ERIC Number: EJ1046310
Record Type: Journal
Publication Date: 2014
Reference Count: 15
What Is the Latent Variable in Causal Indicator Models?
Howell, Roy D.
Measurement: Interdisciplinary Research and Perspectives, v12 n4 p141-145 2014
Building on the work of Bollen (2007) and Bollen & Bauldry (2011), Bainter and Bollen (this issue) clarifies several points of confusion in the literature regarding causal indicator models. This author would certainly agree that the effect indicator (reflective) measurement model is inappropriate for some indicators (such as the social interaction or exposure to media violence examples they present) and that to try to model these using a reflective model would result in a misspecified model. He also agrees that clearly distinguishing between causal indicator models and composite indicators is beneficial. However, he disagrees fundamentally with regard to the interpretation of the latent variable (?1 in Figures 1 and 2 of Bainter & Bollen) in terms of its causal indicators, that is, as "the latent variable defined by causal indicators" (p. 133). In this author's view, it is not, and such an interpretation is misleading at best. He states that the latent variable in a causal indicators model is a common factor accounting for the covariance among its outcomes, and that it should not be construed in terms of the content of the causal indicators. If the outcomes included to identify the model are reflective measures such that the factor has a clear interpretation, all is well. The so-called causal indicants are simply predictors, call them what you may. If the outcomes are variables of interest whose covariance lacks a clear interpretation, there is a problem. The latent variable's meaning is indeterminate, but it cannot be inferred from the content of the causal indicators. Doing so can lead to interpretational confounding. Forming a composite with weights not determined with respect to an outcome variable or variables is a possible solution, as is using the causal indicators individually in a model without creating a composite. The use of the causal indicators individually may sacrifice parsimony, but could provide useful information and conceptual clarity (McGrath, 2005).
Descriptors: Statistical Analysis, Measurement, Causal Models, Data Interpretation, Predictor Variables
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: http://www.tandf.co.uk/journals
Publication Type: Journal Articles; Opinion Papers; Reports - Evaluative
Education Level: N/A
Authoring Institution: N/A