Solving the Interpretational-Confounding and Interpretational-Ambiguity Problems of Formative Construct Modeling in Behavioral Research: Proposing a Two-Stage Fixed-Weight Redundancy Approach
Recently, information systems research has devoted increasing attention to formative measurements. However, current approaches to modeling formative constructs have potential validity problems and thus limited applicability. Here, we highlight two major problems in formative measurement—interpretational confounding and interpretational ambiguity—and propose a novel resolution. Interpretational confounding occurs when using the traditional free-estimation approach, because the weights of different formative indicators vary as the dependent variable changes, resulting in the distortion of the measurement weights of the focal formative construct and thus jeopardizing the generalizability of empirical tests. Another way to alleviate the interpretational-confounding issue is to include the multiple indicators multiple causes (MIMIC) construct in the path model (i.e., MIMIC-path). Unfortunately, this method has led to the second major problem of interpretational ambiguity, the existence of more than one potential explanation of the formative model. More specifically, reflective indicators in the MIMIC model can be viewed as (1) indicators of the MIMIC construct, (2) dependent variables of the formative construct, or (3) indicators of a reflective construct affected by independent variables (formative indicators). To resolve these issues, we propose a two-stage fixed-weight redundancy model (FWRM) approach. We demonstrate the applicability of the FWRM approach with a set of survey data. We conducted a simulation study evaluating the FWRM approach by comparing it with the commonly used free-estimation and MIMIC-path methods. The results indicate that our FWRM approach can indeed improve the validity of formative construct modeling by mitigating confounding and ambiguity issues.