ABSTRACT
This chapter introduces two procedures maximum covariance (MAXCOV); and maximum eigenvalue—that are conceptually and mathematically similar. It reviews the available options, highlights their implications, and describes what is known about the performance of each procedure under various data conditions. MAXSLOPE examines the local regression curve in a scatterplot of the association between two indicators, whereas MAXCOV examines the conditional covariance between two indicators, traditionally within nonover-lapping subsamples that have been ordered along a third indicator. Curve shapes can be difficult to interpret, and latent parameters can be difficult to accurately estimate when one proceeds on the basis of curves containing a small number of data points. The chapter describes traditional features of each procedure and demonstrated the ways in which it is typically implemented. It demonstrates the potential advantages of using a composite input indicator—one constructed by summing across all variables not serving as output indicators—instead of using individual variables in the input role.
