I have a data set containing the entirety of the population I'm interested in, not just a sample. However, there is an abundance of outliers in this data, which we have determined is due to a lot of incorrect reporting. I'm interested in running linear regression for this data (even though the data is not normally distributed) and I want to see which type of residual would be more beneficial to use.

From russellpierce's answer here, "Studentizated scores uses Student's/Gosset's calculation for estimating the population variance/standard deviation from the sample variance/standard deviation (s). In contrast, Standardized scores (a noun, a particular type of statistic, the Z score) are said to use the population standard deviation ?(σ)."

I'm working in Python on this, and already have both types of residuals automatically generated. I know studentized residuals are typically used for outlier detection, but I assume that's usually because data scientists only have a sample to work from? Which one would be better to use for outlier detection when I know I have the data for the entire population?

  • $\begingroup$ You might find it more constructive to conceive of your data as a sample, too, if only to create the opportunity to model and account for the errors of "incorrect reporting." $\endgroup$
    – whuber
    Commented Jun 26, 2019 at 15:19


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.