Details

Autor(en) / Beteiligte

• Baumgartner, Hans
• Weijters, Bert
• Pieters, Rik

Titel

The biasing effect of common method variance: some clarifications

Ist Teil von

• Journal of the Academy of Marketing Science, 2021-03, Vol.49 (2), p.221-235

Ort / Verlag

New York: Springer US

Erscheinungsjahr

2021

Link zum Volltext

Quelle

EBSCOhost Business Source Ultimate

Beschreibungen/Notizen

• There are enduring misconceptions in the marketing and management literature about the potential biasing effects of Common Method Variance (CMV). One belief is that the biasing effect of CMV is of greater theoretical than practical importance; another belief is that if CMV is a potential problem, it can be easily identified with the Harman one-factor test. In this article, we show that both beliefs are ill founded and need correction. To demonstrate our key points with greater generality, we use analytical derivations rather than empirical simulations. First, we examine the effects of CMV on correlations between observed variables as a function of measure unreliability and the sign and size of the “true” trait correlation. We demonstrate that, for negative trait correlations, CMV leads to a substantial upward bias in observed correlations (i.e., observed correlations are less negative than the trait correlation), and under certain conditions observed correlations may even have the wrong sign (assuming that the method loadings are both positive or both negative). We also show that, for positive trait correlations, the downward bias due to measurement unreliability does not always mitigate the upward bias due to CMV (again assuming that the method loadings are either both positive or both negative). Importantly, our results indicate that the inflationary effect of CMV is larger at lower levels of (positive) trait correlations, whereas the deflationary effect of unreliability is larger at higher levels of trait correlations. Second, we demonstrate analytically the serious deficiencies of the popular Harman one-factor test for detecting common method variance and strongly recommend against its use in future research.