Is Logistic Regression model fit output correct?

[czopluoglu]

I can't make sense of SPSS output for model fit indices in logistic regression. Deviance and -2LL statistics do not match with the earlier version and other software. I had the same issue with many datasets. Here is a simple example. Let x (predictor) be 1,2,3,4,4,5,6,7,8,8,9 and y (dependent var.) be 0,0,0,0,0,0,1,1,0,1,1.

The output from an earlier version of SPSS (I don't know how early it is) is




from R



 and from SPSS 19




You will notice that parameter estimates are all identical. However, while the deviance and -2LL statistics are matching for the earlier version of SPSS and R, those statistics are so different for SPSS 19, and make no sense.

Do you have any idea what is wrong here?

Thanks for your help in advance.

[Bruce Weaver]

Judging from that SPSS 19 output, you used GENLIN (rather than LOGISTIC REGRESSION) to estimate the model, with syntax something like this:

* Generalized Linear Models.
GENLIN y (REFERENCE=LAST) WITH x
  /MODEL x INTERCEPT=YES
 DISTRIBUTION=BINOMIAL LINK=LOGIT
  /CRITERIA METHOD=FISHER(1) SCALE=1 COVB=MODEL 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.


If you use LOGISTIC REGRESSION, like this...

LOGISTIC REGRESSION VARIABLES y
  /METHOD=ENTER x
  /PRINT=CI(95)
  /CRITERIA=PIN(0.05) POUT(0.10) ITERATE(20) CUT(0.5).

...you'll find a Model Summary box in the output that shows -2LL = 7.664.  

I know that doesn't answer your question, but it might at least clarify for some other readers how you estimated your model in v19.  

czopluoglu wrote

I can't make sense of SPSS output for model fit indices in logistic regression. Deviance and -2LL statistics do not match with the earlier version and other software. I had the same issue with many datasets. Here is a simple example. Let x (predictor) be 1,2,3,4,4,5,6,7,8,8,9 and y (dependent var.) be 0,0,0,0,0,0,1,1,0,1,1.

The output from an earlier version of SPSS (I don't know how early it is) is




from R



 and from SPSS 19




You will notice that parameter estimates are all identical. However, while the deviance and -2LL statistics are matching for the earlier version of SPSS and R, those statistics are so different for SPSS 19, and make no sense.

Do you have any idea what is wrong here?

Thanks for your help in advance.

[czopluoglu]

Thanks for your help. I was not aware of the syntax used. I just followed the point-and-click procedure:

Analyze -- Generalized Linear Models -- Generalized Linear Models... Then, I chose Binary Logistic as a type of model, and specify the response and predictor variables etc. I was not aware that GENLIN is the function used.

So, how should we conclude? GENLIN and LOGISTIC REGRESSION procedures use two different estimation methods. What bothers me is how could we have two different -2LL values while we have exactly same estimates. From GENLIN, LL value was -3.139, so the deviance is 6.278, not even close to 7.664 obtained from LOGISTIC REGRESSION, but we have identical parameter estimates. Is this usual, or something strange?

Not all students appreciate using syntax language, so I am trying to figure out to reproduce results by point-and-click procedure.

Thanks again for the help.

cengiz