Hi Claire, could you give us a sketch of your design. I don’t understand your problem, usually current softwares may accommodate with different sample or group sizes.

Claire,

I am also not sure what exactly you are looking for, but if you are trying to model a multilevel structure with outcomes at the between level, maybe this will help:

Article Expanding the methodological toolbox of HRM researchers: The...

Thank you for your replies.

The design is the following:

- 2 conditions

- Groups of 2 or 3 randomly assigned to one of the 2 conditions

- Groups do a creative task together and their "creation" is rated (1 score for the whole group)

I'm trying to see if the manipulation has an effect on group creativity.

So the manipulation was at the group level.

And the outcome, creativity, is also at the group level.

My question is: can I do an ANOVA to run this analysis?

I am bit bothered by the fact that groups of 2 or of 3 would be treated as the same.

What do you think?

Thank you,

Claire

Claire,

so, if I understand correctly, you (could) have a group-level IV expressing the treatment condition, and a group-level DV expressing creativity.

Correct?

Do you have any individual-level variables?

By the way, how many groups do you have?

This is correct. I do have covariate variables at the individual level and other DVs at the individual-level but I wanted to start by a simple analysis at the group-level as my main DV, creativity, is at the group-level.

I have 67 groups, half is one condition, half in the other. Some are groups of 2 people, some are groups of 3.

Thanks

If you were only interested in the group level, you could use ANCOVA, including covariates that capture some of the group attributes, such as number of members. Of course, that would be using only part of the data you have.

If you want to use both levels, I would recommend looking into the "latent bathtub models" described in the article I linked to in my first post.

Your sample may actually be large enough for those models, and there are tools for dealing with small samples at level 2:

Article Modeling Clustered Data with Very Few Clusters

Thank you so much for your help!

I will read these articles carefully.

you could manage the different sizes of your groups in many ways, below some :

1 - just neglect it;

2 - use the size as a covariate as already said;

3 - use the size as another factor of your design (you have only two levels, right?), your design will probably be unbalanced, which is not a serious problem for any decent software. To comment the analysis may be tricky in case of strong interaction(s) between condition/manipulation and group size - but that could precisely be of some interest.

Regards.

Thank you for your answer. I'm not sure if I can put group size as covariate as it is a dichotomous variable. But as another factor definitely. Thank you!