I recommend

https://www.quantpsy.org/pubs/preacher_zhang_zyphur_2011.pdf

http://www.dumay.info/pdf/SEM/7.pdf

I agree with you that SEM is probably the best way to analyze the data.

But as far as I know, a 2 - 1 - 2 mediation using SEM requires building measurement models at the between-level analysis, both for the predictor and for the mediating variable. In your case, I think it would therefore be necessary to have a measurement model for the mediator at both the within- and the between-level of analysis. Then you would be able to show the indirect relationship. Since your mediator appears to be a single manifest variable, you could try to create a measurement model using one indicater. However, you need to specify/fix the factor loading, the residual variance, and the latent variance, otherwise your model will not be identified.

With regard to imputed data, there should be no problem when using them in your models. But your should be able to explain carefully how you have proceeded.

Dear,

Thank you for the question and answers. It will also help me.

Although I knot that the following R package can conduct it, I have to confess that I did not manage to make it works (Lu et al. 2017). But maybe it helps you.

I have another set of material that perhaps helps:

Lu, J. G., Page-Gould, E., & Xu, N. R. (2017). Package “MicroMacroMultilevel” Title Micro-Macro Multilevel Modeling. https://cran.r-project.org/web/packages/MicroMacroMultilevel/MicroMacroMultilevel.pdf

Bennink, M. (2014). Micro-macro multilevel analysis for discrete data. Ridderprint.

Bennink, M., Croon, M. A., & Vermunt, J. K. (2015). Stepwise latent class models for explaining group-level outcomes using discrete individual-level predictors. Multivariate Behavioral Research, 50(6), 662–675. https://doi.org/10.1080/00273171.2015.1074879

Croon, M. A., & van Veldhoven, M. J. P. M. (2007). Predicting group-level outcome variables from variables measured at the individual level: A latent variable multilevel model. Psychological Methods, 12(1), 45–57. https://doi.org/10.1037/1082-989X.12.1.45

Cristian Ramos-Vera Thanks for the papers, but that paper specifically mention that it can't manage level-2 outcome.

"these MLM approaches do not accommodate mediation pathways with Level-2 outcomes and may produce conflated estimates of between- and within-level components of indirect effects."

Hi Hanish;

Implemented and perceived high-performance work system and its effect on branch performance: A 2-1-2 mediational multilevel approach.