I'm struggling to determine whether constructs should be specified as formative or reflective in SEM, especially for higher-order constructs like Subjective Well-Being (SWB). Some researchers treat SWB using dimensions, while others use components.
Reflective almost always makes more sense in my opinion, unless you have an a priori well-defined index that is a weighted or unweighted composite of specific components. For a thorough discussion of formative vs. reflective measurement, see the series of articles in Psychological Methods Volume 12, Issue 2:
Howell, R. D., Breivik, E., & Wilcox, J. B. (2007). Reconsidering formative measurement. Psychological Methods, 12, 205–218.
Bagozzi, R. P. (2007). On the meaning of formative measurement and how it differs from reflective measurement: Comment on Howell, Breivik, and Wilcox (2007). Psychological Methods, 12, 229 –237.
Bollen, K. A. (2007). Interpretational confounding is due to misspecification, not to type of indicator: Comment on Howell, Breivik, and Wilcox (2007). Psychological Methods, 12, 219 – 228.
Howell, R. D., Breivik, E., & Wilcox, J. B. (2007). Is formative measurement really measurement? Reply to Bollen (2007) and Bagozzi (2007). Psychological Methods, 12, 238–245.
See also:
Borsboom, D., Mellenbergh, G. J., & Van Heerden, J. (2003). The theoretical status of latent variables. Psychological Review, 110(2), 203-219.
Edwards, J. R. (2011). The fallacy of formative measurement. Organizational Research Methods, 14(2), 370-388.
Instead of viewing a construct as either reflective or formative it is better to instead focus attention to the relationship of each indicator (whether they be observed or latent) to the construct of interest. From here we wish to determine if a given indicator causes the consturct of interest (often called a causal indicator) or is an effect of the construct (often called a reflective or effect indicator). Often times the best way to do this is to conduct a thought experiment where you can directly intervene on only one of the indicators of a construct. If you imagine that the intervention would only influence the indicator then it is likley an effect indicator, but if you believe that the changes in the indicator would cause (indirectly) changes in the consturct then it is a causal indicator. Another possability is imagine being able to directly manipulate the values of the construct. If the manipulation would indirectly lead to changes of the values of the indicator then it is an effect indicator, but if the manipulation shouldnt have any influence on the indicator then it is a causal indicator. Do this for each indicator and the end result may be a set of indicators that are all effect indicators, all causal indicators, or a mix of both.
A good set of citations have already been provided so I will just add
Bollen, K. A., & Diamantopoulos, A. (2017). In defense of causal-formative indicators: A minority report. Psychological methods, 22(3), 581-596.
Polites, G. L., Roberts, N., & Thatcher, J. (2012). Conceptualizing models using multidimensional constructs: a review and guidelines for their use. European Journal of Information Systems, 21(1), 22-48.
Bollen, K. A., & Davis, W. R. (2009). Causal indicator models: Identification, estimation, and testing. Structural Equation Modeling: A Multidisciplinary Journal, 16(3), 498-522.
Hi, Xiaohan! I recommned you to follow the criteria mentioned by Alejandro Martinez . Additionally, you should asses if correlation between the indicators is expected. In reflective measurement models, high correlation is expected, while in formative measurement models indicators shouldn't be correlated or have a low correlation. The most important criteria to decide the type of measurement is the theory. Because in reflective measurement model, the indicators are explained by the construct, they are its reflections, deleting indicators shouldn't change its content and definition. In formative measurement models it is expected that the chosen indicators explain all the content of the construct, so deleting one indicator would change its meaning.
You can also check the book Primer on PLS-SEM book for more criteria.
Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2022). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM), 3rd ed. Thousand Oaks, CA: Sage.
- Badges
- Science topic
- Similar topics
- Social Science
- Sociology