Our latest report of New Hot Papers in Essential Science Indicators shows that the paper “A new criterion for assessing discriminant validity in variance-based structural equation modeling,” (J. Acad. Mark. Sci. 43 : 115-135 January 2015) is the hottest paper in the field of Economics & Business for the period ending October 31, 2016. Currently, this paper has 111 citations in the Web of Science.
Here, we talk with the corresponding author, Professor Marko Sarstedt, about the paper and its implications for the field of Economics & Business as well as other fields in science and social sciences.
Why do you think your paper is highly cited?
The paper challenges the standard results assessment criterion in partial least squares structural equation modeling (i.e., the Fornell-Larcker criterion for assessing discriminant validity). Instead of just showing that the existing procedure does not work properly, the paper proposes a new criterion, called heterotrait-monotrait ratio of correlations (HTMT), which works extremely well across many research settings. The paper also provides a detailed explanation of how applied researchers can use the new criterion and offers guidance on the steps to follow if it indicates a lack of discriminant validity. Our findings are not limited to business and management, but are relevant to all scientific fields that use partial least squares structural equation modeling, such as biology, medicine, engineering, political science, and psychology, all of which contribute to the paper’s popularity.
Does it describe a new discovery, methodology, or synthesis of knowledge?
The paper describes a new method for assessing discriminant validity in partial least squares structural equation modeling, which is one of the key building blocks of model evaluation. If discriminant validity is not established, researchers cannot be certain that the results confirming hypothesized structural paths are real, or whether they are merely the result of statistical discrepancies. Our new method clearly outperforms classic approaches to discriminant validity assessment, which are largely unable to detect a lack of discriminant validity.
Would you summarize the significance of your paper in layman’s terms?
When estimating statistical models that include latent variables (e.g., measuring unobservable phenomena such as perceptions, attitudes, or intentions), researchers must ensure that a measure is empirically unique, represents the phenomena of interest, and that the other measures in a statistical model do not capture it. For example, from a conceptual perspective, customer satisfaction and customer loyalty are two different phenomena. When measuring customer satisfaction and loyalty in a statistical model, we must ensure that the measures representing these two concepts are also empirically distinct. Our paper presents a new method for testing whether they actually are.
How did you become involved in this research, and how would you describe the particular challenges, setbacks, and successes that you’ve encountered along the way?
The starting point was a response to a paper published in Organizational Research Methods (ORM). In this response, we ran a simulation study on the efficacy of model evaluation criteria in partial least squares structural equation modeling, but discovered that long-established criteria for discriminant validity assessment did not perform well. The biggest challenge was developing a better criterion and confirming its superior performance in an extensive simulation study. The reviewers of the Journal of the Academy of Marketing Science (JAMS) offered very useful guidance to develop the paper further, particularly with regard to our recommendations on how to deal with a lack of discriminant validity. These guidelines certainly contributed to the paper’s success.
Where do you see your research leading in the future?
We will certainly continue our research on partial least squares structural equation modeling. One research avenue deals with model evaluation statistics and ways of disclosing model misspecifications. We also seek to better understand the measurement philosophies underlying the different approaches to structural equation modeling, which have vast implications for the methods’ areas of use. Finally, we work on issues such as endogeneity and common method bias.
Do you foresee any social or political implications for your research?
This paper contributes to the safeguarding of empirical results’ soundness in disciplines that deal with unobservable phenomena. While our study has no immediate political or social implications, it ensures that the implications derived from studies using structural equation modeling are appropriate with regard to discriminant validity. Importantly, our results are not limited to the social sciences, but extend to other fields of research, which frequently use methods such as structural equation modeling in disciplines like biology, medicine, engineering, political science, and psychology.
Marko Sarstedt, Professor of Marketing
Otto von Guericke University Magdeburg
University of Newcastle
Newcastle, New South Wales, Australia