Robust option in a ols model

[Mahdie Raj]

Hi everyone,
I was wondering if one can use robust option in a ols regression model that includes both individual and higher level data. In other words, does robust option in such a ols model provides unbiased SEs in a two-level model or a multilevel model must be used. Thanks

[Jeff Wooldridge]

OLS is not a model. It is an estimator that can be applied any time you have a model linear in parameters. It could be cross section data, time series, panel data, or have a nested structure — the case you’re interested in.

With out getting into subtleties, you’ll probably want to use the vce(cluster id) option where id is the highest level. If you tell me more about your data structure I might have additional thoughts.

[Mahdie Raj]

Thanks. I am trying to use the ols regression to estimate the impact of a combination of individual level and organizational level variables on a continuous outcome (job satisfaction).

[Jeff Wooldridge]

How many organizations and individuals?

[Mahdie Raj]

3000 individuals and 80 organizations

[Jeff Wooldridge]

You can get away with running OLS with those dimensions and clustering at the organization level. That’s also robust to heteroskedasticity. You can also use -mixed- to estimate a two-level model. But then you should still cluster your standard errors at the organization level. Doing both is a good robustness check.