In the initial versions of LOG REG, we followed some claims in the literature to the effect that you could use the -2LL and Pearson GOF statistics to test model fit. This does not work for case by case data; the statistics don't follow the presumed distributions under the null. So we dropped the df and significance values for these for later releases. The Model chi^2 is like the overall F-test in a linear regression (assuming the model was fitted with one METHOD subcommand), so it's comparing two nested models and is fine, but it doesn't test goodness of fit of the model. There's a macro you can use to do a Hosmer-Lemeshow goodness of fit test. The read me file is hl.rm, in

ftp://ftp.spss.com/pub/spss/statistics/nichols/macros.

David Nichols Senior Support Statistician SPSS Inc. nichols@spss.com

>---------- >From: Sang-Gyun Lee[SMTP:sglee@welfare.net] >Sent: Friday, September 19, 1997 10:30 AM >To: SPSSX-L@UGA.CC.UGA.EDU >Subject: Help! goodness of fit in Logistic Regression > >I need your avices urgently. >In Logistic regression, statistics provided regarding goodness of fit >are -2LL, goodness of fit, and Model chi-square. When I want to measure >goodness of fit, which statistics should I check? >The manual of SPSS ver 4.o recommends using of both goodness of fit and >model chi-square, but SPSS ver. 6.0 which I used doesn't provide >significance level of goodness of fit. Only model chi-square provide it. > >If I use only model chi-square for the test of goodness of fit, are >there problems? > >In addition, explain the concept of goodness of fit, please. > >Thanks in advance, > >Sang-Gyun Lee >