Random effects binary logistic regression model with two(?) level 2 variables

I have a dataset of interviews and want to determine whether interviewee characteristics predict how they record interview data collected from respondents. This requires a multilevel analysis because interviews are nested in interviewers - every interviewer (level 2) conducts many interviews (level 1).

However, I also have respondents who did more than one interview, and some of these respondents have been interviewed multiple times by the same interviewer.

How do I model my regression?

If every interviewee did only one interview, interviewer ID would be the level 2 variable and all other variables would be level 1. However, since my observations are also not independent because some belong to the same respondent, this approach still violates the independence assumption.


Less is more. Stay pure. Stay poor.
I am not sure if you may have power issues given how many people are nested in multiple groups, etc. but there is a MLM called "cross" something for this scenario.

@Jake - what is this called again?

Correction, I am be wrong - I was thinking of cross-level interactions.