different underlying distributions

[Lilia Wasserka-Zhurakhovska]

Hi!

I have data from a game in which subjects first roll a die and afterwards make a report. I have two treatments. However, the true realized die rolls are not identical in the two treatments. How can I correct for that? Maybe by bootstrapping? Or any other simulation? Has anyone an idea?

Thx,
Lilia

[Carlo Lazzaro]

Lilia:
welcome to this forum.
Please read and act on the FAQ on how to post more effectively. Thanks.

[George Ford]

So you begin with 3 randomly assigned groups: control, treat1, treat2

Each person rolls a die (a random outcome) then makes a report.

You want to know whether something about the report differences by treatment.

You've discovered that the mean of the die outcome is not equal across the groups.
Did you conduct a hypothesis test of equality?
What's the sample size?

You'd think with 2 random stages, the mean die value would be nearly the same.

If they are very different in the means and the sample size is large, something weird is going on, or else this process isn't as random as I've laid out.

Check the data to make sure there's no coding errors (impossible values for die).

There are plausible solutions when the mean of X differs among control/treated (matching), but I think you need to get to the root of this difference. I'd reject a paper that had a big difference in mean die toss in an apparent random process with large samples, since the result suggests a problem with the experiment.