Construct validity explains how the results are obtained from the use of a measure in accordance with the theories around which the test is designed (Sekaran & Bougie, 2010). The question here is “does the instrument tap the concept as theorised?”. This can be evaluated through convergent and discriminant validity. First, in order to asses if there any problems with any particular items, the loadings and cross loadings were checked. Table 2 shows the respective loadings and cross loadings of the items used in this study. The cut off value of below 0.5 was considered significant (Hair et al., 2010). Thus, any items with a loading of higher than 0.5 on two or more factors, then they will be deemed as having significant cross loadings. From Table 2, it can be confirmed that all items measuring a particular construct loaded highly on that construct and loaded lower on the other constructs thus confirming to construct validity. In this study, two items in innovation and one item in risk taking were dropped due to low loadings (less than 0.5). The next steps are elaborated in the following sections.
The convergent validity can be established when there is a high degree of correlation between two different sources responding to the same measure (Sekaran & Bougie, 2010, p. 327). Thus, it is important that the items share more variance with it’s’ measure than other variables in a particular model. The assessment of convergent validity requires the examination of the average variance extracted (AVE) measure (Fornell & Lacker, 1981). The purpose of the AVE is to measures the amount of variance of the indicator which is accounted by the construct relative to the amount due to the measurement error. Thus, the AVE should exceed 0.5, which is indicating that, more than 50% of the indicators’ variance can be captured by construct (BoRow-Thies & Albers, 2010, p. 596 ). From the Table 3, the AVE values are exceeded the recommended value of 0.5 (Hair et al., 2010) which was in the range of 0.606 and 0.765. The results illustrated adequate convergent validity and unidimensionality, thus, all construct were valid measures of their respective constructs based on their parameter estimates (Chow & Chan, 2008).
The discriminant validity is the complement of the convergent validity. It is indicates the degree to which one construct differs from the others. The square root of the AVE is calculated to determine the construct discriminant validity. Thus, the square root of AVE should be greater than each of the construct correlations (Compeau et al. 1999). As shown in Table 4, the reflective variables of this study fulfil these conditions because the diagonal elements are greater than the off-diagonal elements in the corresponding rows and columns. Thus, it can be concluded that the measurement model demonstrated adequate discriminant validity.
The Cronbach’s alpha coefficient and composite reliability were used to assess the inter-item consistency of the measurement items. The composite reliability takes individual loadings, whereas Cronbach’s alpha assumes priori that each indicator contributes equally to its construct (Barclay, Thompson & Higgins, 1995, p. 297). As shown in Table 3, the alpha values range between 0.795 and 0.924, which are more than 0.6 as suggested by Nunnally and Berstein. Interpreted like a Cronbach’s alpha for internal consistency reliability estimate, a composite reliability of 0.70 or greater is considered acceptable (Fornell & Larcker, 1981). As such, we can conclude that the measurements were reliable. easy payday loans
Table 4: Discriminant Validity of Constructs