Page 1 of 1

Using weights from a SmartPLS model to estimate LVs in a second sample

Posted: Thu Jun 21, 2018 8:43 pm
by spangler
Hi all,
I have estimated several SmartPLS models including a reflective dependent latent variable, behavior, and several formative independent latent variables, all of which have are measured with a number of manifest variables. I would like to use the weights from these models to estimate the dependent and independent latent variables in a large sample for which I have manifest variables. Here’s the issue: The small test sample is not a random subset of the large population I want to generalize to. The sample I use to get the estimates consists of two groups (types of individuals) which are fairly homogeneous within and very different between in terms of behavior and personality. I get huge R2s and path coefficients. The large sample for which I want to estimate the latent variables is much more heterogeneous. Can I legitimately use the estimates from the test sample to calculate latent variables for the large sample? Thanks.

Re: Using weights from a SmartPLS model to estimate LVs in a second sample

Posted: Mon Jun 25, 2018 10:56 am
by jmbecker
Well technically it is possible to do that. I think the best source for some problems around these issue is the following article: Shmueli, G., Ray, S., Estrada, J. M. V., & Chatla, S. B. (2016). The elephant in the room: Predictive performance of PLS models. Journal of Business Research, 69(10), 4552-4564.

Whether it makes sense to take the estimates of your sample to generalize to another sample when those samples are not from the same population is certainly questionable. You need to ask yourself how well you expect your samples to share similar characteristics and therefore, the estimates to generalize from the one sample to the other.

Re: Using weights from a SmartPLS model to estimate LVs in a second sample

Posted: Wed Jun 27, 2018 6:55 pm
by spangler
Thanks for your response and for the reference.