Colleagues,
I am working with a national telephone sample which seems to
overrepresent certain segments of the population. Education,
liberalism, political interest are all coming in too high; not
surprising given what we know about the response process. Final
sample size is approximately 1200.
My collaborator wants to weight the sample by education. Of course
this assumes that we know that education is the biasing variable, an
assumption I'd rather not have to make. Nonetheless, if we do use an
education weight, the population figure to use comes from the 2000
Census. I took this off their website, and it mentions that it is
derived from a monthly sample (taken during 2000). Is this correct?
Is Education not on the long form?
Second question: I'd rather use a inverse probability of selection
weight, since this takes the substantive decision out of my hands, and
simply over/underweights the data according to the actual and
predicted probability of selection. I am not sure I have the
information in order to calculate this weight, however. From the
survey administrator I have this, and am not sure what to do with it:
GENESYS RDD is a single stage epsem methodology and does not
employ M-W
clustering. The RDD method simply uses working 100 series banks
as the
Measure of Size. The banks are divided up into even sized
intervals (based
on number of pieces required) and each interval is randomly
sampled.
Seems like I'd need to know the proportion of the population in each
"interval" and then know the proportion of the sample actually in each
interval. Does that sound right? I also have the area code for each
respondent -- is there a way to aggregate these to some higher level,
knowing the proportion of the population in the higher level of
aggregation?
Thanks for any help and advice,
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
