Latent variable path modeling with partial least squares

From MaRDI portal





The monograph emphasizes model construction by formulating linear stochastic relationships within unobservable latent variables (LVs) and between LVs and observable manifest variables (MVs). If \(\eta\) and \(y\) represent the LVs and MVs as deviations from mean values, the model may be written in the form \(\eta=B\eta+v\), \(y=\Pi\eta+\varepsilon\), where \(v\) and \(\varepsilon\) are error vectors. The elements of the matrix \(B\) are called path coefficients and those of \(\Pi\) the factor loadings. The model is completed by the generalized weight relation \(\eta=\Omega' y+\delta\). The problem is to estimate \(B\), \(\Pi\) and \(\Omega\), which provide the estimates of the other unknowns such as the elements of the variance-covariance matrix of \((v,\varepsilon)\). Chapter 1 deals with some philosophical and basic issues in modelling. The formulation of the model and the estimation of the parameters are discussed in Chapter 2. The method is illustrated with analysis on data on ``home environment and intelligence. Chapter 3 deals with the statistical foundations of Wold's partial least squares (PLS) method of estimation. This is done by a recapitulation of Wold's predictor specification for one relation, reconstruction of the psychometric factor score problem from the PLS theory, unifying several traditional two-block models like canonical correlations and introduction of the split principal component theorem as a basis for reflections on properties of multi-block models. The PLS modelling is extended to contingency tables in Chapter 4, and the differences between the PLS method which is data-structure oriented and the maximum likelihood (ML) method which is covariance-structure oriented are discussed in Chapter 5. The LV modelling is extended to three-way classified data in Chapter 6. The final Chapter 7 gives information on computer programs for the application of the PLS method. The author has done an excellent job in putting together the various developments that took place in LV path modelling with partial least squares during the last 20 years. An attractive part of the book is the discussion of live examples explaining the method of estimation and interpretation of results. The monograph is addressed to both theoretical and practical research workers. It should be of interest to all those involved in multivariate data analysis.




Cited in
(68)


Describes a project that uses

Uses Software






This page was built for publication: Latent variable path modeling with partial least squares

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3995608)