Mixed effect model with binary response variable

Hello,

I have a binary dependent variable(correct and incorrect decision) and a binary independent variable(correct and incorrect memory). Each of my 20 participants had to answer a memory question 240 times (independent variable) and a had to take a decision 240 times (dependent variable). My goal is to explain the decision with memory correctness, so that in the end I have one beta value for each participant.

I am struggling with the specification of my model, it should be a logistic mixed effects model, right? Logistic because the DV is binary and mixed effects because of the repeated measures. I tried using fitlme and random slope and intercept(DV is dependent variable, IV independent variable):

    lme = fitlme(tbl, 'DV ~ IV + (IV|subject)')

Is this how I should do it?

Thank you very much in advance! Robert

Answer 1

[Jeff Miller]

It sounds like you have a 2x2 table for each participant with a total of 240 observations across the four cells for that participant. If so, the beta value is probably not your best measure of association between memory correctness and decision. Google "measures of association for contingency tables" and look at some simpler-to-calculate options. For example, it will certainly be easier to calculate phi coefficient for each participant than a beta value, and the phi coefficient might also be easier to interpret.

[Robert Eisenbecker]

Thank you very much for the answer, I followed your advice and extracted the phi values. But still, if I were to run a mixed effect model, would the specification that I stated in the question be reasonable?

[Jeff Miller]

Sorry, but I am not confident enough about fitlme to answer this question.