Asymptotic inference for semiparametric association models

TL;DR

This paper establishes that asymptotic inference for the odds ratio parameter in semiparametric association models remains valid regardless of whether sampling is conditional on X or Y, extending previous results and linking to common regression models.

Contribution

It generalizes asymptotic inference results for association models, showing sampling conditional on Y yields the same inference as on X, and connects these models to standard regression frameworks.

Findings

Asymptotic inference for $olds heta$ is invariant to sampling condition.

Inference for regression parameters can be derived from association models.

Results extend to common regression models like GLMs and logistic regression.

Abstract

Association models for a pair of random elements $X$ and $Y$ (e.g., vectors) are considered which specify the odds ratio function up to an unknown parameter $\bolds\theta$. These models are shown to be semiparametric in the sense that they do not restrict the marginal distributions of $X$ and $Y$. Inference for the odds ratio parameter $\bolds\theta$ may be obtained from sampling either $Y$ conditionally on $X$ or vice versa. Generalizing results from Prentice and Pyke, Weinberg and Wacholder and Scott and Wild, we show that asymptotic inference for $\bolds\theta$ under sampling conditional on $Y$ is the same as if sampling had been conditional on $X$. Common regression models, for example, generalized linear models with canonical link or multivariate linear, respectively, logistic models, are association models where the regression parameter $\bolds\beta$ is closely related to the odds ratio parameter $\bolds\theta$. Hence inference for $\bolds\beta$ may be drawn from samples conditional on $Y$ using an association model.