Online Program Home
My Program

Abstract Details

Activity Number: 123 - Binary and Ordinal Outcome Regression
Type: Contributed
Date/Time: Monday, July 30, 2018 : 8:30 AM to 10:20 AM
Sponsor: Biometrics Section
Abstract #327092
Title: Improving the Hosmer-Lemeshow Goodness-of-Fit Test
Author(s): Nikola Surjanovic* and Thomas Loughin
Companies: and Simon Fraser University
Keywords: Goodness-of-fit; Hosmer-Lemeshow test; Logistic regression; Poisson regression

The real relationship from which data are drawn may not adhere to a proposed model for this structure. Goodness-of-fit (GOF) tests help to identify when a model is a poor fit for a given dataset. A special GOF test for logistic regression models, the Hosmer-Lemeshow test, is used extensively in applications in many disciplines. It has been extensively studied, and some of its properties are somewhat surprising. There is evidence that the test functions poorly under certain circumstances. We will explore the origins of some of these properties and suggest ways to improve the test based on these findings. For example, the test statistic consists of a sum of squared Pearson residuals computed on grouped data. Preliminary evidence suggests that using adjusted Pearson residuals may allow the test to maintain its size better under certain circumstances. We will examine this, and other possible corrections to the test. Also, we will discuss the performance of a parallel test for Poisson regression models.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program