B
Benoit
Grubbs test
Hi
I search the method to perform Grubbs test (test detects one outlier).
Many thanks
Benoit
Hi
I search the method to perform Grubbs test (test detects one outlier).
Many thanks
Benoit
D
Darius
It's like Chauvernet's criterion (Journal of Chemical Engineering Feb 13,1967 pag. 182,184).
It normalizes the data, and discart the datum according to the probability of being from a normal distribution and using a table of critical values.
https://graphpad.com/articles/grubbs.htm
I tink it's great to determine outliers, I use it very offten, it's better to make the statistical estimates strong, you can obtain very good estimates of the true variation of the process (without special events), but keep on mind "what make that datum different from the others?"
It normalizes the data, and discart the datum according to the probability of being from a normal distribution and using a table of critical values.
https://graphpad.com/articles/grubbs.htm
I tink it's great to determine outliers, I use it very offten, it's better to make the statistical estimates strong, you can obtain very good estimates of the true variation of the process (without special events), but keep on mind "what make that datum different from the others?"
C
Campy
See ASTM E 178-08: Standard Practice for Dealing with Outlying Observations. The first technique in the practice is Grubbs' test although the name is not mentioned in the text. Grubbs is referenced at the end of the document however.
D
Darius
Trying to add new stuff to this thread.
The Grubbs and the Chauvernet take the distribution as Gaussian (aca. Normal), I found some time ago (on the NIST site) that there is another way to detect outliers using quartile.
OUTLIERS - NIST
The interesant part is that the use of quartile, work for non Gaussian distributions. Also there is the link to Grubbs method.
I found also in
WAPEDIA Outliers
An Excelent recompliation of methods, and a good reference to Peirce's criterion. And Cook's distance in regression problems, only exclude points which exhibit a large degree of influence on the parameters.
The Grubbs and the Chauvernet take the distribution as Gaussian (aca. Normal), I found some time ago (on the NIST site) that there is another way to detect outliers using quartile.
OUTLIERS - NIST
The interesant part is that the use of quartile, work for non Gaussian distributions. Also there is the link to Grubbs method.
I found also in
WAPEDIA Outliers
An Excelent recompliation of methods, and a good reference to Peirce's criterion. And Cook's distance in regression problems, only exclude points which exhibit a large degree of influence on the parameters.
U
Umang Vidyarthi
Similar threads
