SPSS Statistics

 View Only
  • 1.  Complexe samples logistic regression: Odds ration differ from Exp (b)

    Posted Tue March 21, 2023 07:21 PM

    Dear colleagues, 

    While using the complexe samples modules, I did a logistic regression. 

    I was surprise to discover that the Odds Ration differ from the Exp(b) when there is only one variable (with 2 categories in the model).

    Is it common? If yes, which one should be reported?

    Cordially



    ------------------------------
    M-A
    ------------------------------


  • 2.  RE: Complexe samples logistic regression: Odds ration differ from Exp (b)

    Posted Wed March 22, 2023 10:55 AM

    Do you mean one binary outcome with one binary predictor?

    and Do you mean different OR (Exp b) from same analysis without complex samples routine? Or somehow different in same analysis?

    In the complex samples logistic set up (which I have used a lot lately) it matters a lot which outcome (1, 0) you say is the reference category. Which you choose determines if you are getting an odds increase OR or an odds decrease answer. 
    And if you asked for the separate output of the OR in that routine (which you can do), it matters again which category you set as high and low for that. 
    I am thinking you might be seeing the mirror of the pos and neg odds ratios. On is going one way in main output and other is other way in separate listing of the OR. See if dividing the number 1 by one of them gives you same vale as the other of them. I think you may find one is just the inverse of the other. If so, they are the same result. For example, an OR of .6 is equivalent to an OR of 1.667. I forget if you can make them be different in one run or not, but I know you can easily flip what you get depending on which category you want to talk about as the reference. 

    I hope that helps.  S



    ------------------------------
    Scott Stanley
    ------------------------------



  • 3.  RE: Complexe samples logistic regression: Odds ration differ from Exp (b)

    Posted Thu March 23, 2023 12:32 PM

    Dear Scott,

    It is exactly that, I was seing the mirror image. It is strange that in the same analyses the Exp (b) and the OR are not coded the same.

    I'll have to pay more attention in the further.

    Very helpful, 



    ------------------------------
    Marc Blanchette
    ------------------------------



  • 4.  RE: Complexe samples logistic regression: Odds ration differ from Exp (b)

    Posted Thu March 23, 2023 06:32 PM

    I'm so glad to hear that, Marc. 

    I've been living that routine for about 9 months now for a project, and I thought it likely you were describing something I'd seen scores of times.  GREAT.  Glad that helped.

    Scott



    ------------------------------
    Scott Stanley
    ------------------------------