Partial Least Squares Modeling in Research on Educational Achievement

This paper contains a discussion of partial least squares (PLS) path modeling with latent constructs as a general method for research on educational achievement. To the extent that such research requires the analysis of comparatively large and complex models under mild supplementary assumptions, PLS is an extremely flexible and powerful tool for statistical model building. The formal specification, estimation, and evaluation of PLS models i s described with special emphasis on the features that distinguish PLS from other methods for path analysis. This specifically concerns distribution-free least squares estimation and distribution-free model evaluation using jackknife techniques.