Reconciling macro and micro perspectives by multilevel models

Abstract

The authors demonstrate "that multilevel models are useful in dealing with clustering, a feature of data that is often present in empirical economics and in the social sciences. If outcome variables are correlated within groups or clusters these models are superior to ordinary regression since they take this correlation explicitly into account. The results from a multilevel analysis of regional wage differences differ from those obtained by OLS. In particular, the estimated standard errors of the coefficients of the aggregate-level variables are considerably larger in the multilevel method. Thus, this method on random coefficient models can help to avoid spurious regressions. The estimate of the affregate-level covariance matrix is of interest because it describes the variation of between-sex differences across the regions." (Author's abstract, IAB-Doku) ((en))

Cite article

Blien, U., Wiedenbeck, M. & Arminger, G. (1994): Reconciling macro and micro perspectives by multilevel models. An application to regional wage differences. In: I. Borg & P. Ph. Mohler (Hrsg.) (1994): Trends and perspectives in empirical social research, p. 266-282.