I am trying to decide on an approach to estimate design effect for a multi stage cluster survey. The clusters were selected with probability proportional to size sampling WITH replacement. The primary sampling units (districts) are large enough that some were sampled multiple times. When defining the PSU variable should the unique districts within the sample be used, or a cluster variable representing the true number of clusters sampled?

For example: A district gets sampled twice, should the PSU variable described, be one or two individual clusters?



Update: I'm sorry to say that my 2014 answer was wrong. In Stata or elsewhere, specify the original primary sampling unit as the PSU.

2018-06-04 I Now believe that my original answer is correct.

The relevant sample size for estimating the design effect (and other precision-related measures) is the number of draws, not the number of unique PSUs selected. If a PSU was drawn twice, it counts as two draws. Note that it is the total number of draws that is fixed by a with-replacement design.

Terminology can get ambiguous when sampling is with replacement. I think it safest to reserve the terms "primary sampling unit", "PSU", "cluster", "first-stage unit" to denote the fixed geographic unit that is selected at the first stage. What distinguishes the data from different draws of the same PSU? Each contains a different set of second stage sampling units (SSUs). Moreover, each second stage sample must be drawn independently of the other second stage samples, even those from the same PSU.

So when a sampling analysis program like Stata asks you to specify a "psu", substitute the name of the variable that designates the draw.

| cite | improve this answer | |
  • $\begingroup$ Thanks to Stephen Hornbeck who noticed that I had answered this question differently on Statalist. $\endgroup$ – Steve Samuels May 24 '18 at 15:07
  • $\begingroup$ Steve, I found this source which indicates that your initial answer is correct...“In PPS sampling with replacement, a given cluster (i.e., hospital, in this instance) can be sampled more than once, and each drawing of a cluster is considered a primary sampling unit.” Levy, Lemeshow Sampling of Populations p. 346 $\endgroup$ – david rae Jun 1 '18 at 19:45
  • $\begingroup$ I think that you are right, David. The problem I was concerned about in Stata was how to designate the PSU. But I realize that anybody who sampled with replacement would provide the draw numbers as well as the actual cluster ID. $\endgroup$ – Steve Samuels Jun 4 '18 at 17:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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