log-likelihood distance

[kathrin-2]

Hello,

has anyone of you ever heard about a distance measure called log-
likelihood d? I've read this in conjunction with anomaly detection and
cluster analysis. This should be the sum of all of the variable deviation
indices where the variable deviation indix of a case is a measure of the
deviation of the variable value X from its cluster norm.
What I search is a formula of how to calculate this log-likelihood d!
Any help would be appreciated! Thanks!

Kathrin

[Reutter, Alex]

Kathrin,

Have a look at the TWOSTEP CLUSTER and DETECTANOMALY algorithms -- from the SPSS menus, choose Help > Algorithms, and you should be able to get to them.

Alex


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of kathrin
Sent: Thursday, February 08, 2007 2:09 AM
To: [hidden email]
Subject: [BULK] log-likelihood distance
Importance: Low

Hello,

has anyone of you ever heard about a distance measure called log-
likelihood d? I've read this in conjunction with anomaly detection and
cluster analysis. This should be the sum of all of the variable deviation
indices where the variable deviation indix of a case is a measure of the
deviation of the variable value X from its cluster norm.
What I search is a formula of how to calculate this log-likelihood d!
Any help would be appreciated! Thanks!

Kathrin

[parisec]

Hi all,

I have question of "which is more appropriate" versus a direct SPSS question.

I have a very large crosstab (8x20) table and i have always relied on the SRESID of +-1.96 to give me a clue as to which cell(s) contributed to the Chisquare and to see patterns in the data. But you can also get ASRESID which are the adjusted standardized residuals.

I just finished interpreting my table and am writing up the results and started looking for a citation for "standardized residuals". I pulled out Christensen's "Log-linear Models.." book and Agresti's "Intro to cagetorical data analysis" book and now i'm wondering if if i've been doing it wrong all these years.

Christensen just states that there are differences between the 2 Pearson and adjusted formulas but Agresti says "Adjusted residuals are preferable to Pearson residuals"...but doesn't really explain why.

For those of you who use these residuals, how do you decide on which one? There is not alot of difference in the values in some cases but sometimes it's enough for a cell to go from sig to or vice versa.

Thanks a bunch for your wisdom

Carol