Non-parametric regression with categorical dependent variable

[Eleonora Facibeni]

Hi everyone, I have a question about estimating a non-parametric regression using a categorical dependent variable (10-point Likert scale). I should also consider survey weights and clusters. I'm not sure whether I can use npregress, lpoly, or lowess commands in this context. If it is possible, please can you suggest to me a well-suited command to use? Thank you for your help!

[Nick Cox]

It's documented or in two cases not documented that none of those commands supports svy weights. A bigger question is what you consider to be nonparametric regression for that kind of data, because for each of those commands, nonparametric means a smooth fitted response but one that takes the response literally, which many researchers consider unacceptable for Likert scales.

You may be nearer what is approved with ologit or a sibling or a cousin.

[Jeff Wooldridge]

How many covariates are there? What is the compelling reason to use nonparametric methods? I agree with Nick that you should try ologit or oprobit and include the controls flexibly — squares of continuous covariates and interactions of any covariates. If you have many, you’ll use up degrees of freedom. But nonparametric methods also do that.

[Eleonora Facibeni]

Thanks Nick and Jeff for your help!