Micro-macro multilevel analysis

Interfaces brms
1

Hi all,
ithere is someone who has experience to use brms (or other interfaces) for micro-macro multilevel analysis?
I mean that multilevel analysis that allows to predict group-level outcome variables from variables measured at the individual level.

here some references:
-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

  • Bennink, Margot, Marcel A. Croon, and Jeroen K. Vermunt. “Micro–macro multilevel analysis for discrete data: A latent variable approach and an application on personal network data.” Sociological Methods & Research 42.4 (2013): 431-457.

There is an R package names “micromacromultilevel” , but it only handles continuous response variables, and I need to predict a dichotomic outcome.

Thanks a lot
Massimo

2

I’ve only just glanced at this a bit, but I suspect that if you include the standard deviation of the mean score of any within cluster-varying covariates and use brms’s me() function in the group-level model, you would probably deal with the concern raised in C & vV. However, it might be worth noting that Foster-Johnson & Kromrey suggest the C & vV approach appears to perform no better than using a standard group level analysis using group means of the lower level variables with robust standard errors. In that case, a glm with the sandwich package would probably be enough (or something that does robust se automatically).

In Stan proper (and possibly brms in the future), if you really want to do the C & vV approach more explicitly, you could code the latent variable as part of the model, feeding the estimates (latent variable scores) from that lower level part to the group level model. If I was going that route, I would maybe just do the lower-level variables as mixed model (rather than as a factor analysis, but either way is fine), and use the estimated random effects, which are the latent variable in the C & vV approach, in the group level analysis.

3

thank you very much for your suggestion. I’ll think it over
Massimo