Lack of method variance in self-reported affect and perceptions at work: Reality or artifact?

Abstract

Spector (1987) recently concluded that there is little evidence of method variance in multitraitmultimethod studies of self-reported affect and perceptions at work. In this article we propose that this conclusion was incorrect and was the result of improper analytical procedures. Specter's data were reanalyzed by using a more powerful approach: confirmatory factor analysis. Model comparisons and variance partitioning indicated that method variance is present and accounts for approximately 25% of the variance in the measures examined by Spector.Researchers in organizational behavior have shown an increasing interest in the effects of method variance on organizational research outcomes (e.g., Podsakoff & Organ, 1986). Most recently, Spector (1987) examined a series of multitrait-multimethod (MTMM) matrixes to determine the extent to which method variance is problematic in research on self-reported affect and perceptions of work. By using data from 10 published MTMM studies, Spector compared the correlations between different traits or constructs that were measured using the same method with the correlations of the different traits measured across methods. This analysis revealed that the first set of monomethod correlations was not significantly different from the second set of heteromethod correlations, and Spector concluded that there was little evidence of method variance.Specter's conclusion is contrary to widely held beliefs that method effects are omnipresent (e.g., Fiske, 1982;Peter, 1981) and findings from organizational behavior research concerning the importance of method variance. For example, Glick, Jenkins, and Gupta (1986) demonstrated through structural equation methodology that common method effects were present that inflated relationships between job characteristics and attitudinal outcomes. Similarly, Wagner and Gooding (1987) used meta-analysis to show that studies based on self-report measures of both participation in decision making and performance ratings yielded higher correlations between these two variables than those studies in which performance was measured by using a technique other than self-reports. Furthermore, Harvey, Billings, and Milan (1985) determined that the inclusion of method factors significantly improved the fit of confirmatory factor analysis models of the Job Diagnostic Survey.We would like to thank the Editor and two anonymous reviewers for their comments on previous versions of this article.

Outcome-based contracts as new business model: The role of partnership and value-driven relational assets

“…Our results reveal that when adding a latent variable that represents common method, model fit improved (χ 2 difference = 8.65, df = 492, p < .01) but the variance accounted for by the common method latent variable was only 5.9% of the total variance. Together, these results suggest that common method bias is not a serious threat to our analysis (Calson & Perrewe 1999;Williams, Cote & Buckley, 1989). …”

Who retires early and why? Determinants of early retirement pension among Danish employees 57–62 years

“…Though the cohort is representative, one should have in mind the relatively small sample size when interpreting the risk estimates. The baseline was questionnaire-based, whereas the outcome was established using register data: this eliminates possible common method variance and the related positive bias (Spector 1987;Williams et al 1989). The study comprises information on both physical, psychosocial work environment exposures, and a number of socio-demographic variables.…”

Managerial self-awareness in high-performing individuals in organizations.

“…This inflation was also evident among importance ratings between self-reports and direct reports, f(477) = 12.49, p < .001. The high correlations between total behavior and importance scores for both self-reports (r = .63) and direct reports (r -.63) indicated some degree of common method variance (Mitchell, 1985;Specter, 1987;Williams, Cote, & Buckley, 1989) between the two sets of ratings as well. In terms of the overall level of managerial self-awareness vis-a-vis congruence between self-reports and direct report ratings, the absolute average profile difference for all managers was 0.82 (SD = 0.32) and a range from 0.32 to 2.13.…”