Dummy dependent: ANOVA v logit

[Pratap Pundir]

This is a statistics question rather than a Stata question, but would appreciate any advice.

After running an between-subject experiment, in which I had participants randomly assigned to one of four groups, I am trying to analyse my data.
The outcome was a binary response from the participants. Hence my dependent variable is a dummy variable.

Usually in experiments, with a small number of observations, ANOVA works well as the predictor variable (group to which the participant was assigned) is categorical.

Since the dependent variable is a dummy/binary variable, can ANOVA (or a variation of it) be used? Can it be suitable? Or should I simply run a logit with my eyes closed?

[Jared Greathouse]

I know nothing about ANOVA, MANCOVA, or any of that.

Use logit analysis.

[Joseph Coveney]

. . . a small number of observations . . . should I simply run a logit with my eyes closed?

I don't recommend fitting any model with your eyes closed. In your case, what does "a small number of observations" mean?

Logistic regression is considered a large-sample estimation method. If you're interested in null hypothesis significance tests, then you might want to verify that the operating characteristics of the test method are acceptable to you for your intended use. In this case, you'd use simulation under the null hypothesis with your obtained sample sizes in order to check that the so-called test size is not too far off from nominal.

And if you're just interested in significance testing, then there are other methods for use with binomial outcomes and multiple categories of predictor that might have better operating characteristics than logistic regression. But again you'd need to verify that, either through the literature or simulation.

[Pratap Pundir]

Joseph Coveney
To answer your question, I have a total of 240 participants, randomly assigned to 4 groups. And group is the main predictor. So each predictor has approximately 60 observations.