4
$\begingroup$

Good morning all,

I'm working on a project related to the occupations that university graduates are likely to go into. I have a large data set (N in the hundreds of thousands) where the unit of analysis is an individual person. For each person, I have two categorical variables - a code representing the field of their university degree and a code representing their occupation. I'm looking for a statistically valid way to find out what fields of study and occupations "go together." In other words, what courses of study prepare people for which jobs?

So far, I've considered doing this with simple descriptive statistics... pull, say, the top 10 occupations for every subject area while ruling out occupations like cashiers, fast food workers, etc. But it would be great if there was some sort of more rigorous test that could be used for this. Perhaps one regression per occupation with tons of dummy variables representing each subject area? If a subject area dummy variable is statistically significant, then that subject area is connected with that occupation?

I'm open to any and all suggestions.

$\endgroup$
3
$\begingroup$

Sounds like multinomial logistic regression is the tool to use.

The dependent variable would be "field of work" and the independent variable would be "field of study". With such a large N you can be fairly specific in defining the levels of the variables, but you should probably start with frequency counts of each and then a crosstabulation of the two, not for statistical testing but to see what's going on and whether you want to combine some categories of either variable.

$\endgroup$
  • $\begingroup$ Great advice. Sounds like it is exactly what I need. $\endgroup$ – John Chrysostom May 30 '14 at 12:22
  • $\begingroup$ Hm. SAS is throwing out of memory errors. That will be fun to navigate... $\endgroup$ – John Chrysostom May 30 '14 at 14:47
  • $\begingroup$ Oh boy. That probably means you have to collapse categories. $\endgroup$ – Peter Flom - Reinstate Monica May 30 '14 at 20:51
  • $\begingroup$ You could also look into multiple correspondence analysis (and extensions) $\endgroup$ – kjetil b halvorsen Mar 4 '18 at 11:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.