R2/explained variance for ACS data

I received a reviewer comment from a paper in which I am using IPUMS data that reads:

“What is typical for regression R2 for American Community Survey Microdata? The R2 for the four models seem quite high, and providing a reference point would be useful for readers.”

Currently, I’m exploring research on aggregation bias to attempt to respond to this comment, but I wanted to see if any other users or staff have thoughts on how to approach the question. In the study, the dependent variable is an MSA-level variable created using IPUMS microdata, and the independent variables are also derived from the ACS (estimates downloaded directly from the Census).

Thanks for any thoughts!

I don’t have an answer for the question of what is a typical R^2 for regressions using ACS data and it seems unlikely to me that you will be able to find an answer to such a broad question. Some relationships may involve only a few well observed variables and thus will generate higher R^2 values when the correct variables are specified in a regression model. Other relationships are much more complex and involve unobserved variables that cannot be included in the model, causing a decrease in the R^2. The expected R^2 will depend on the relationships you’re evaluating in your paper and what previous research has found on the specific issue you’re studying.