turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Grubbs outlier test and distributions

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Feb 4, 2013 10:23 AM
(1529 views)

Hello all. I have a very high value in a univariate dataset and wanted to test if it were an outlier. I have downloaded and used the Grubbs outlier test script. However, the assumption for the Grubbs outlier test is that the data come from a normal distribution -- but I am unclear from the references I have read if this refers to the raw data or the residuals. It has been suggested that what could look like an outlier could occur if the data are distributed log-normally.

If anyone has used this test before, could you specify if I need to test the distribution of my raw data or of my dataset's residuals? I'm assuming for either of these, I should be testing the distribution including the value I suspect. Thanks!

1 REPLY

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

May 30, 2014 9:00 AM
(963 views)

The script adds the result of the Grubb's test to a Distribution platform and opens the normal quantile plot to so that you can assess normality of the sample, inferring that the population is normally distributed. If the data are otherwise normally distributed but contain a discordant outlier, it might fail a normality test but you should still see linearity in the plot. Regardless of the outlier, non-normal data should not appear linear in this plot. The normal quantile plot should make it clear if this is the case.

If this example is univariate, then where do 'residuals' come from?

Learn it once, use it forever!