Cross Sectional Data and Unobserved Heterogeneity

[Leon Wildon]

Hi. I’m doing a cross sectional study and know that for observed heterogeneity I can add control variables e.g. population. I was wondering however if I could adapt the Least Squares Dummy Variable method used for panel data on my cross sectional data to control for any unobserved heterogeneity.

Also, if the answer is I can, I assume I don’t need to use control variables like population anymore as this heterogeneity is accounted for in the differing slopes?

If I can’t, is there any other way to control for unobserved heterogeneity with cross sectional data? Or can I only account for the observed by including useful control variables?

Thanks.

[Andrew Musau]

Unobserved heterogeneity is endemic in cross-sectional analyses. Without repeated observations of the same units (panel data), all you can do is to find control variables that will capture some (hopefully much) of the heterogeneity.

I was wondering however if I could adapt the Least Squares Dummy Variable method used for panel data on my cross sectional data to control for any unobserved heterogeneity.

As you observe each unit only once, including an indicator for each unit will result in an indicator for each observation. Not only will you not have enough degrees of freedom to estimate a regression with these indicators but they will also be collinear with your main variables. So this is not a viable solution.