# Multi-level data query

We want to find the relationship between supervisor's narcissism and subordinate's job commitment. We have collected data from 600 supervisors and 1800 subordinates reporting to them (three subordinates per supervisor). When we run the regression, will it be run on n= 600 or n = 1800. In other words, for every three subordinates, the supervisor's narcissism will be same. So can we really use n = 1800?

• It will run for n=1800. Use for what? Commented Aug 21, 2019 at 10:18
• We have collected the Narcissism score from the Leader and the Job commitment from the team members. We wanted to know which of the following approaches we should take to do normal regression. Option 1 - Keep the 1800 entries of Team members as-is and bring the Team leader scores against each of the team member enter (in which case, the leader score will be repeated for more than 1 member) and regress OR Option 2 - Keep the unit of analysis at the leader level (600 entries) and average out the Team members scores and roll it up to the Leader level and do the regression. Commented Aug 21, 2019 at 10:21
• What approach? You don't need to input n in a regression problem. So what do you want to "use" n for? Commented Aug 21, 2019 at 10:23
• @user2974951 - Sorry... our responses crossed as I was editing it while you posted the above. Does this help? Commented Aug 21, 2019 at 10:27
• Given that you have multilevel / hierarchical data you should use a multilevel / hierarchical model. So a GLM with supervisors as random effects or something, which means 3 samples per supervisor. Commented Aug 21, 2019 at 10:30