I have been trying really hard to find some sources that can help me develop the first-order reflective second-order formative (Reflective-Formative) model in AMOS. Unfortunately, I am unable to find any useful source which can help me. Kindly suggest if the model in Figure 1 can be developed in AMOS. Entrepreneurial in Figure 1 is the dependent variable with first-order reflective and second-order formative. The variables on the left are the independent variable. I have an idea that a reflective-formative model can be developed in AMOS as some papers suggest doing so, but how can we do it?
Any help will be much appreciated.
Hello Chaudry,
Given your research aim, you'd likely have an easier time by moving away from ML/covariance estimation models to partial least squares methods, as used in software packages like:
semPLS in R (freely available software)
SmartPLS (commercial software)
Good luck with your work.
Hello Chaudry,
Reflective factors assume the measures of latent variables (factors) depend on the latent variable (or factor). Reflective factors are defined by the common variance of their dependents, not the predictors. The formative indicators are just predictors – they do not form or define a common factor, but rather a composite variable. Both are used, but carry heavy assumptions about how the variable works with the measures.
A second-order formative variable is quite valid, but only if your variable is a) formed by the measures (first-order factors, in your case) and b) a good proxy for the conceptual variable (Entrepreneurial). Thus a formative composite must be defined by the observed variables, which must be defended theoretically (or commonly accepted past practice).
SES is a common formative composite defined by such elements as parents’ education, income, and job (there are other elements sometimes added in). The level of SES does not cause the indicators. Ken Bollen refers to reflective indicators as a “type of indicators that are more accurately viewed as influencing the latent variable. An example that I developed and have written about is the amount of social interaction of a person as the latent variable and the time spent with friends, time spent with family, time spent with coworkers, etc. as causal indicators of the amount of social interaction. In other words, these indicators affect social interaction, not the reverse.” This is the type of argument you will need to make in order to defend your use of formative modelling for “Entrepreneurial”. The contrary view that a reviewer will likely take, or ask you to respond to, is … shouldn’t info gathering be reflective of entrepreneurial? In other words, if someone scores high on “Entrepreneurial”, shouldn’t this cause them to be more likely to gather information?
I have added some references below, which may be helpful.
Wishing you well in your research,
Tim
Edwards, J.R. (2011). The fallacy of formative measurement. Organizational Research Methods, 14(2), 370-388.
Bollen, K. A., & Diamantopoulos, A. (2017). In defense of causal-formative indicators: A minority report. Psychological Methods, 22(3), 581–596.
Hardin, A. (2017). A call for theory to support the use of causal-formative indicators: A commentary on Bollen and Diamantopoulos (2017). Psychological Methods, 22(3), 597-604.
Bollen, K. A., & Diamantopoulos, A. (2017). Notes on measurement theory for causal-formative indicators: A reply to Hardin. Psychological Methods, 22(3), 605-608.
See also this special issue of Measurement: https://www.tandfonline.com/toc/hmes20/14/3?nav=tocList
David Morse One of the conditions of developing the reflective-formative model in SmartPLS is applying consistent PLS on the model to check the convergent and discriminant validity as both the exogenous and endogenous constructs are reflective. Unfortunately, consistent PLS requires extremely highly reliable data which in real cases is difficult to achieve. I initially did it in SmartPLS, but later I shifted to AMOS. The bootstrap never gave me any results even the composite reliability of constructs was above 0.7 and AVE was above 0.5.
- Badges
- Science topic
- Similar topics
- Papers