The use of formative
measurement in the field of Information Systems has increased,
arguably due to statistical tools (e.g., PLS) that can test such
models. However, in the literature, there exist two contradictory
views on the potential deficiency of formative measurement. While
opponents who are critical of formative measurement argue that there
are native weaknesses of the formative approach in model estimation,
proponents who are in favor of using formative measurement counter
that opponents’ research methods in measurement model specification
are flawed. The goal of this work is to empirically test these
opposing views on whether the alleged estimation instability of
formative measurement is due to measurement model misspecification
or simply the shortcoming of formative measurement. To assess the
integrity of arguments of both parties, we adopt a research design
in which four different cases are tested in terms of
interpretational confounding and external consistency. We find that
regardless of whether there is a specification issue, formative
measures can lead to misleading outcomes. Based on the results, we
offer guidelines that researchers may adopt in planning and
executing data analysis with structural equation modeling. Given
that the use of formative measurement is at a critical juncture in
the IS field, we believe that the guidelines in this research note
are important to promote appropriate use of the approach rather than
relegate it to a bandwagon effect.
Keywords: Formative
measurement, formative indicators, measurement models, measurement
instability, external consistency, interpretational confounding,
information systems measures