Outer Weights Initialization and Pre-specification
When you run the PLS or consistent PLS (PLSc) algorithms, SmartPLS uses a value of +1 to initialize all outer relationships (default) in the PLS path model. This initialization supports the common a priori assumption that all indicators of a measurement model are (equally) relevant and that they have a positive relationship with their construct.
However, in some special situation, you may want to use pre-specified outer weight to initialize PLS or consistent PLS (PLSc) algorithms. For example, you may assume that a measurement model’s outer relationships have positive while others have negative signs. To avoid unexpected sign changes, you may select an indicator per measurement model, which is dominant over the other indicators. For this specific indicator, you should have particularly large confidence (e.g., based on prior research) that it has a significant and relatively strong outer weight and you should be very sure whether it has a positive or negative relationship with the construct. In that case, you would choose a +1 or -1 for the outer weight of the dominant indicator and zero for all other indicators of the construct’s measurement model. When running the PLS or consistent PLS (PLSc) algorithms, SmartPLS uses these pre-specified outer weights to initialize the algorithm.
In SmartPLS, you can pre-specify the outer weights in the dialog that includes the settings of the PLS and consistent PLS (PLSc) algorithm. At the bottom of the dialog, click on the initial individual weights hyperlink.
Then, a dialog with all initial values of the outer weights appears. The default value is +1.
You can manually change the initial value. For example, in the corporate reputation model example, if you assume that qual_6 represents the dominant indicator of the latent variable QUAL (i.e., it has a significant relationship which you assume to be positive in this example), then select an initial outer weight of +1 for qual_6 and zero for all other indicators of the construct QUAL (i.e., the seven indicators qual_1 to qual_5, qual_7, and qual_8).
Note: No matter which kind of initialization you choose, the absolute values of estimated coefficients in the PLS path model should be identical.