repeated measures & dichotomous dependent variable

bandini

New Member
I am currently working with a dataset from a study comparing two versions of an alcohol screening. One screening was administered in a face-to-face format, the other was done on the computer. Each participant was given both versions of the screening with a 2-week time interval between assessments. Participants were randomized into one of four conditions: 1) face-to-face x2, 2) computer x2, 3) face-to-face, computer, 4) computer, face-to-face.

The purpose of the study is to determine if participant responses to questions about their patterns of alcohol use are different depending on the method of assessment: face-to-face versus computerized.

The DV is dichotomous, with participants either being flagged as at risk for alcohol abuse or not.

I can handle the between-subjects portion of the experiment, that is, looking at T1 and T2 independently, using chi-squares to see if the face-to-face condition differs from the computer condition.

However, I am not sure how to test the question of whether or not a particular participants responses change with method of assessment. The only "solution" I have come up with so far is to recode the two alcohol time scores into a single variable with 4 levels instead of two: 1)no/no, 2)yes/no, 3)no/yes, 4)yes/yes, were each level represents the participants alcohol assessment score on T1 and T2 respectively. Again, I used a chi-square to analyze the results. However, I don't like losing the repeated measures component of the study.

Does anyone have alternative strategies for dealing with repeated measures and dichotomous dependent variables? I have used logistic regression in the past, but not with a within-subjects components.

Savvy

New Member
What you have got here is 2 experiments with your groups; one repeated measures (controlling for order effects), and one between groups design.

I think here you need to think more simply, and consider a between groups t-test or a repeated measures t-test, depending on which data you want to use.

bandini

New Member
I agree, separating the study into its between and within components is one way to handle the data. However, t-tests don't seem to be the appropriate test due to the dichotomous dependent variable.

Furthermore, I'm not entirely sure how to handle dichotomous repeated measures. For example, is it at all appropriate to take the difference score of two variables that are dichotomous?

Savvy

New Member
So, what you are essentially measuring is how well each type of screening detects risk for alcohol abuse. How are you measuring that? That is, a chi sqare could be useful, but if you have an actual score to work with, you could run a t-test and then examine mean differences. From that you could then tell if there was a significant difference between the groups, and which one was more effective. It's difficult to say as whether or not that's appropriate though without knowing how you are scoring.

Hope that makes sense.