The Statistical Analysis of Pairwise Experiments with Qualitative Responses

TL;DR

This paper develops a statistical method to estimate interaction values from ordered categorical responses in pairwise experiments, providing optimal properties and applications in biology, medicine, and imaging.

Contribution

It introduces a novel estimation technique for qualitative pairwise interaction data with optimal properties and broad applicability.

Findings

Method explains maximum variance with minimal parameters

Interactions interpreted as correlation, error size estimable

Single run per pair suffices for experiment

Abstract

Suppose an experiment is conducted on pairs of objects with outcome responses a continuous variable measuring the interactions among the pairs. Furthermore, assume the response variable is hard to measure numerically but easy to be coded into ordered categories such as low, moderate, and high levels of interaction. In this paper we estimate the unknown interaction values from the information contained in the coded data and the design structure of the experiment. The method of estimation is shown to enjoy several optimal properties such as explaining maximum variance in the responses with minimum number of parameters and for any probability distribution underlying the responses. Other properties of the method include: the interactions have the simple interpretation of correlation, size of error is estimable from the experiment, and only a single run of each pair is needed to carry out the experiment. We also explore possible applications of the technique. Three applications are presented, one on protein interaction, a second on drug combination, and the third on computer imaging. The first two applications are illustrated using real life data while for the third application the data are generated via binary coding of an image.