Which regression is useable for an ordinal dependent and multiple discrete/ordinal/binary independent variables?

Question

For a paper of mine I am trying to figure out if there is a correlation between success as a musician and multiple other factors. These are my variables:

1. Dependent variable: Success from 2016 based on RIAA certifications. (ordinal, key: 0 = No cert;
   1 = Gold; 2 = Platinum; 3 = Multiplatinum)
2. Independent variables:
   • network degree (discrete, non-gaussian)
   • male (binary)
   • Success from 2013-2015 (ordinal, key: same as above)
   • 5 dummies for different regions (binary, made using One Hot Encoding)

So my datatypes for the regression look like this: $ordinal = discrete + binary + ordinal$ .

I am unsure which type of regression to use. An ordered probit looks like the best option, although I am unsure if it is the best possible choice (my statistics are rusty). Are the any limitations that I need to be aware of in regards to the different data types and what are other useable options that could help me?

Answer 1

Welcome to the site Ranga. To be honest, with an ordinal response you can actually use an ANOVA, as it is quite robust to deviations in terms of the response variable being non gaussian. ANOVA has the advantage that it is quite easy to interpret in terms of effects, although it is not greater for prediction, especially so if your data are not normally distributed. An alternative would be to use a regression tree or some variation there of (i.e. boosted regression tree) which are much more robust than ANOVA's, if a little more opaque in terms of interpretability....

Let me know if you need any clarification.