forum.smartpls.com

I am relatively new to SmartPLS and am trying to test a mediated model with a number of independent and dependent variables. Most variables are at the individual level, but some independent variables I would like to include are at the company or country level. Reviewers have raised concerns about doing this. Would it be appropriate to include higher-levels-of-analysis variables with those at the individual level, or do all variables need to be at the same level? Any advice and references regarding this would be greatly appreciated.

Multi-level PLS analyses are still under development. You may use the latent variable scores and, then, run multilevel analyses for the partial regression models using Stata, SPSS, Statistica, R etc.

Alternatively, you could use these variables and run a multigroup analysis. The advanced PLS-SEM book has a chapter on this topic:
https://www.smartpls.com/documentation/ ... sem-issues

Best
Christian

Dear Christian,

Many thanks for your prompt response! I'll consider these options and look at the chapter! I look forward to multi-level analysis becoming available.

I've read more about this issue and wonder whether multi-level or multi-group analysis is necessary for me. My variables are not clearly nested (for multi-level) and I am not sure what categories would make sense (for multi-group). My model includes IVs at different levels of analysis, such as perceived prestige of a subsidiary (individual level) and cultural distance between headquarters and subsidiary (arguably national level, but certainly not individual level). Note that the sample includes multiple companies and subsidiaries, so there are different headquarters and subsidiaries for different respondents. All DVs as well as mediators are at the individual, attitudinal level. In this case, would it be appropriate to include all these variables in the same model, or must I only include the individual-level, perceived variables? If it would be appropriate to include all variables (i.e. individual and national level), then references to support this would be helpful. Again, any advice and insight is highly appreciated.

Thanks. I cannot really comment on this. PLS-SEM only uses the full set of data for the model estimation. It is possible to run a moderator or multigroup analysis to account for observed heterogeneity. Also, you can use the latent variable scores (if you run the IPMA, you get the unstandardized latent variable scores) for subsequent (e.g., complementary analyses).

Your research goals and data rather sound like a multi-level analysis.

Best regards
Christian

Thank you for the prompt response and advice, Christian!

Dear Prof. Dr. Ringle,

you mentioned in an earlier post from 2017 that Multi-level PLS analyses are still under development. Hence, I was wondering if there were any further developments in the meantime that would enable me to run Multi-level PLS analyses in SmartPLS.

Best regards

We did not follow-up on this direction. As as work around, you may extract the unstandardized latent variable scores from the IPMA results report and run a "standard multilevel analysis" for the partial regression models.