Prism offers two standard Goodness-of-fit tests, Hosmer-Lemeshow and likelihood ratio test (LRT). You should be cautious when utilizing these tests, as they do not test the same null hypotheses,and the P values need to be interpreted very differently. Each test is briefly explained below, while some additional information is provided in the results interpretation section of this guide.

Hosmer-Lemeshow test

This test uses the null hypothesis that the specified model is correct. As such, a small P value would suggest that the model is incomplete. Perhaps you would need to add additional independent variables or maybe interactions. This test is commonly included in other software, so we provide as an option. However, there has been much criticism regarding this test, and it is not recommended. Prism does the test as it was initially described, dividing the data into 10 groups by predicted outcome. The value 10 is arbitrary, and some programs (not Prism) let you choose another value (which can change the P value a lot).

Likelihood Ratio Test (LRT)

This null hypothesis is that an intercept-only model is correct. In other words, the null hypothesis is that none of the independent variables provides any information to help predict the outcome. As such, a small P value would suggest that an intercept-only model is insufficient. As implied by the name, this test uses the ratio of the log-likelihood values calculated for the specified model and the intercept only model.