Testing treatment effects in unconfounded studies under model misspecification: Logistic regression, discretization, and their combination
Authors:
- 1 July 2009
- journal article
- research article
- Published by Wiley in Statistics in Medicine
- Vol. 28 (20) , 2531-2551
Abstract
Logistic regression is commonly used to test for treatment effects in observational studies. If the distribution of a continuous covariate differs between treated and control populations, logistic regression yields an invalid hypothesis test even in an uncounfounded study if the link is not logistic. This flaw is not corrected by the commonly used technique of discretizing the covariate into intervals. A valid test can be obtained by discretization followed by regression adjustment within each interval. Copyright © 2009 John Wiley & Sons, Ltd.Keywords
Funding Information
- NSF (SES-0550887)
- NIH (10006441691)
This publication has 35 references indexed in Scilit:
- Risk of respiratory morbidity in term infants delivered by elective caesarean section: cohort studyBMJ, 2007
- Robit Regression: A Simple Robust Alternative to Logistic and Probit Regression2004
- Factoring vs Linear Modeling in Rate EstimationEpidemiology, 1998
- Testing Statistical Hypotheses.1997
- A Comparison of the Performance of Model-Based Confidence Intervals When the Correct Model Form Is UnknownEpidemiology, 1994
- Generalized Logistic ModelsJournal of the American Statistical Association, 1988
- Reducing Bias in Observational Studies Using Subclassification on the Propensity ScoreJournal of the American Statistical Association, 1984
- Graphical Methods for Assessing Logistic Regression ModelsJournal of the American Statistical Association, 1984
- The central role of the propensity score in observational studies for causal effectsBiometrika, 1983
- Cumulative Frequency FunctionsThe Annals of Mathematical Statistics, 1942