Canonical Correlation Analysis of Data on Human-Automation Interaction

Abstract

Canonical correlation analysis is a type of multivariate linear statistical analysis, first described by Hotelling (1935), which is used in a wide range of disciplines to analyze the relationships between multiple independent and multiple dependent variables. We argue that canonical correlation analysis is the method of choice for use with many kinds of datasets encountered in human factors research, including field-study data, part-task and full-mission simulation data, and flight-recorder data. Although canonical correlation analysis is documented in standard textbooks and is available in many statistical computing packages, there are some technical and interpretive problems which prevent its routine use by human factors practitioners. These include problems of computation, interpretation, statistical significance, and treatment of discrete variables. In this paper we discuss these problems and suggest solutions to them. We illustrate the problems and their solutions based on our experience in using canonical correlation in the analysis of a field study of crew-automation interaction in commercial aviation.