2
$\begingroup$

I have data from an experiment in which participants played a repeated game 50 times. That is, I have 50 data points per participant. I want to run a logistic regression and I have of course to take into account that I have data points that are correlated, but I don't know if I should use repeated measures or if I should work with clustered sample. I have found some very clear articles on why clusters are better than other techniques (Williams, 2000; Petersen, 2009), but I have found nothing on repeated measures or on the comparison between the two. I would appreciate it if you could help me on this one!

Thanks!

$\endgroup$
4
  • $\begingroup$ In repeated measures, between-subject factor and RM factor are crossed: every individual comes through all the RM levels. In clustered sample, between-subject factor is nested within cluster factor: clusters are comprised of different individuals. $\endgroup$
    – ttnphns
    Jul 23 '14 at 6:33
  • $\begingroup$ Thank you for your answer and for editing my question. I realized I misused the term "cluster". $\endgroup$
    – Deborah
    Jul 23 '14 at 8:50
  • $\begingroup$ @ttnphns: I wonder if you could add a reference for your last sentence? $\endgroup$
    – Michael M
    Jul 20 '15 at 19:40
  • $\begingroup$ en.wikipedia.org/wiki/Multilevel_model; en.wikipedia.org/wiki/Cluster_sampling. $\endgroup$
    – ttnphns
    Jul 20 '15 at 21:53
1
$\begingroup$

Expanding ttnphns comment by examples, repeated measures are if you observe a measurement at certain specified time points (fixed effects) on each (independend) individual. Instead of certain time points you can also have measurements at certain places in space: The classical split-plot design from agriculture is a repeated measures design, too. You can formulate hypotheses comparing the time points.

Clustered data are if you have e.g. young mice from different litters. There, the litter is a (random!) cluster factor. The nesting structure of the random factors says that you usually don't design a clustered data experiment in order to test something like the effect of being the 5th mouse against the effect of being the 2nd mouse.

Note that sometimes repeated measures are understood as having equal correlations between the time points. This however is not necessary, as statistics like the one from Geisser and Greenhouse (1958) can deal with unknown covariance matrix structures between the time points (and even different covariance matrices between samples).

So in your case, if all the players had the same sequence of games, you'll rather have a repeated measures design.

$\endgroup$
1
  • $\begingroup$ Thanks a lot! I think I was confused between clustered sample and using a robust variance (in stata vce cluster). I now understand why ttnphns edited my question. I realized that having repeated measures influences my analysis at two stages: 1) when choosing the correct estimation procedure (GEE vs. OLS) 2) when computing the covariance matrix. $\endgroup$
    – Deborah
    Jul 23 '14 at 8:49

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.