Hello everyone,

I have a question about the ordered probit, ordered probit random effect, ordered logit fixed and random effects. I am wondering which one of the regressions is the best for me to use. My dependent variable is sovereign credit ratings which range from 1-22 so they are of ordinal nature. My dataset consists of around 80 countries with 20 year of data, so panel data. My independent variable is cross border Bank flows, in addition to this i am using 7 control variables (mostly based on country level gross domestic product). The goal of the study is to see what effect the Bank claims have on Sovereign Credit ratings.
To see what the effect is i have computed the:
  • Ordered probit
  • Ordered logit
  • Ordered probit, random effects
  • Ordered logit, random effects
  • And the fixed effects ordered logit (with the feologit command)
All of these regression give me in one way or another significant effects, what im wondering is what is the best model to use and how do i know it is the best to use, ive read that there is not much difference between the outcomes of the ordered logit and probit so they are in a way interchangeable. What i really dont understand is how to know whether its better to use a regular ordered probit/logit or one with fixed or random effects. To me fixed effects control for differences across countries that do not differ over time. What random effects do exactly is still a bit foreign to me. For random effects i have read this: "If you have reason to believe that differences across entities have some influence on your dependent variable then you should use random effects" from: https://www.princeton.edu/~otorres/Panel101.pdf slide 25. So for my case this would imply if i think the difference across entities: i.e countries, have some effect on sovereign credit ratins i should use a random efffect version of the ordered model?

Thanks you for taking the time to read this!

Kind regards,

Max Verheijen