forum.smartpls.com

Hello everyone,

I'm using SmartPLS 3 for my master's project. Since I collected data in a geographically broad area I ran the PLS and the bootstrap algorithms for the complete dataset (n tot=249) and also for 2 subgroup (n1=179 and n=76); I also checked for heterogeneity by conducting Multigroup Analysis. Until I had in my PLS model 7 variables pointing at the final endogenous variable everthing worked.

However, once I included 1 and/or 2 more variable(s) to my model (one mediator and one moderator), the boostrap report for the smaller subgroup (n=76) gave me NA values for standard deviation and blank values for T and p values.

Do you think the reason is the sample size of this subgroup? According to Hair, Jr et al. (2014) who cited Cohen (1992), n=76 should work for PLS at 1% significance, mininum R^2= 0.5 and 9 arrows pointing at a construct. Does this work also for the bootstrap analsysis?

Thank you!

There can be potentially many reasons. I would guess that the new variables do not have sufficient variance for one of the subgroups. That can especially happen with moderators that are very similar to your grouping variable. Within the groups you may have a variance of zero for those variables (only the same values) or a very limited variance that leads to zero variance in only some of the randomly drawn bootstrap samples. Variables with zero variance cannot be used in a PLS model.
Your variables might also be highly collinear in the subgroups which could lead to problems in some of the bootstrap samples.

Both problems are indirectly related to sample size as they are more likely to occur if you have only a few observations.