Identifying outliers based on second stage residuals

[Ashani Abayasekara]

Hi everyone,

I'm running fixed-effects IV regressions to identify the causal effects of housing costs on individual health. I want to identify outliers in the housing cost measure, based on second stage residuals.

How would I identify these residuals, and the residual mean of housing costs, to determine residuals outside 3 SD of the residual mean? The command I'm currently using for the IV regressions is xtivreg28. (Schaffer, M.E., 2012 http://ideas.repec.org/c/boc/bocode/s456501.html).

Any help you can provide on this would be much appreciated.

Many thanks,
Ashani

[George Ford]

xtivreg2 does not allow for the predictions of residuals. might be a way to use the coefficients and predicted fe to calculate it.

[George Ford]

xtivreg does permit the prediction of the things you need. the results are not exactly the same between xtivreg and xtivreg2.

[Ashani Abayasekara]

Ok, I will look into this, thank you.

[Nick Cox]

In plain or vanilla regression outliers can pull the fit towards them. It's not always easy to identify outliers from residuals. I doubt that this flavour of regression is different in that respect, although please correct me otherwise.

[Jeff Wooldridge]

I agree with Nick -- it's a tricky business determining what is an outlier based on residuals. I'm guessing using fixed effects helps with outliers because it removes unit-specific averages. The problem becomes outliers within a unit, but not so much across units. If you want the residuals after netting out the "fixed effects" -- which I think are the ones you should look at if you must -- then just use the following:

xtivreg y x1 ... xk (w = z1 ... zm), fe
predict ehat, e

[Ashani Abayasekara]

Thank you very much, Prof. Cox and Prof. Wooldridge for your inputs. I will try this approach.