Covariates specific to one person

[michalovadek]

Thanks for making this great package. I am using it in my PhD thanks to its superior performance over methods like NOMINATE. I think the package was taken off CRAN for some reason but I hope you continue to develop it, because it's a fantastic contribution to the R ecosystem.

My question concerns the use of covariates in the estimation. In your working paper you show an example when covariates affect all legislators in the Congress. I was wondering whether you would have any advice for a situation where I need to include covariates just for one of the "persons" in the data (binary response). A set of independent variables - some binary, some continuous, including interactions - affects the "votes" of only this one person, while the rest are unaffected by any variables.

Would you recommend to add a dummy variable for the person of interest and interact it with every covariate? Or perhaps is it better to set the variables of all unaffected persons to the mean? I am new to Bayesian inference and am not 100% sure whether there is a recommended way of doing this. Additionally, can I still meaningfully use id_plot_cov when I only want to see the marginal effects on one person's ideal point (in which case the liberal/conservative distinction is irrelevant)?

Thanks again for all your great work on this package.

[saudiwin]

Hi michael -

Sorry I just saw this. Don't worry, development on the project is very active. I will have a major release (v. 1.0) within the next two months. It will have the ability to do within-chain parallelization to fit much bigger models.

Because the data are entered in long-form, you can code the covariates however you want - they can vary by person, by bill (item), by time point, etc. In your case, I would probably add a dummy variable for that one person and then interact the covariates with that dummy variable. You would then have a baseline (all of the legislators) plus an interaction effect specific to that person.

I'll also have better covariate plotting in the next version. Unfortunately, a lot of my time is currently taken up with a coronavirus-related project :(, but getting this out is a top priority.

[michalovadek]

That's great to hear! And thanks for the recommendation, I will try out the dummy interaction. I might open a few other issues to suggest some features that I thought would benefit the package, although you may have already had these in mind for the next release.

[saudiwin]

Sure thing! I'm happy to hear your suggestions.