1
$\begingroup$

I want to do a logistic regression in SPSS. However, since I analyse unemployment spells the subjects are sometimes repeated (violating the independence assumption of the regression). One way of removing the within subject variation is by applying a Genlin model with the repeated subject subcommand (in essence a GEE model). Thus, I tried out a Genlin model with binomal probability and the logit link, comparing it to a standard logistic regression. I used the exact same variables in the two procedures.

However, the results that was delivered from the Genlin procedure was inverted relative to that of the logistic regression. For instance: Exp(B) for women (of the independent variable sex/gender) was just above 2.0 in logistic regression while being at 0.49 in Genlin. The same happened with every independent variable.

  • Any suggestions to why the results of the Genlin procedure is inverted?
  • Is there any way to get the Genlin results in accordance to the logistic regression?
$\endgroup$
3
  • 2
    $\begingroup$ Check the other tables of the outputs and look for the coding schemes. These two modules might have coded your sex variable differently; one might have used male as reference while the other one used female. $\endgroup$ – Penguin_Knight Oct 15 '13 at 1:10
  • $\begingroup$ No, that is not it. I have been very careful to get the same reference for every variable in both logistic and Genlin. In every respect (since I have left out the repeated subcommand) the models should be the same. $\endgroup$ – Cookie Monster Oct 15 '13 at 6:10
  • $\begingroup$ It's the reference of the outcome that matters. How is this coded? $\endgroup$ – Jeremy Miles Feb 19 '14 at 17:00
3
$\begingroup$

This is because the default for the reference category in GENLIN is FIRST. Try changing this to LAST.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.