1
$\begingroup$

First of all, my background in statistics is a bit shaky these days due to a trauma to the brain. I am considering a study that examines at least three independent variables (e.g., creativity, locus of control, and assertiveness) on course grades (only grades were obtained not percentages . . . so this would be a ranking) in a college setting. The independent variables can be pushed into three different categories if needed (interval, ordinal, or nominal). Thus, I can use an interval score for creativity, for example. I can switch this to low versus high creativity as well (ordinal or nominal). Most would probably assume high creativity is better than low creativity (ranked), but some might think it is just nominal data.

I am thinking that ordinal regression is the best fit. Am I off base here?

$\endgroup$

1 Answer 1

3
$\begingroup$

Ordinal logistic regression is a fine choice for this situation. Your one issue is what to do with your independent variables. I would recommend against collapsing categories into just high vs. low, in general. For ordinal categories, you could represent them as nominal, via dummy codes, or represent them with a set of continuous numbers that represent your theoretical belief about the appropriate underlying values that are represented by each category. The latter is an approach advocated by Agresti (you could use his Introduction to Categorical Data Analysis as a citation, if you need one). He makes this suggestion based on the fact that your models will be approximately right unless your posited continuous values are way off.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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