4.1. Unidimensionality and Reliability of Scales for Measuring Brand Relationships, Reputation and Corporate Identity
The first-order model had three factors (trust, commitment, and motivation) and nineteen corresponding reflective indicators, as listed in Table 2
and Table 3
. The goal of most research projects is not just to develop unidimensional and reliable measurement scales, but to build and test theory. To summarize the data in terms of a set of underlying constructs, a factor analysis was conducted. We measured the unidimensionality and reliability of the proposed scales. To measure unidimensionality, we conducted principal component analysis with varimax rotation and Kaiser normalization to each scale. The scale items that did not show factorial stability were candidates for elimination. To measure reliability, we selected Cronbach’s alpha.
Next, we analyze the measures of the brand relationships construct. We start by analyzing Trust, commitment and Motivation. Then we define guidelines and criteria to assess a model for Brand Relationships.
This scale was adapted from Morgan and Hunt
) and Gurviez and Korchia
) and had eight reflexive items. We measured the reliability of the scale defined by the selected items. Cronbach’s alpha was 0.898 (higher than the 0.8 suggested by Nunnally
)). Dekovic et al.
) and Holden et al.
) characterized reliabilities of 0.60 or 0.70 as good or adequate. However, Ping
) stated that higher reliability measures tend to avoid low average variance extracted (AVE) when running the CFA. Regarding dimensionality, the scale was shown to be unidimensional, with an explained variance of 59.213 percent extracted by that component.
This was a new scale proposed for this research and consisted of four reflexive items. Regarding the reliability of the scale, Cronbach’s alpha was high (0.819). We then analyzed the dimensionality of the scale and found that the scale was unidimensional, with an explained variance of 64.898 percent by that component.
This was also a new scale proposed for this research and consisted of seven reflexive items. Assessing the reliability, the Cronbach’s alpha was high (0.886). Analyzing the dimensionality, we found that the scale was unidimensional, with an explained variance of 60.417 percent.
Results regarding the other constructs (external brand identity and brand reputation), the initial measures, the analysis of the dimensionality of reputation, and the final research measures are summarized in Table 2
and Table 3
. More information regarding the technical procedures can be provided on request.
4.2. Guidelines and Criteria to Assess Model for Brand Relationships
We used the following guidelines:
where CMIN/DF = Chi-square value/degrees of freedom, CFI = comparative fit index, RMSEA = root mean square error of approximation.
Following these guidelines, we applied the first-order measurement model to the brand relationships concept. A summary of the psychometric properties for the first-order constructs is provided in Table 4
. Discriminant validity was tested, and after dropping item T4, no problems were reported, as can be seen in Table 5
. Taking these results into account, we tested the second-order model for the brand relationships construct. The results showed robustness regarding the selected indicators (see Table 6
We assessed the reliability and validity of the second-order factor for the brand relationships construct. Construct validity is demonstrated by plausible correlations of the second-order construct with first-order indicators, whereas convergent validity can be suggested by an AVE for the second-order construct that is greater than 0.5 (Bagozzi et al. 1991
; Ping 2004
The values of CR = 0.87 and AVE = 0.68 are greater than the recommended values, suggesting higher reliabilities and convergent validity for the second-order construct. In line with this, we can conclude that the results support the first hypothesis (H1) and state that the constructs of trust, commitment, and motivation are a part of a higher dimension construct of brand relationships.
4.3. Model Evaluation
The first analysis of the proposed measurement model suggested that the item Rep2.4 (innovation) be dropped. We re-calculated the reliability and unidimensionality of the scale and found the following for the new three items. The brand reputation scale had a Cronbach’s α of 0.777 (higher than the threshold of 0.7 defined by Bland and Altman 1997
; DeVellis 2003
; Nunnally 1978
; Nunnally and Bernstein 1994
) and a percentage of explained variance of 68.481 percent, which is highly acceptable. In line with these findings, we re-specified the model and conducted CFA again. The results are summarized in Table 7
These fit indices were satisfactory according to the selected guidelines. This means that the second-order construct named brand relationships was related to the second-order corporate brand identity construct (external part) and to the brand construct reputation formed by three measures.
An analysis of all loadings showed that all except one were higher than the threshold of 0.5. The “physical” dimension was the exception; it contributed poorly to the external part of the corporate brand identity construct (0.420 < 0.5). Even so, the model fit was satisfactory. We can conclude that, in contrast to what Kapferer
) suggests, the used sample did not greatly value the physical dimension of corporate brand identity (external part). This is consistent with the sample, which was composed of goal-oriented engineering students. They demonstrated that they assign more value to the dimensions reflecting consumer (loading: 0.784) and relation (loading: 0.750), because they believe that these dimensions are more connected with their lives as students and future professionals. The reflected consumer dimension (the one with the highest loading) was strictly connected with the aspirations of students. However, this finding should be further investigated in other contexts, using other samples. The following standardized residual values also deserve further attention: 2.629 between F4 and Rep2.2; 2.731 between R5 and C3; and 2.716 between R5 and C2.
Rep2.4 (innovation), that was immediately deleted because it had a high standardized residual. Rep2.2 (network performance) also had a relatively high standardized residual, yet we had to maintain one of them because CFA demands at least three items to run an analysis. We considered Rep2.2 more in line with the theoretical background and the factor loadings gave us the same cue (Rep2.2 0.813 vs. Rep2.4 0.685). All the other standardized residuals were below the cut-off point of 2.58, as suggested by Jöreskog and Sörbom
). The other items were a part of other second-order constructs, which were previously analyzed and evaluated and revealed as valid (convergent, discriminant, and nomological). Therefore, considering that the mentioned values were far from the cut-off point of 4.0 (Hair et al. 2006
) and required no further considerations and that the model fit was satisfactory, we decided to keep these items and test the structural model.
Regarding the modification indices, the one between R5 and C3 had a value of 11.588 (>11). This was expected, given the standardized residual value between both items. However, as mentioned above, the difference was very small, and it was decided to keep both items. All other modification indexes (Mis) had values below 11. No problems regarding multicollinearity were found, and no other indices required our attention; with these findings, we tested the structural model.
4.4. Final Structural Model Estimation and Testing
By developing this causal model, we aimed to demonstrate that universities/institutes of higher education need to invest in and select recognized brands for developing relationships, as well as manage the corporate brand identity in the part that is more exposed to interaction with the public.
In the proposed model, the brand relationships construct was an antecedent of the corporate brand identity construct (external part), and the brand identity (external part) was an antecedent of the brand reputation construct. Corporate brand identity (external part) and reputation were latent variables. Consistent with Hair et al.
), and James et al.
), we added a parsimony fit index (PCFI) to the analysis. We selected PCFI because it represents the result of applying James et al.’s
) parsimony adjustment to the CFI:
is the degree of freedom for the model being evaluated, and db
is the degree of freedom for the baseline model. Values are between [0–1], and better fits are closer to 1. Table 8
summarizes the indices of fit of the structural model:
As expected, the χ2
was higher than the one calculated with the measurement model, because a recursive structural model cannot fit better (to have a lower χ2
) than the overall CFA. The difference between both χ2
was quite small (727.239 − 726.149 = 1.09), demonstrating that the model was strongly suggestive of adequate fit (Hair et al. 2006
). The loadings, standardized residuals, and modification indices maintained approximately the same values. Regarding the standardized residuals: 2.704 between F4 and Rep2.2; 2.805 between R5 and C3; and 2.787 between R5 and C2.
The problematic items relating to the modification indices are:
These small differences did not require further analysis, because, at this stage, the focus was on diagnosing the relationships among constructs. A good model fit alone is insufficient to support a structural theory. It is also necessary to examine the individual parameter estimates that represent each specific hypothesis (Hair et al. 2006
). Table 9
summarizes the main indicators and conclusions.
Examining the paths among constructs showed that they were all statistically significant in the predicted direction. The path that represented the weight between brand relationships and external corporate brand identity was characterized by βBR.ECBI = 0.652; S.E. = 0.135; βBR.ECBI = 0.876; p < 0.001. This means that the regression weight for brand relationships in the prediction of external corporate brand identity was significantly different from zero at the 0.001 level (two-tailed). The path that represented the weight between external corporate brand identity and reputation was characterized by βECBI.Rep = 1.302; S.E. = 0.260; βECBI.Rep = 0.824; p < 0.001, meaning that the regression weight for external corporate brand identity in the prediction of reputation was significantly different from zero at the 0.001 level (two-tailed).
We analyzed the variance explained estimates for the endogenous constructs in Table 10
and found that the predictors of the physical construct explained 17.7 percent of variance. This means that the error variance of the physical dimension was approximately 82.3 percent of the variance of this dimension itself. As for the other constructs, no problems were found. We can conclude that our model supported both Hypotheses 2 and 3. Therefore, the relationships among brands (brand relationships) influenced external corporate brand identity, and later, the brand reputation.
Because theory has become essential in assessing the validity of a structural model, we examined an equivalent model, with the purpose of testing an alternative theory. For the previous model, we dropped the physical dimension, for comparison purposes. In line with these findings, we accepted the second and third hypotheses and concluded that the brand relationships construct influences the external part of corporate brand identity (H2) and that the brand identity influences brand reputation (H3). Therefore, the management of corporate brand identity depends on the investment and selection of strong relationships with reputed brands, to attract students and increase brand reputation.