Reification In Factor Analysis: A Plasmode Based On Human Physiology-Of-Exercise Variables

Abstract

This study investigates the validity of multiple factor analysis in a known model in biology, and -- by extension -- in other fields. A plasmode was constructed by subjecting three test subjects to 100 different combinations of temperature, humidity and exercise levels, and measuring 45 physiological response variables, some of these three times during the experiment. Factor analysis of the correlation matrix among response variables with the putative causes included yielded seventeen factors, the majority of which were interpretable. Two of the factors were identified by the factor solution to be related to the putative causes. Exercise affects a suite of respiratory and cardiovascular variables, while experimental temperature affects skin temperatures of the subjects. Humidity did not show any direct effect on any of the response variables. Factor analysis of a second correlation matrix excluding the putative causes yielded very similar results. From plots, partial regression coefficients and analyses of variance relating the response variables to the putative causes one could have predicted the outcome of the factor analysis. We conclude that the factor analysis resolved the data into meaningful groups of correlated variables validating the use of factor analysis in a linear dynamic model. Putative causes associated with these groups of variables can be identified by means of tagging techniques.