Aggregated data

noetsi

Fortran must die
#1
We have to work with a form of regression where the data is first aggregated from an individual level to a state level and then a regression is run on this. I have never encountered a regression like this and wonder how it is done. Does anyone know what type of techniques does this I never heard of it before today.
 

noetsi

Fortran must die
#3
I don't have a link, this is a federal agency we answer to in part. I will see if I can send the actual document.

This is not meta analysis. It is some type of linear fixed effect regression. Of that I am certain. I have never heard of linear regression done this way, with aggregate DV, but then I am learning lots of new things. :p

I think what they do, this is a guess, is that they estimate an employment rate for all individuals from a given state and month (the ones that are employed as a percent of all). That is every individual who leaves the organization in that month and state gets the same DV. The problem with that is they are using predictors that are individual like gender and age. I am not sure how you would link those (someone suggested they were making the predictor the percent that were each dummy variable such as the percent that were men for gender. I have never seen regression done that way).

They provide limited details of what they do in their documentation. That has been the problem all along.