stacked data and binary logistic regression

Question

I am conducting research where I investigate gap acceptance by pedestrians. One of the statistical tests which I would like to carry out is binary logistic regression or any similar test. I am asking the question: what factors contribute to the pedestrian's decision to reject or accept a gap in the traffic. I have both demographic variables and traffic variables (speed, type of car, etc). I have observed many pedestrians and each pedestrian rejects 1 - 10 or even more gaps... and accepts only one. As you would imagine the binary choice is REJECT or ACCEPT.

My data is in stacked format, so each gap is at a separate row. This means that while each gap has its own associated vehicle speed, traffic and other "between-gap" variables, the demographic characteristics, as well as the location for crossing, often repeat in a similar way to a repeated-measures design study where the demographic characteristics of the participants remain the same.

So, my question is: would it be appropriate to use binary logistic regression in this case? It has been done before in similar literature, although I am not sure exactly how they structured their data. Am I violating the assumptions? Should I resort to Generalised Estimated Equations (which actually wouldn't run I guess due to the amount of data, number of variables and the computer SPSS is on).

Thank you for your time!

Answer 1

If I understand your data correctly, it means that each row presents one trial, and multiple are associated with a single individual. In this case, trials will be non-independent on a per-individual basis. Therefore, you should use something to accommodate this, like a GEE as you identified. However, if you have R, you could also fit a Generalized Linear Mixed Model e.g. with glmm() or with glmer() from the lme4 package, with individual as a grouping factor or random effect. In this case, as the response is binary, you would use family = binomial()