Social Science & Medicine 62 (2006) 1215–1218 Short Report Multi-level modelling in health research: A caution and rejoinder on temporally mismatched data Michael Buzzelli à , Jason Su Department of Geography, University of British Columbia, 1984 West Mall, Vancouver, BC, Canada V6T 1Z2 Available online 18 August 2005 Abstract In a review of the multilevel modelling literature (MLM) we find that data on individuals and their social environment contexts (neighbourhoods, municipalities) are often drawn from different years/time periods. This temporal mismatch has scarcely attracted any attention though it can significantly influence modelling results and interpretation. We demonstrate the influence of temporal mismatch first by outlining the degree of neighbourhood mobility in large metropolitan areas in Britain, Canada and the United States and second with a brief MLM example. We conclude that researchers ought to provide more study context when such mismatch is unavoidable. r 2005 Elsevier Ltd. All rights reserved. Keywords: Multi-level modelling; Temporal mismatch; Social environments Introduction The growth of multilevel modelling (MLM) reflects, and reinforces interest in the individual health effects of different types of social context, including neighbour- hoods. But a problem arises when neighbourhood data do not correspond temporally to data on individual subjects: this mismatch requires that we assume, rather tenuously and often uncritically, stable neighbourhood populations for deriving contextual variables. The purpose of this paper is to issue a caution and rejoinder on MLM research that relies on temporally mismatched data. This paper first highlights the pervasiveness of temporal mismatches in the MLM—health literature. We then turn to an analysis of neighbourhood census data for metropolitan areas of Canada, the UK and US (2000) showing that residential mobility—the problem behind mis-matched data—is high and significantly different between and within countries. Consequently, standard area-level variables also change significantly over short periods. A brief MLM case study is then used to demonstrate the consequences of data mismatch in MLM. We argue that multilevel models must be applied with greater local sensitivity and the issue should at a minimum be acknowledged and discussed in published case studies. MLM context and overview MLM-health research seeks to elucidate the analytical and empirical influence of social contexts over individual health. The premise of MLM is to overcome both the atomistic fallacy of individual risk factor epidemiology and the ecological fallacy of aggregated data; in short, to diminish the impact of cross-level bias in statistical modelling. After controlling for individual risk factors, do higher level contexts bear on individual health, and to what degree? To answer this, individuals—‘composi- ARTICLE IN PRESS www.elsevier.com/locate/socscimed 0277-9536/$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2005.06.056 à Corresponding author. Tel.: +1 604 822 6620. E-mail addresses: buzzelli@geog.ubc.ca (M. Buzzelli), jasonsu@geog.ubc.ca (J. Su).