analyzing data from within subject experiment design

[Anwesha Chakrabarti]

Dear users,

I have a dataset from an induced value lab experiment and I am trying to find the best way to analyse the data. The data has the following structure:

1. I use a within subjects design where each subject responds to 6 different treatments.
2. Each treatment is played for 30 rounds.
3. There are 12 sessions in total.
4. Each session has 10 participants.

So I have 10 participants playing 6 treatments for 30 rounds (each) in every session, which gives me 10*6*30*12 (= 21600) individual level observation. The participants are playing an auction game where they are contributing to sell an asset each round. My goal is to find the difference in average contribution across six different treatments. And I also need to control for the order of the treatments as the order was randomized. Can anyone suggest me what type of model I should use?

Thanks,
Anwesha

[Clyde Schechter]

This sounds like a hierarchical linear model, with participants nested in rounds and rounds nested in sessions. The treatment and treatment order are bottom-level fixed effects, and you probably need an interaction term between them, at least initially. It isn't clear from your description whether your contribution outcome is a continuous or discrete variable. In any case, you will likely be using one of the -me- models to analyze this, -mixed- or perhaps -melogit-.

[Anwesha Chakrabarti]

Thanks, Clyde. Yes I started reading about the mixed models and just learning about -xtmixed-

The contribution variable is continuous. I started with

xtmixed contribution, || session: || period: || treatment:

as the starting point and hope to build on this. I will post results as I proceed and include other explanatory variables.

Anwesha

[Clyde Schechter]

As of version 13, -xtmixed- was renamed -mixed-. The old name still works, for now, but that could change in the future.

I would not structure the model the way you have shown. There are only 6 treatments, and you are explicitly interested in their exact effects. For both of those reasons, representing it as a random effect in the model is not a good idea. Your model in #2 also refers to a new level, period, that is not mentioned in #1, and it fails to represent the round level. Based on what you described in #1, I would do this as:

mixed contribution i.treatment##i.order_of_treatments || session: || round:

I don't know where in this hierarchy the variable period sits, but you can insert it in the appropriate place.

Note that there is no comma before ||.

[Anwesha Chakrabarti]

Thanks, Clyde. "Period" variable is the same as "round", sorry for causing confusion. And I agree with your observation that I should use i.treatment.

Anwesha