What is the difference between "repeated cross-section" and "pooled cross-section"?

Pooled cross-section is defined e.g. here as "randomly sampled cross sections of individuals at different points in time. Example: Current population survey (CPS) in 1978 and 1988".

Repeated cross-section is defined e.g. here as "one type of survey design and uses data, in which the same information is asked to an independent sample at each wave". Which is the same as the previous definition for "pooled cross-section".

This is an answer to a question about "pooled cross sectional data" and the answer uses the term "repeated cross sections".

So it looks like the two terms are synonymous?

  • 1
    $\begingroup$ Subtlety: I find that 'repeated cross-section' (aka 'cross-sectional time series'), in the political science, econometric, and other literatures often includes both (1) repetitions of cross-sectional designs (sampling & measurement), and (2) repetitions of prospective designs (sampling & measurement; i.e. 'cohort designs' which capture change during the period, e.g., annual mortality rates). These last are sometimes called 'repeated cohorts' or 'serial cohorts' designs by epidemiologists who care about differentiating inference on rates from inference on status. $\endgroup$
    – Alexis
    Nov 1, 2022 at 21:26
  • $\begingroup$ robertspierre: you should "unaccept" my answer (I can't delete it while it is accepted, see the comments under it for more info). @Alexis Thanks for the comments under my answer. I still don't understand the difference Lebo & Weber make between repeated and pooled design, as apparently it's not the probability of recruitement that differ, according to what you say and the paper by Rafferty et al. (footnote 8 p. 14). $\endgroup$
    – J-J-J
    Nov 10, 2022 at 9:58
  • $\begingroup$ @JJJ all right...... $\endgroup$ Nov 10, 2022 at 10:15
  • $\begingroup$ robertspierre: sorry about that. I find things quite confusing actually, as on the surface sources seem to disagree. I put a bounty on the question to get more feedback, as I'm interested in the question too. $\endgroup$
    – J-J-J
    Nov 10, 2022 at 10:19
  • 1
    $\begingroup$ To add to the complexity, some sources sometimes use "pooled" to refer to panel data, e.g. see the first lines of reed.edu/economics/parker/312/notes/Notes10.pdf (HGL is short for "Hill, Griffiths, and Lim" and their 2012 book "Principles of Econometrics"). It doesn't really answer your question, but at least it shows that the meaning of the terms may vary depending on context. $\endgroup$
    – J-J-J
    Nov 10, 2022 at 12:03

1 Answer 1


This is an excellent question that illuminates the variations of terminology and conventions across all the disciplines that use (rely on) statistics.  The answer I share here is my experience and opinion...though I believe it is pedagogically and pragmatically sound.

To answer the question asked at the end of the original post:  Yes, the terms may be treated synonymously...however they may be used contextually to highlight two potential differences in the data collection protocol or the samples obtained.  The first difference is to distinguish if the “waves” were collected at the same time or not.  The second difference is to distinguish if there is any chance of “overlap” between the samples.  While I personally would be inclined to use “pooled” for the first distinction (the data collection protocols were pooled in that they occurred at the same time), whereas I would be inclined to use “repeated” for the second distinction (allowing for the fact that some individuals may appear in more than one of the cross-sections).  However, it appears the author of the first reference cited in the original post would use pooled for the second of my situations.

To elaborate slightly, I will reference an experimental design I unpack in some detail with my students...the posttest-only design using an "independent" pretest sample (see Shadish, Cook & Campbell, 2002, p.117).  In brief, the experimental design is two pre-existing groups where one group receives a treatment (and the other is the control).  When pretesting is not possible, this design proposes sampling different subsets from the groups for a pretest measure.  The key point is that it is not uncommon for this design to have data collected in such a manner that it is not possible to determine if any of the pretest participants also appeared in the posttest groups.  In this context, I would be more inclined to say this design is an example of a repeated cross-section design instead of a pooled cross-section design.

Happy to elaborate more if possible.


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.