Full-Information Covariance SEM
Partial least squares path methods elicit only pairwise relationships between latent constructs, though they allude to a more complete “plausible” structure. In the 1950s, the Cowles Commission at the University of Chicago and in the 1970s Karl Jöreskog o
- PDF / 221,547 Bytes
- 11 Pages / 439.36 x 666.15 pts Page_size
- 29 Downloads / 184 Views
Full-Information Covariance SEM
Sewall Wright’s path coefficients were conceived as dimensionless binary indicators of whether a genetic trait was passed to an offspring, or not; they were “bits” of information about whether or not a gene was present. Correlations in Wright’s context were almost overkill, though their magnitude might have been considered to suggest varying degrees of confidence in heritability of a trait. As path analysis began to find application in analyzing relationships that were multivalued or continuous, limitations in the ability to resolve effects began to reveal themselves. It was in this context that (Cochran et al. 1954) advocated systematic replacement in path analysis of the dimensionless path coefficients by the corresponding concrete path regressions. In the early days of data processing, both Herman Wold and his student Karl Jöreskog developed software, building on Turner and Stevens’ mathematics, that took advantage of computer intensive techniques becoming available to universities in the 1970s and 1980s. Jöreskog substantially advanced the utility of SEM by creating the LISREL full-information method for SEM analysis, which allowed fit statistics and with proper argumentation, the testing of causal models and hypotheses on those models.
3.1 LISREL Wold’s student, Karl Jöreskog, extended Wold and Lohmöller (Lohmöller 1988, 1989) methods in software implementations of covariance structure path analysis methods. Jöreskog’s LISREL (an acronym for linear structural relations) software was the early trendsetter in computer intensive path model, having appeared in a mainframe form in the late 1970s. Later in the 1980s (prior to the Windows 3.1
© Springer Nature Switzerland AG 2019 J. C. Westland, Structural Equation Models, Studies in Systems, Decision and Control 22, https://doi.org/10.1007/978-3-030-12508-0_3
39
40
3 Full-Information Covariance SEM
graphical interface) Lohmöller released desktop computing software that implemented both Jöreskog’s as well as his own interpretation of Wold’s ideas for path modeling in LVPLS. LISREL followed shortly with a desktop version, and these two packages set the standards for future path modeling software. Lohmöller’s proponents have projected an ongoing animosity towards the LISREL and AMOS approaches. PLS path analysis is to this day considered to be a competitor of Jöreskog’s LISREL approach. Vinzi et al. (2005) complained: “The goal of LISREL (or hard modeling) is actually to provide a statement of causality by seeking to find structurally or functionally invariant parameters, i.e., invariant features of the mechanism generating observable variables that define how the world of interest to the model at hand works. These parameters are supposed to relate to causes describing the necessary relationships between variables within a closed system. Unfortunately, most often real data do not meet the requirements for this ideal.” Whether or not you accept Lauro and Vinzi’s complaints about LISREL (and covariance structure modeling approaches
Data Loading...
