Hi folks

I posted this in the SPSS section, but its actually more a question about which test to use so I think it belongs here.

I'm working with a series of survey questions on job openings that share the same response scale. This is similar to a likert-style series of questions, except the scale for responses is a set of company names the respondent can pick from to match preferred HR recruiter companies with specific sample job openings. So for an IT opening, the respondent might indicate a preference for company 'A' but for an Admin job they might prefer company 'B' and so on. Recruiter companies are coded numerically. They can only select one company for each sample opening.

The respondent also rates each recruiter in terms of performance. I'm interested to see whether that performance rating (ordinal 1-10) impacts whether a particular recruiter gets selected more often across all sample job openings. So - in a nutshell, if the recruiter is rated as very good, do they get chosen more often regardless of the type of job opening?

My thought was to look for 'convergence' towards a particular numerical recruiter code when performance rating is high. Even though these are categorical data a means-based comparison could still work I think. So for instance, the code for recruiter B is '5.' My hypothesis is that respondents who rate recruiter B's performance at 10 (highest level) will pick that recruiter more often, and thus the mean across all of their sample job opening questions will be very close to '5' with a smaller SD. I would expect that mean to deviate towards other responses and the SD to increase as performance rating goes down.

Selecting for a particular recruiter, I can see this relationship graphically with a scatter plot. A simple comparison of performance score to the mode for each respondent also suggests there is a relationship here. Higher performance = higher number of times the recruiter is selected.

But I'm wondering how to test that relationship statistically and all at once for all recruiters? I'm thinking some form of cluster analysis? Or possibly discriminant analysis across job categories? Someone who works a lot with likert scales and common responses across several questions will probably have an answer.