The Relationship between Self-Concept and Brand Personality

Research Model Test

The Partial Least Squares approach is chosen because the distribution of data is not normal; as one of the reasons for using the PLS method is the non-normal distribution of data. Fortunately, PLS-SEM is less stringent when working with non-normal data because the PLS algorithm transforms non-normal data in accordance with the central limit theom. When applying PLS-SEM, researchers need to follow a multi-stage process which involves the specification of the inner and outer models, data collection and examination, the actual model estimation and the evaluation of results. In the following, this study centers around the three most salient steps: (1) model specification; (2) outer model evaluation; and (3) inner model evaluation.

First of all, it is necessary to show the results of the model implementation displaying standard coefficients and significant coefficients. These coefficients are shown in Figures 2 and 3.

Examining the Fit of Measurement Models (Outer Model Evaluation)

When assessing reflective outer models, researchers should verify both the reliability and validity. The first step is using composite reliability to evaluate the construct measures internal consistency reliability. While traditionally assessed using Cronbach's α, composite reliability provides a more appropriate measure of internal consistency reliability for at least two reasons. First, unlike Cronbach's α, composite reliability does not assume that all indicator loadings are equal in the population, which is in line with the working principle of the PLS-SEM algorithm that prioritizes the indicators based on their individual reliabilities during model estimation. Second, Cronbach's α is also sensitive to the number of items in the scale and generally tends to underestimate internal consistency reliability. By using composite reliability, PLS-SEM is able to accommodate different indicator reliabilities (i.e. differences in the indicator loadings), while also avoiding the underestimation associated with Cronbach's α.

In Table1, the Cronbach's alpha and the composite reliability coefficient are shown. The Cronbach's Alpha coefficient of all research structures is higher than 0.7 and the reliability of the structures is confirmed. The composite reliability coefficient of all research structures is higher than 0.708, so the combined reliability of the structures is also confirmed. The Cronbach's alpha and CR coefficients are displayed in Table1.

The second step in evaluating reflective indicators is the assessment of validity. Validity is examined by noting a construct's convergent validity and discriminant validity. Support is provided for convergent validity when each item has outer loadings above 0.70 and when each construct's Average Variance Extracted (AVE) is 0.50 or higher. The AVE is the grand mean value of the squared loadings of a set of indicators (Hair et al., 2014) and is equivalent to the communality of a construct. Put succinctly, an AVE of 0.50 shows that the construct explains more than half of the variance of its indicators.

The average coefficient of extracted variance for all research structures is higher than 0.5, so the convergent validity of the structures is confirmed. The AVE index is shown in Table1.

Discriminant validity represents the extent to which the construct is empirically distinct from other constructs or, in other words, the construct measures what it is intended to measure. One method for assessing the existence of discriminant validity is the Fornell & Larcker criterion. This method states that the construct shares more variance with its indicators than with any other construct. To test this requirement, the AVE of each construct should be higher than the highest squared correlation with any other construct. Table 2 shows the Fornell-Larcker method.

According to Table 2, the second root of the AVE coefficient for each element is higher than the correlation values of that structure with other structures, so divergent validity exists for all components.