Ordinal Logistic Regression Analyses

Besides the binomial logistic regression analyses discussed above, UniLogistic macro can be employed for conducting multivariable ordinal (proportional-odds) logistic regression analyses when the outcome is ordinal. The mechanics of conducting analyses is exactly similar to that of the binomial logistic regression discussed above, except the modeltype argument:

modeltype = ordinal, /* Type of outcome variable */

The rest of the procedure remains the same.

Example

We will use the dataset LOWBWT (Low Birth Weight) used for conducting univariable ordinal logistic regression analyses using the UniLogistic macro. This dataset is from a study conducted to investigate factors associated with low birth weight and is described in detail in Applied Logistic Regression by Hosmer and Lemeshow (2000) (© John Wiley & Sons Inc).

You can download this dataset and the implementation code used to analyse this data from here (You do not need to download the dataset again if you downloaded it for conducting univariable ordinal logistic regression analyses).

• Dataset.
• Dataset description (from the University of Massachusetts website): http://www.umass.edu/statdata/statdata/data/lowbwt.txt
• Implementation code (sas).
  We have also included the MultiLogistic macro, although you do not need to download it again if you downloaded it earlier for conducting multivariable binomial logistic regression analyses.
• MultiLogistic macro(sas)
  We have included here a video demonstrating use of the macro for conducting multivariable ordinal logistic regression analyses and the output files produced by the macro.
• Video (wmv) (32.3MB; 2.07min)
• Output file (xls)

Note that the results are presented in a similar format to the binomial logistic regression. Of course, interpretation of some estimates, particularly odds ratios, will be different for ordinal analyses from binomial logistic regression. Please consult a statistician or read the other resources, if unsure.