Improved Conditional Logistic Regression using Information in Concordant Pairs with Software

2026-02-09

AI Summary

This study addresses the limited statistical power of conventional conditional logistic regression (CLR) in estimating the additive effect of a binary treatment on the log-odds, which arises from its exclusive reliance on discordant pairs and disregard for covariate information in concordant pairs—particularly problematic in small samples or under nonlinear log-odds specifications. The authors propose a novel approach that systematically incorporates covariate information from concordant pairs into informative Bayesian priors, seamlessly embedded within the standard CLR framework. This integration substantially enhances estimation efficiency and testing power. The method demonstrates superior performance in both small-sample settings and models with nonlinear log-odds structures, and is efficiently implemented via the open-source R package bclogit.

Technology Category

• Machine Learning: Statistical Relational/Logic Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Description Logics

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

We develop an improvement to conditional logistic regression (CLR) in the setting where the parameter of interest is the additive effect of binary treatment effect on log-odds of the positive level in the binary response. Our improvement is simply to use information learned above the nuisance control covariates found in the concordant response pairs'observations (which is usually discarded) to create an informative prior on their coefficients. This prior is then used in the CLR which is run on the discordant pairs. Our power improvements over CLR are most notable in small sample sizes and in nonlinear log-odds-of-positive-response models. Our methods are released in an optimized R package called bclogit.

Problem

• conditional logistic regression
• concordant pairs
• binary treatment effect
• log-odds
• nuisance covariates

Innovation

• conditional logistic regression
• concordant pairs
• informative prior
• discordant pairs
• binary treatment effect