NotesFAQContact Us
Search Tips
Peer reviewed Peer reviewed
Direct linkDirect link
ERIC Number: EJ1036653
Record Type: Journal
Publication Date: 2014
Pages: 6
Abstractor: ERIC
Reference Count: 21
ISSN: ISSN-1536-6367
The Vanishing Tetrad Test: Another Test of Model Misspecification
Roos, J. Micah
Measurement: Interdisciplinary Research and Perspectives, v12 n3 p109-114 2014
The Vanishing Tetrad Test (VTT) (Bollen, Lennox, & Dahly, 2009; Bollen & Ting, 2000; Hipp, Bauer, & Bollen, 2005) is an extension of the Confirmatory Tetrad Analysis (CTA) proposed by Bollen and Ting (Bollen & Ting, 1993). VTT is a powerful tool for detecting model misspecification and can be particularly useful in cases in which causal indicators are suspected. However, the results of VTTs (and CTA more generally) should not be interpreted to be more than they are. VTTs are a type of test of global (or local) model fit, and a variety of model misspecifications can yield a significant VTT. Willoughby et al. (this issue) present VTT results from multiple studies, each with a different sample (notably, with important demographic differences that have substantive implications for the measurand, such as respondent age and sampling frame), as well as models with different indicators. This makes interpretation of patterns of VTT significance across these different studies impossible, particularly in cases in which the number and procedure of indicators varies. Any problems with global model fit may be due to sample-specific issues, indicator-specific issues, model-structure issues, or any combination of these. Thus, attributing the significant VTT results to causal versus effect indicators, rather than sample differences, between-study indicator differences, or to differences in the measurands is premature. The SBIC is both a stand-alone index and a comparative one and assumes that models are nested with the saturated model (Raftery, 1995). While originally designed to approximate a Bayes factor, the SBIC adjusts for the sensitivity of the ?[superscript 2] test in large samples with a penalty for model complexity when the number of indicators is large. In cases in which a model with causal indicators is otherwise identified, comparisons of global fit between models with both causal and effect indicators may be made using the SBIC. In this commentary, J. Micah Roos argues that while Willoughby et al. call attention to the intriguing possibility that the indicators of EF may be of the causal type (or formative) rather than the effect type (or reflective), the evidence is inconclusive and much more work is needed before moving forward with this speculation. Roos recommends a rigorous analysis of nested Vanishing Tetrad Tests (VTTs) (Bollen, Lennox, & Dahly, 2009; Bollen & Ting, 2000; Hipp, Bauer, & Bollen, 2005) and SBIC comparisons of both causal and effect models for EF, using the same sample and identical indicators as a fruitful direction for this line of research.
Psychology Press. Available from: Taylor & Francis, Ltd. 325 Chestnut Street Suite 800, Philadelphia, PA 19106. Tel: 800-354-1420; Fax: 215-625-2940; Web site:
Publication Type: Journal Articles; Reports - Evaluative
Education Level: N/A
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: N/A