Non-linear effect on sole authorship

[AaronGullickson]

The relationship of sole authorship in the between models is clearly a diminishing (negative) returns type model, so we should think about how to fit that. One obvious solution is logging the x variable, but logging percents always feels weird to me. Another option is a squared term. Lets play with it.

[AaronGullickson]

I am kind of curious if non-linear effects there kind of fuck up the fixed effects interpretation of the within effect, so probably want to leave it as linear in the final models.

[AaronGullickson]

I just added a commit that does some exploratory work here. There is not doubt that logging the disciplinary sole authorship variable provides a better fit. I don't know if its worth switching to that for the paper though as its still the same basic story and harder to interpret.

Its also clear that even when transformed, the within discipline effects are identical which is kind of cool and unexpected.

[AaronGullickson]

Ok, I decided I prefer the transformed variable better since it fits better and made that change as part of PR #14.