Residuals for fracreg (with fractional outcome dependent variable)

[Kwali Wood]

When using fracreg, Stata 14 does not want to generate residuals in postestimation, saying

option resid not allowed.

But after fracreg with the dependent variable "depvar," it is happy enough to generate residuals using

predict yhat
generate residual = depvar - yhat

That generated thing is the residual, right? What am I missing? (maybe a lot!)

[FernandoRios]

Hi kawali
i don’t think your are wrong with your definition of residuals.
however fracreg does not use that definition of residuals for the model estimation which is why stata doesn’t give you an option for reporting it.
that being said, perhaps the question. Is, what do you want to use the “residuals “ for?
fernando

[Kwali Wood]

Hi kawali
i don’t think your are wrong with your definition of residuals.
however fracreg does not use that definition of residuals for the model estimation which is why stata doesn’t give you an option for reporting it.
that being said, perhaps the question. Is, what do you want to use the “residuals “ for?
fernando

Thanks! -- here's the story: A paper in the Journal of Education for Business (https://www.tandfonline.com/doi/pdf/...B.82.6.357-362) showed how some undergraduate business schools "punch above their weight" in generating good outcomes for their graduates. Specifically, the authors characterized graduates' starting salaries as a function of the school's measured inputs and then analyzed the residuals of this ordinary regression estimation. Fordham's graduates, for example, had a starting salary of $52,500 -- which was $9865 above the predicted value. After regress, getting residuals is straightforward using predict .

But in my case, attempting to apply this idea, the problem is that I have a fractional outcome variable -- like "percentage of each college's law school applicants admitted to top-100 law schools." That variable necessarily lies from 0 to 1 at most. I'm trying to identify the undergraduate schools that "punch above their weight" in law school admissions and so I am looking for residuals -- what colleges do much better than predicted?

And that's why I'm looking for residuals, so to speak, in a model that doesn't use the residuals in estimation.

[FernandoRios]

ok that makes the question more clear.
If that is your purpose, then yes. that idea of residuals you propose is valid, because what you will be doing is finding which schools have an outcome above or below of what the model predicts.