.  I tried PROC NEIGHBOR (in the old
SAS v.5) but I do not get satisfying results (I wonder why
PROC NEIGHBOR has been removed...?). Next, I tried the
following: I fitted a linear model (using PROC CATMOD), with
dependent variable=ordinal variable 'to predict', and independent
variables=3 background vars, for the respondent group. The fit
is very good with close predicted cell frequencies. Given the
good fit, I thought I could use the same response distribution
for suitable subsamples of non-respondents based on the 3 dependent
variables. Assuming the explation is clear, is this OK?

I am personally suspicious of 'predicted' responses for
non-respondents, but I am also curious about potential
methodologies on the subject.

A heartfull 'thank you' (in advance) to the...respondents.

Nino