Construct Measurement and Validation Procedures in MIS and Behavioral Research: Integrating New and Existing Techniques

Evaluate goodness of fit of the measurement model Examine significance of the χ (p > .10), CFI (> .95), RMSEA (< .06), and SRMR (< .08). Conduct a simultaneous test of the vanishing tetrads implied by the model (Bollen and Ting 2000). Examine significance of the χ (p > .10), CFI (> .95), RMSEA (< .06), and SRMR (< .08). For models including 2+ reflective indicators, conduct a simultaneous test of the vanishing tetrads implied by the model (Bollen and Ting 2000). Assess validity of the set of indicators at the construct level For first-order constructs (Figure 3, Panel A): Examine whether the average variance extracted (AVE) for the set of indicators is greater than .50 (Fornell and Larcker 1981). For first-order constructs (Figure 3, Panel B): Assess the validity of the set of indicators using Edwards’ (2001) adequacy coefficient (Ra). Values of Ra greater than .50 would mean that, on average, a majority of the variance in the indicators is shared with the construct. Alternatively, if the only antecedents of the composite latent construct are its own formative indicators, the magnitude of the construct level error term could be used to assess validity (Diamantopoulos et al. 2008; Williams et al. 2003). The construct level error variance should be small and constitute no more than half of the total variance of the construct (the smaller the better). For second-order constructs (Figure 3, Panel C): Assess the validity of the set of sub-dimensions using Edwards’ (2001) multivariate coefficient of determination (Rm). Alternatively, the average variance For second-order constructs (Figure 3, Panel D): Assess the validity of the set of sub-dimensions using Edwards’ (2001) adequacy coefficient (Ra). Values of Ra greater than .50 would mean that, on average, a majority of the variance in the