Repeated measures and response times: Can I include a variable that is not balanced across individuals as a predictor in my model?


I’m currently working on some observational data. My outcome variable (y) is a response time which was measured repeatedly for each individual four times. Initially I was thinking to fit a multilevel model to understand how response time changed over each repeated exposure (Var1 = exposure). However, I’ve also realized that another observed (i.e. this variable was not controlled but just happened to be present for some individuals and not others due to observational nature of the experiment) binary variable (Var2 = 1 if present, 0 if not present) that seem to have a greater influence on response time than the exposure. The issue is that this variable is not equally 1 or 0 for each individual and exposure. Thus, one individual could have 1 for every exposure, another could have 1, 1, 0, 0 and a third could only have 0:s.

If I include Var2 as a predictor in my model together with Var1 I was thinking that I could understand something about the influence of both predictors on response time. However, will I ever be able to tell apart the effect of Var2 and the effect of the individuals that have more 1:s being just slower individuals in general?

Any ideas of how to deal with this type of data? If needed I can also provide a more detailed example with a toy dataset and code.

Thanks in advance! :innocent: