The TETRAD Approach to Model Respecification

The TETRAD project revives the tetrad analysis developed almost a century ago. Vanishing tetrads are overidentifying restrictions implied by the structure of a model. As such, it is possible to examine a model empirically by these constraints. Scheines, Spirtes, Glymour, Meek, & Richardson (1998) advocate using vanishing tetrads as a tool for automatic model searches. Despite the search algorithm proving to be superior to those from LISREL and EQS in an earlier report, it is argued that TETRAD II, the search program, is still a datamining procedure. It is important that substantive justifications should be given before, not after, a model is selected. This is impossible with any type of automatic, procedure for specification search. Researchers should take an active role in formulating alternative ' models rather than looking for a quick fix. Finally, the tetrad test developed by Bollen and Ting (1993) is discussed with its application for testing competing models or their components formulated in I advance.