Advanced multilevel modeling for a science of groups: A short primer on multilevel structural equation modeling

Christ Oliver, Mario Gollwitzer, Eva G. Green, Hewstone Miles, Sarasin Oriane, Katharina Schmid, Wagner Ulrich

Producció científica: Article en revista indexadaArticleAvaluat per experts

22 Cites (Scopus)


A science of groups needs to take different levels of analysis into account since only multilevel perspectives provide a full and realistic picture of processes within and between social groups. A multilevel perspective, however, requires appropriate statistical models. Conventional multilevel regression models suffer from a number of limitations. Among these are the restriction to include only manifest variables and only one Level 1 outcome variable. Moreover, it is not possible to test complex models (i.e., with multiple mediators and outcomes in a single step). In this paper, we introduce multilevel structural equation modeling (MSEM) as a new and promising development within psychological methods that helps overcome the limitations inherent in conventional MLM. MSEM combines SEM with MLM and offers the best of both worlds. Because MSEM allows using latent instead of manifest variables, measurement error can be taken into account. Moreover, the measurement model can be tested on both the within and between-levels of analysis. MSEM enables researchers to specify Level 2 outcome variables and allows the researcher to test complex multilevel models (i.e., simultaneous tests of multiple direct and indirect effects). We illustrate the potential of MSEM using three examples from our own research, provide the corresponding software code as online supplementary material, and discuss important practical issues.
Idioma originalAnglès
Pàgines (de-a)121-134
RevistaGroup Dynamics: Theory, Research and Practice
Estat de la publicacióPublicada - 1 de set. 2017


Navegar pels temes de recerca de 'Advanced multilevel modeling for a science of groups: A short primer on multilevel structural equation modeling'. Junts formen un fingerprint únic.

Com citar-ho