4
$\begingroup$

I have a stratified random sample based on sampling frame formed from our CRM systems data.

Now when I look responses in different strata they seem to differ. Some stratas have much higher response rate.

My survey variables are mostly preference related, there are few single value answers and many multi value answers.

Should I do some calibration based on weighted inverses of inclusion probabilities so that answers in stratas where response rate is lower will get higher weight when reporting results?

Improve this question
$\endgroup$
3
  • $\begingroup$ Non-response, the ways to diagnoze and overcome it have been THE main topic in public opinion research in the past 20 years or so. The most cited article has 850+ citations at this point, there is a number of books on it, etc. I believe Lohr's book must have a few sections on this. $\endgroup$
    – StasK
    Nov 26, 2013 at 14:08
  • $\begingroup$ Thanks StasK for links. I have consulted Cochran's classic and Särndal et al, but did not find exact solutions.. $\endgroup$
    – Analyst
    Nov 26, 2013 at 14:22
  • $\begingroup$ I think both are too mathematical. Unfortunately, there is really no middle ground between say Cochran (or even Lohr), on one hand, and the "softer" methodology on survey methodology. You can look at what other people are doing in the grey literature of method reports -- if you want a serious smack on your head with tons of details, you can read the one from California Health Interview Survey. $\endgroup$
    – StasK
    Nov 26, 2013 at 14:28

1 Answer 1

Reset to default
2
$\begingroup$

The most commonly used non-response adjustment method is that of weighting classes. In this method, you split the population/the sample into non-overlapping groups that (you think) have similar response propensities. These groups can, and most of the time do, cut across strata. Then you adjust the weights within each of these groups so that the sum of the adjusted weights over the responding units is equal to the sum of weights of the sampled units. All the weights here are on the scale of the population total, not the sample size as the market research community likes to have them (but which are harder to interpret).

Improve this answer
$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.