repeated measure ttest/exact/chi2 cluster

[Christine Alamaa]

Dear Forum!


I have experimental data in which the subjects perform an incentivised guess 3 times. The measure I would like to test is the guess- the true so it is discrete ranging from negative to positive. I would like to check for gender differences and would like to use all the three rounds "pooled". However that of course cause other problems due to not clustering at "subject" level. Is there a way to do this?

Additionally I have the same question regarding a binary variable and wonder If there is a way to correct for autocorrelation with panels in an exact fisher or chi 2 test?

[George Ford]

any t-test can be done by regression, and you can cluster on subject.

Is guess-true discrete or continuous?

[Christine Alamaa]

George Ford - The variable that I would like to test is discrete. My second part of the question relates to binary (indicators). Would you mind to expand a bit on your very prompt and nice answer? Thanks in advance.

I think I should have used a permutation test instead, but if I am going down the xtreg(?) road or reg (cluster id), that is not the way forward, right?

[George Ford]

post an example of the type of data you are working with using dataex. It can be made up, but should be the same sort of thing.

This is what I meant about t-test by regression, in a very simple case.

sysuse auto, clear

ttest mpg, by(foreign)

reg mpg foreign

[Joseph Coveney]

I think I should have used a permutation test instead . . .

What is your sample size? Do you have balanced data, that is, do all participants have complete data for the three rounds?