Donnerstag, 16. Februar 2012

Multi-level modelling vs. cluster-robust standard errors: Same same, but different?

Presentation by Merlin Schaeffer (WZB) at the CO:STA Colloquium

Social scientists are frequently interested in the importance of the social context for people’s actions, attitudes, interests and so on. We want to know for example whether pupils in small school classes perform better than those in large ones, whether people living in socio-economically deprived neighbourhoods face health hazards, or whether a well functioning and strong economy is important for people’s civic and political participation. Outcome and explanatory factor are situated at two different levels, namely the individual and contextual. In testing respective claims quantitatively, sociologists and political scientists are usually taught to use multi-level modelling. In contrast, economists tend to rely on cluster-robust standard errors. Are these merely different terms for the same thing or are there meaningful differences?



Heisig, Jan Paul, Schaeffer, Merlin and Giesecke, Johannes (2015): "Multilevel Modeling When the Effects of Lower-Level Variables Vary Across Clusters. A Monte-Carlo Comparison of Mixed-Effects Models, Cluster-Robust Pooled OLS and Two-Step Estimation". Available at SSRN:

Angrist, Joshua D. and Pischke, Jörn-Steffen (2009). “Mostly harmless econometrics. An empiricist’s companion”. Pinceton: Princeton University Press: Chapter 8.2, 308-32

DiPrete, Thomas A. and Forristal, Jerry D. (1994). “Multilevel models: Methods and substance”. Annual Review of Sociology 20: 331-357

Special Issue of Political Analysis on Multilevel Modeling for Large Clusters