Repeated measures, clchi2 right order of variables?

[Andreas Orland]

Dear all,

I conducted an experiment where subjects repeatedly set prices. My price variable is quasi-continuous (with 37 possible values). Now I want to compare the distributions
in the two conducted treatments. (Just by eyeballing, the two CDFs look quite different.)

Using Stata 16.1, I thought about using the user-written command clchi2 (part of the cltest package):

If I run

clchi2 price treatment, cluster(id)

I get an output that makes sense. However, price is not my dichotomous event variable. That is treatment (which takes either 0 or 1).
When I change the order of variables

clchi2 treatment price, cluster(id)

I receive "too many values".

Do I overthink this (and should go ahead with the first code snippet)? Or is there another command for clustered chi2 tests?

Best,
Andreas

[Leonardo Guizzetti]

-cltest- is for cluster randomized clinical trials using cluster-adjusted versions of the t-test and Chi-squared test. I don't see an obvious reason from your post as to why this should apply.

[Andreas Orland]

Dear Leonardo,

I am sorry for having missed some crucial info.

I randomly assign clusters (of two participants) to treatments. Thus, two participants in either one of the treatments repeatedly set prices.
My idea is to compare the distributions of prices between the treatments. It seems there is no Kolmogorov-Smirnov test for repeated measures
(also, my variable of interest is only quasi-continuous). It seemed to me that -clchi2- is the next best step.

Thanks for your time.

Best,
Andreas

[Leonardo Guizzetti]

I'm not sure how to compare distributions in this case. Is this really necessary though? I'm not an economist, do someone with more expertise could jump in. I wonder if it is sufficient to show a difference in means between treatments?

[Andreas Orland]

I did compare means between treatments. They do not differ significantly. However, theory predicts different prices in different regions of the price range.
I will test if my participants actually did set prices differently in those regions. But I wanted to start using a comparison of the whole distribution.

So -clchi2- tests for a difference in relationships between the k groups in 2xk tables where the outcome variable is binary.
It is not possible/advisable to use this test to compare two groups in kx2 tables where the outcome can take k different values?

Thank you for your help!

Best,
Andreas