Residual plots interpretation

[Janka Zitha]

Dear all,

I built linear regression models in which I controlled for confounding effect of several variables. I investigated associations between race and C-reactive protein and sex and C-reactive protein.
In order to validate final regression models I obtained residuals plots. From what I know, residuals are supposed to fluctuate randomly around 0 horizontal line. Yet, I have a problem in interpreting my plots. Would anyone kindly explain me whether there is much of a heteroscedasticity? Thank you

[Clyde Schechter]

In the Race adjusted model, there is a hint of heteroscedasticity, with the variance getting a bit smaller as the fitted value increases. But it is really pretty minor. I would probably just ignore it, but if you are not comfortable doing that, you can use the -vce(robust)- option on your regression. Then you don't even have to think about heteroscedasticity.

The second model looks just fine to me.

[Bruce Weaver]

I agree with Clyde. With real world data, you'll rarely see residual plots that look much better than that.

[Carlo Lazzaro]

Janka:
as an aside to previous helpful advice, you may complete your -regression postestimation- analysis on residuals distributions via:

estat hettest

[Nick Cox]

I agree with the consensus that these plots are (very) well behaved.

For checking for heteroscedasticity it sometimes helps to plot |residual| (use abs()) or even its square root.

This isn't a difficult calculation but rvfplot2 (SJ) makes it easy to look at this just through an option.

search modeldiag

gives download locations.

[Janka Zitha]

Thank you all for a valuable advice and suggestions. Really much appreciated!