Subscribe Bookmark RSS Feed

Mahalanobis Distance

bluenose

Community Trekker

Joined:

Sep 11, 2011

Hello,

I have a big dataset containing 94 variables. I ran the multivariate platform on JMP 9, and calculated the correlations. A lot of the variables are correlated.

My next step was to seek for outliers using the Mahalanobis distance. I ran the function, but out of approximately 70 subjects (rows), only 8 had a disance being calculted. All the rest were missing values. My dataset contains missing values, buy minority in compare to the data available. Why did I get only 8 distances and over 60 missing values ? Can't JMP handle missing values or there is another reason ?

Thank you in advance

4 REPLIES
bluenose

Community Trekker

Joined:

Sep 11, 2011

I figured that one out already, if there are missing values in a row, it doesn't calculate the MD.

I took a few columns out, and now got more reasonable results. What I don't understand is, how both MD and T^2 give the same outliers, but jacknife gives different outliers...

michael_jmp

Staff

Joined:

Jun 23, 2011

Hello,

Thanks for using JMP software!  We are aware of the issue of the jackknife giving different outliers than both MD and T^2, and we are working to correct that for future releases of JMP.

Thanks,
Michael

Michael Crotty
Sr Statistical Writer
JMP Development
michael_jmp

Staff

Joined:

Jun 23, 2011

You should be able to track when this is fixed in future JMP releases on the JMP support site:

http://www.jmp.com/support/notes/45/845.html

Thanks,
Michael

Michael Crotty
Sr Statistical Writer
JMP Development
bluenose

Community Trekker

Joined:

Sep 11, 2011

Thank you Michael, I will rely on the MD's for now.