SPSS Statistics

 View Only
Expand all | Collapse all

Binary logistic regression, testing for linearity of the logit

  • 1.  Binary logistic regression, testing for linearity of the logit

    Posted Fri December 10, 2021 12:59 PM
    Hi! I am investigating littering behaviour using a binary logistic regression model in SPSS 27. One predictor variable is the number of trash bins that are available at the test sites. They are either 0, 1, 2, 3 or 5. I thought that the number of trash bins is a continouse variable that requires the testing for linearity of the logit. Is that right?

    While its of course not possible to calculate the Ln for 0, the test for the remaining cases shows non-linearity for the number of trash bins. But number of trash bins is a significant and important predictor in my model. Is there something I could do about it or does the violation of assumption indicates I can not use number of trash bins as predictor variable?

    Kind regards!

    ------------------------------
    Therese
    ------------------------------


  • 2.  RE: Binary logistic regression, testing for linearity of the logit

    Posted Fri December 10, 2021 01:37 PM
    The question would be whether the relationship is linear in the context of the logistic regression that includes other variables with bins as a predictor.  Two things come immediately to mind.

    First, you can treat the number of bins as a factor (categorical), in which case linearity is irrelevant.  LOGISTIC and NOMREG have different ways of expressing this - categorical or factor.

    Second, the Hosmer and Lemeshow test tests whether the form of the table conforms to the logistic model.  See this link in the logistic help

    You could use this on both the categorical and continuous forms of the equation.

    --





  • 3.  RE: Binary logistic regression, testing for linearity of the logit

    Posted Mon December 20, 2021 04:13 AM
    Thank you!

    ------------------------------
    Therese Nitschke
    ------------------------------