Skip to content

Emails I never finished reading

This one came in today (i.e., 2 months ago):

Dear colleague,

Can I ask you a quick question? I am seeing more and more research projects (many in the economic sciences) in which researchers use dummy-codes to account for non-independence due to higher-order units. So when the researchers have data from employees in 15 different departments, they include 14 dummy-codes in the OLS regression.

I have heard and read many times that this is not the right data analytic strategy. As far as I remember the parameter estimates are correct, but the standard errors of the parameter estimates are too large and that the dfs tend to be too high. In short, multilevel modeling is the right strategy, whereas the use of dummy codes produces incorrect inferential statistics (at least with a large number of higher-order units). Can you point me toward a publication that examines this question?
Sorry to bother you with this. I know you are probably very busy. I have looked around quite a bit and despite all my efforts I have not been able to find a relevant publication. . . .

Ummmm, yeah I’m busy! I’ll answer pretty much any question, but not if you don’t even bother to cut-and-paste my name into the salutation!


  1. Anonymous says:

    On the plus side, this is a question I hear asked a lot.

    If your _colleague_ did ask in a nice way, what would you have recommended?

Leave a Reply