forum.smartpls.com

I am using SmartPLS and a fictitious dataset in a research methods course I am teaching. The means of the item loading estimates from the pseudo samples for some of the latent variables, one in particular, are much lower than the estimate from the original whole sample. The reason is that many of these pseudo sample estimates are negative and many are positive, rather than being all negative or positive as the original estimate. Thus, the means are much lower than the original and closer to zero. Also, the standard errors for the tests are inflated and the subsequent ts for the item loadings are nonsignificant although the items really are significantly related to the underlying factors (I created them that way in the dataset).

Is this a bug or I am I doing something wrong?

Anyway, .

This problem should be diminished if you use a correction option in the bootstrap settings (e.g., construct level changes).

Are your LVs modeled as formative or reflective?