Greening the competitive advantage: antecedents and consequences of green brand equity

As both constructs, brand attitude and green brand equity, are evaluative constructs concerning the same entity, that is, the brand, we used confirmatory factor analysis models to examine whether they really capture different aspects. We compared a two-factor model, with the two factors brand attitude and green brand image, to a one-factor model with just one overall evaluative factor. Analyses showed that the two-factor model fitted the data very well, χ²(19, N = 358) = 25.11, p < .157, χ²/df = 1.31, CFI = 1.00, RMSEA = .03, 90 %CI_RMSEA = [.00; .06], SRMR = .03, AIC = 59.11, and substantially better than the one-factor model, χ²(20, N = 358) = 533.93, p < .001, χ²/df = 26.70, CFI = .67, RMSEA = .27, 90 %CI_RMSEA = [.25; .29], SRMR = .16, AIC = 565.93. A χ²-difference test indicated a significantly worse fit of the latter model, χ²diff (1) = 508.82, p < .001. These analyses confirm that brand attitude and green brand equity are not the same concept and should be treated as separate constructs.

Including sex, age, student status (yes/no), people per household, and brand category as covariates into the analyses did not change the pattern of results obtained: (non)significant effects remained (non-)significant and significant effects remained significant. Thus, our findings did not depend on the presence or absence of covariates (Simmons et al. 2011).

As our analyses are based on cross-sectional data sharing common method the obtained relations might at least to some extent be a result of common method variance, rather than of associations between the underlying constructs themselves (see for review: Podsakoff et al. 2003). Therefore, within each of the reported mediation analyses this study tested each path (i.e., from predictor to criterion, from predictor to mediator, from mediator to criterion) by using the correlational marker technique (see Richardson et al. 2009) developed by Lindell and Whitney (2001) and recommended by Grant and Campbell (2007) as well as Sendjaya et al. (2008). We used the five dimensions of the BFI-10 (Rammstedt and John 2007) as marker variables which were partialed out from the tested relations. In no case was a regression weight substantially influenced in its size or significance (i.e., in no case did we observe that a previously significant association was no longer significant or vice versa), thus indicating that our mediation findings are relatively unlikely to be explained by methodological artifacts instead of actual associations between the observed constructs.