logistic regression

[Margot Marin]

Hi List,

 

I’ve executed a logistic regression with two explanatory variables, with the next procedure:

 

* Modelos lineales generalizados.

GENLIN Ydep (REFERENCE=FIRST) BY grupo X2 (ORDER=DESCENDING)

  /MODEL grupo X2 INTERCEPT=YES

 DISTRIBUTION=BINOMIAL LINK=LOGIT

  /CRITERIA METHOD=NEWTON SCALE=1 COVB=ROBUST MAXITERATIONS=100 MAXSTEPHALVING=5

    PCONVERGE=1E-006(ABSOLUTE) SINGULAR=1E-012 ANALYSISTYPE=3(WALD) CILEVEL=95 CITYPE=WALD

    LIKELIHOOD=FULL

  /MISSING CLASSMISSING=EXCLUDE

  /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION.

 

The outcome says that:

 

It is possible that there is a separation almost complete within the data. There are no estimations of maximum likelihood for the file.

 

I’ve read that a solution for this is to use a logistical regresión hidden created by Rousseeuw and Christman. Can you guide me if this is a correct solution and if this command is implemented on SPSS.

 

Thank you very much.

--
Magi

[Bruce Weaver]

This is not an answer, just more information for other list members.

The "in press" version of the relevant article can be downloaded from Christmann's website:

   http://www.stoch.uni-bayreuth.de/de/CHRISTMANN/

Look for "Robustness against separation and outliers in logistic regression".

A document describing the S-Plus package (hlr) can be found here:

   http://cran.r-project.org/web/packages/hlr/hlr.pdf

Margot Marin wrote

Hi List,




I’ve executed a logistic regression with two explanatory variables, with
the next procedure:



* Modelos lineales generalizados.

GENLIN Ydep (REFERENCE=FIRST) BY grupo X2 (ORDER=DESCENDING)

  /MODEL grupo X2 INTERCEPT=YES

 DISTRIBUTION=BINOMIAL LINK=LOGIT

  /CRITERIA METHOD=NEWTON SCALE=1 COVB=ROBUST MAXITERATIONS=100
MAXSTEPHALVING=5

    PCONVERGE=1E-006(ABSOLUTE) SINGULAR=1E-012 ANALYSISTYPE=3(WALD)
CILEVEL=95 CITYPE=WALD

    LIKELIHOOD=FULL

  /MISSING CLASSMISSING=EXCLUDE

  /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION.



The outcome says that:



It is possible that there is a separation almost complete within the
data. There
are no estimations of maximum likelihood for the file.



I’ve read that a solution for this is to use a logistical regresión hidden
created by Rousseeuw and Christman. Can you guide me if this is a correct
solution and if this command is implemented on SPSS.



Thank you very much.
--
Magi