|
|
Hi,
I have a large survey dataset featuring responses from a number of
different geographic areas and the analysis requires that separate
weights are applied for each question (since different numbers of
respondents answer each question) to give each area an equal (not
proportional) influence on the overall estimates. This is fine for
estimates on individual questions, but I also need to produce Xtabs of
different questions based on this kind of weighting approach. Since n
for Qx*Qy is generally < n for either Qx OR Qy, I think this means that
what I need to do is identify the number of respondents in each area
answering BOTH questions to be included in any given Xtab and then
weight based on these [eg, area (A) weight for Xtab = average number per
area answering both / number answering both in area (A)]. This isn't
something I've thought much about before though so I'm wondering:
a) is this an appropriate approach to be using? If not, should I be
doing something else?
b) is there a simple way to do this based on existing individual
question weights or some other method? At the moment the most efficient
approach I can think of is to compute new variables to identify
respondents who do answer both questions, then use aggregates to
identify n of such cases within each area before aggregating again,
merging, and computing weights from this - but this seems rather
laborious...
Any advice would be greatly appreciated!
Thanks,
Chris
|
Locked
