search

lightsociologist / hirsch_test

Code and Data for Light, Gullickson, and Harrison Hirsch Manuscript
1 stars 0 forks source link

Rescale predictor variables #10

Closed lightsociologist closed 1 year ago

lightsociologist commented 1 year ago

I think the last thing that we need to finalize these models is to address the rescale predictor variables error that likely pertains to the University Productivity variable. What is best practice with this issue?

AaronGullickson commented 1 year ago

In commit c33f08831a276a4687f2b1ae9e0f5e979336874c, I rescaled it to use a standardized scale (grand mean centered with a standard deviation of 1). I think the other option would be to rescale it so that 1=2 SD. That is Gelman's recommendation, the idea being that it brings the size more in line with a 0-1 dummy variable. I am happy to do it that way instead, but we aren't scaling all the proportion variables that way (we do sole authored as a percent not a proportion), so not sure if its worth the effort. If we did it this way, I would recommend we move sole-authored back to a [0,1] proportion. That would put everything on the same more or less scale.

AaronGullickson commented 1 year ago

OK, I played around with some rescaling a bit and didn't really care for the results as I think its more intuitive to present the things we care about in terms of percents and gender gaps, so I think rescaling as done in the commit referenced above is the best way to go. If you agree, we can close this.