# Estimating Studentized Residuals (or Another Similar Measure) After Linear Regression With Robust Standard Errors

I have estimated a linear multiple regression with robust standard errors using Stata (regress depvar indepvar1 indepvar2 indepvar3 indepvar4 indepvar5, robust).

Given that I've used robust standard errors, Stata doesn't allow me to estimate studentized residuals. I haven't found a technical note explaining why this is the case, but I believe that studentized residuals are not a statistically well-defined concept after an estimation with robust standard errors.

Is there any other way in which I could obtain a measure such as studentized residuals after linear multiple regression with robust standard errors? I am especially interested in comparing the residuals of the observations in my dataset and seeing whether they seem to fall above or below the model prediction.

• Your Stata syntax example was illegal. It's incidental to the question, so I edited it. Nov 23 '15 at 14:30
• I don't think you need Studentized residuals to see whether values fall above or below model predictions. Any flavour of residuals would do that. Nov 23 '15 at 14:31
• Thanks for your responses @NickCox! Perhaps I didn't specify my question well enough: as internally studentized residuals are the 'normal' residuals divided by their standard deviation (without removing the observation in question), they allow you to see whether the residuals of certain observations are more or less than the expected frequency. Ie, if the int studentized residual is larger than 2 or smaller than -2, the observation's value is larger/smaller than predicted by the model. I thought that only studentized residuals use this scale, where 2 and -2 mark area for normal residuals. Nov 23 '15 at 17:49
• I don't understand your extra point. The question clearly states interest in what is above or below prediction. The precise scale of residuals is secondary and even at best +2 and -2 don't mark any kind of formal confidence limits, if that is what you are seeking. What you mean by "expected frequency" here? Nov 23 '15 at 17:53
• Thanks again for your quick comment @NickCox! Some materials that I've come across seem to state that for internally studentized residuals, values between 2 and -2 contain normal residuals, and any with absolute values larger than that are significantly bigger/smaller. I wanted to be able to look at the studentized residuals from my model and evaluate whether they're above or below this 2 and -2 region. I think that this would allow me to say that eg. the observation with a studentized residual 3 has a higher value on their dependent variable than would be predicted by the model. Nov 23 '15 at 18:07