Measuring Agreement

CHAPTER 8 DATA WITH COVARIATES

8.1 PREVIEW

So far the method comparison data were assumed to solely consist of observations from the measurement methods of interest on a number of subjects. Frequently, however, additional data on a number of influential covariates are also available. These covariates may be categorical or continuous quantities and may affect either the means or the variances of the methods. This chapter coalesces ideas from previous chapters to present a unified method that incorporates covariates in the analysis. Both unreplicated and repeated measurements data are considered. As before, our approach is to first model the data and then use the assumed model for evaluating similarity and agreement, and possibly repeatability as well. A case study illustrates the methodology.

8.2 INTRODUCTION

Quite often in a method comparison study, data are collected on a number of covariates in addition to the observations from the measurement methods of interest. The covariates may be categorical or continuous. They are fixed, known quantities and are assumed to be measured without error. They include subject-level covariates, for example, gender of the subject, and method-level covariates. Of course, we always have the “measurement method” itself as a method-level categorical covariate in the analysis. The other covariates may affect the means of the methods, thereby explaining a portion of the variability in the measurements. We refer to them as the mean covariates. Incorporating ...