Logistics research methods: Employing structural equation modeling to test for construct validity

Journal of Business Logistics, 1999 by Garver, Michael S, Mentzer, John T

INTRODUCTION

The logistics discipline continues to examine the philosophical and methodological concepts of conducting rigorous scientific research.' Researchers suggest that as the discipline matures, rigorous research methods and techniques should become the standard for developing and testing logistics theory.2

The cornerstone of conducting scientific research in the modern positivist paradigm (the predominant logistics paradigm) rests in developing sound theoretical frameworks, followed by rigorous testing of these theories. While testing a priori theory with hypotheses represents a small segment of logistics research from 1978 to 1993, trend analysis indicates theory testing is becoming more prevalent in the logistics literature.3 Furthermore, researchers are calling for future logistics research to have a stronger theoretical foundation and to focus on theory testing research, while still maintaining relevance to practitioners.4 As this trend continues, logistics researchers will continue to increase the rigor of their research endeavors.

An important aspect of increased rigor is testing for construct validity and many of its subdimensions, including unidimensionality, convergent validity, discriminant validity, and predictive validity. Advancements in structural equation modeling (SEM) make this statistical technique useful in testing construct validity within a single research study. However, research on structural equation modeling applications and techniques is often fragmented and extremely technical.5 As a result, few logistics researchers, to date, are employing this valuable statistical technique to test the concept of validity in empirical research. Given the importance of testing for validity in conducting rigorous theoretical research, logistics research needs to more fully utilize this methodological tool.

Thus, the purpose of this article is to introduce structural equation modeling, specifically the measurement model in confirmatory factor analysis, as a research tool to test for construct validity. This purpose is accomplished by first discussing validity, construct validity, and the sub-dimensions of construct validity and, then, presenting a detailed process of how to empirically test for construct validity within a single research study using SEM. For each step in the process, applications and appropriate empirical standards for these applications are presented. Specifically, this article will discuss, in-depth, how the measurement model in confirmatory factor analysis can be used to test for the following dimensions of construct validity: 1) unidimensionality; 2) reliability; 3) convergent validity; 4) discriminant validity; and 5) predictive validity.

VALIDITY

Validity is the foundation of the logistics research process.6 Mentzer and Flint state, "validity in research is actually a hierarchy of procedures to ensure that what we conclude from a research study can be shared with confidence."7 Validity can be accomplished and tested both within a given research study and across numerous research studies. For example, external validity is defined as the degree to which the research findings can be generalized to the broader population.8 While steps taken within a single research study can improve external validity, external validity can only be achieved over a variety of research studies conducted within varying contexts. The focus of this paper is to test the various aspects of construct validity that can be addressed within a single study.

Construct Validity

Construct validity examines the degree to which a scale measures what it intends to measure.9 Construct validity is comprised of numerous sub-dimensions, all of which must be satisfied to achieve construct validity. These sub-dimensions of construct validity include: content validity, substantive validity, unidimensionality, reliability, convergent validity, discriminant validity, and predictive validity.

Content and substantive validity ask the question: "What is the nature and domain of the construct, and do the specified items intended to measure this construct actually match the conceptual definition and tap into the domain of the construct?" Content validity refers to the degree that the construct is represented by items that cover the domain of meaning for the construct.10 Is the construct adequately represented by the scale of items? Since there is no formal statistical test for content validity, researcher judgment and insight must be applied. For instance, a straight-forward logistics construct, such as units in inventory, can be measured by one question, whereas a more elusive construct, such as customers' perceptions of the satisfaction they receive from a certain level of seller logistics service, may require numerous questions to "tap" the construct.

Substantive validity refers to the theoretical linkage between the construct (also called the latent variable) and its items." Whereas content validity refers to the correlation between the latent variable and its scale of items, substantive validity is the linkage between individual items and the latent variable. Thus, a latent variable must have substantive validity if it has content validity.'2