Which Test?

Draft: This page has not been reviewed by a statistician yet. Use it as a starting point and check important choices with a statistician.

Multinomial logistic regression

Models an outcome with three or more unordered categories, giving an odds ratio for each category versus a reference category.

Clinical example

Which treatment patients receive (surgery, radiotherapy or watchful waiting), and how that relates to age and stage.

When to use it

  • The outcome has three or more categories without a natural order.
  • You want to adjust for other variables.
  • Each category is compared with one reference category.

Assumptions to check

Independent observations

One value per patient.

Choose the reference category on clinical grounds

Every odds ratio is relative to it, so pick the category that makes the comparison meaningful, for example standard care.

Enough patients in each category

Small categories give unstable estimates.

What to report

  • OR per category vs reference, with 95% CI

Alternatives

If your situation is…Use
An ordered scoreProportional odds (ordinal logistic) regression

In your software

R
# R
library(nnet)
d$treatment <- relevel(factor(d$treatment), ref = "surgery")
fit <- multinom(treatment ~ age + stage, data = d)
summary(fit)
exp(coef(fit))   # odds ratios versus the reference category
SPSS
NOMREG treatment (BASE=FIRST) WITH age stage
  /PRINT=PARAMETER SUMMARY.

Sources