Ordinal Logistic Regression in SPSS

Question

This is my first time conducting an ordinal logistic regression on SPSS, and I want to check for the assumptions.

Assumption 1: My dependent variable is indeed ordinal. My variable is anxiety symptom severity levels: normal, mild, moderate, severe, and extremely severe.

Assumption 2: My independent variables are either continuous or categorical. I have 3 categorical variables: sex, educational attainment, and marital status. I have 4 continuous variables: age, number of children, and two other variables that are scored on self-report questionnaires.

Assumption 3: I tested for multicollinearity using multiple linear regressions. To do so, I generated seven multiple linear regression each one having one predictor variable of the OLR as a dependent variable. All VIFs were less than 3. So I conclude that there are no multicollinearity among the predictors.

Assumption 4: I am testing for proportional odds. The test of parallel lines on SPSS was insignificant, p>.05. I read that we should do a full likelihood ratio test comparing the fitted location model to a model with varying location parameters and then run separate binomial logistic regressions on cumulative dichotomous dependent variables.

Should I do this given that the test of parallel lines is insignificant? Or do we run separate binomial regressions regardless of the test of parallel lines outcome? Also, how do we conduct these? Please share with me resources that explain the procedures on SPSS.

If the test of parallel lines is violated, what to do next?

Thank you.

Answer 1

The test of parallel lines in PLUM (Ordinal Regression in the menus) is a likelihood-ratio test. If you're not able to assume parallel functions, you can fit a multinomial logistic model in the NOMREG (Multinomial Logistic in the menus) procedure, which fits a separate function for each of K-1 logits for a K-level response, but does them altogether, which is better than doing separate binary response models.