Categorical moderators with small sample size

[katarina86]

Hi everybody,
I have small sample size, around 100 observations. I created a model with 6 latent variables, and I have a total of 23 indicators. It is a simple model, no higher-order constructs. Now I want to test categorical moderators (gender and experience level) for gender I don’t have equal groups… one group is around 30, and another is around 70 respondents. For experience level I have three groups, also not equal… I assume that it is not possible to acquire meaningful results using MGA… For experience level, I think for sure that sub-groups are too small… so I wanted to ask if I can model gender (and maybe experience too) as dummy variables and use the interaction term to model the influence? Does the sub-sample sizes represent a problem when moderators are modeled as dummy variables?
Thank you in advance for your answers.

[jmbecker]

In general, the sample size will not be larger and therefore you will not have more information to estimate the model. Thus, sample size will still be an issue if you go for dummy variables.
However, these dummy moderator models are slightly more efficient than multi-group models, because they restrict moderation to one specific relationship and do not allow all parameters to vary across groups. Nevertheless, the power to detect effects will likely be small and you will have quite high uncertainty in your estimates.
You may also get problems with bootstrapping if one dummy variable category has only very few values because the random sampling process might create sub-samples that have no variation in the dummy variable because by chance none of the rare values might be picked so that all values of the dummy are either only ones or zeros (depending on which is the rare category). Such a dataset will not be estimable and you will get an error.

[katarina86]

Dear Dr. Backer,
Thank you very much the fast response and for the clarification. It is very helpful!
Best regards,
Katarina