Discriminant validity needs to be established in order to confirm that the hypothesized structural paths results are real and not the result of statistical discrepancies. In PLS-SEM where there are reflective constructs, it is important to assess discriminant validity when analyzing relationships between latent variables. Assessing Discriminant Validity Using Heterotrait-Monotrait Ratio of Correlations (HTMT) Also, if there is a huge discrepancy between the traditional PLS and PLSc results, the researcher should rethink if all reflective constructs truly follow a common factor model, or if they should use a composite (formative) model instead. For example, if a researcher’s model utilizes a higher-order construct, he or she should just use the two-stage approach and not the repeated indicator approach as the latter does not work well with PLSc. There are also other considerations when using PLSc. In other words, if the constructs are modeled as factors, the researcher should use consistent PLS (PLSc) instead of traditional PLS with Mode A. If there is a mixture of reflective and formative constructs: use Consistent PLS Algorithm and Bootstrapping.If all constructs are formative: use PLS Algorithm and Bootstrapping (the original one).If all constructs are reflective: use Consistent PLS Algorithm and Bootstrapping.Which to choose depends on whether the researcher’s model has reflective or formative constructs: The original PLS Algorithm and Bootstrapping functions are still available in the software. In SmartPLS v3, the developers have added “Consistent PLS Algorithm” and “Consistent PLS Bootstrapping” to account for the correlations among reflective factors (see Figure 99). It is also found that the R 2 value of endogenous latent variables is often underestimated (Dijkstra, 2010).īuilding on Nunnally’s (1978) famous correction for attenuation formula, the Consistent PLS (PLSc) is proposed to correct reflective constructs’ correlations to make estimation results consistent with a factor-model (Dijkstra 2010 Dijkstra 2014 Dijkstra and Henseler 2015a Dijkstra and Schermelleh-Engel 2014). Dijkstra & Schermelleh-Engel (2014) argue that it overestimates the loadings in absolute value and underestimates multiple and bivariate correlations between latent variables. The traditional PLS algorithm has its shortcomings. Estimating Factor Models Using Consistent PLS (PLSc)