Context, composition and heterogeneity: Using multilevel models in health research

Bringing the individual back to small-area variation studies: A multilevel analysis of all-cause mortality in Andalusia, Spain

“…Our study also gives further evidence concerning the unsuitability of ecological/spatial analyses for investigating contextual effects (Aitkin M & Longford N, 1986;Bullen et al, 1996;Duncan et al, 1993Duncan et al, , 1995Duncan et al, , 1996Duncan et al, , 1998Duncan et al, , 1999Jones et al, 1991;Merlo et al, 2001;Subramanian et al, 2009;Twigg et al, 2000). As expressed by Morgenstern, "Several epidemiologists have recently called for a greater emphasis on understanding differences in health status between populations-a return to a public health orientation in contrast to the individual (reductionist) orientation of modern epidemiology….…”

“…It is today well established that multilevel modelling (Bingenheimer & Raudenbush, 2004;Duncan et al, 1998;Goldstein, 2003;Merlo et al, 2006c;Merlo et al, 2005a, c;Merlo et al, 2005d;Subramanian et al, 2003) is a worthy instrument for the quantitative analyses of place effects on individual health. However, many multilevel analyses performed so far have been mainly focused on the study of associations between contextual variables and individual health, considering the analysis of variance as secondary information (Blakely & Woodward, 2000;Diez Roux, 2008).…”

“…In contrast, others scholars have explicitly concluded that the analysis of variance provides indispensable information for understanding place effects on health (Boyle MH & Willms JD, 1999;Clarke P & Wheaton B, 2007;Duncan et al, 1993;Merlo, 2003;Merlo et al, 2004;Merlo et al, 2009;Riva et al, 2007) Moreover, all around the world, a persistent amount of observational information on place effects is still being obtained from ecological/spatial studies of "small-area variations" , frequently in the form of coloured atlases and disease maps (Benach et al, 2003;Benach et al, 2004;Borrell et al, 2010;Collaboration, 2010;MacNab & Dean, 2002;Middleton et al, 2008;Ocana-Riola & Mayoral-Cortes, 2010;Ocana-Riola et al, 2008a;Pickle et al, 1999;Shaw, 2008;Turrell & Mengersen, 2000). From an empirical perspective, the advantages of multilevel versus ecological regression analyses were clearly identified by the seminal work performed by Aitkins and Longford (Aitkin M & Longford N, 1986) as well as by Jones,4 Duncan, Moon, Subramanian and colleagues (Bullen et al, 1996;Duncan et al, 1993Duncan et al, , 1995Duncan et al, , 1996Duncan et al, , 1998Duncan et al, , 1999Jones et al, 1991;Subramanian et al, 2009;Twigg et al, 2000). We also contributed to this discussion in a previous publication (Merlo et al, 2001).…”

A multilevel study of predictors of student perceptions of school climate: The effect of classroom-level factors.

“…A multilevel modeling technique was selected for the present study because both the data (students nested within classrooms nested within schools) and the hypotheses (the impact of schooland classroom-level factors on students' perceptions) are multilevel in nature (Raudenbush & Bryk, 2002). Single-level models are inappropriate for the current analyses because they assume that regression coefficients apply equally to all contexts (Duncan, Jones, & Moon, 1998;Luke, 2004). In addition, because individuals from the same school contexts will likely have correlated errors, a basic assumption of multivariate regression is violated (Luke, 2004).…”

‘The bigger the better’ – mothers' social networks and child nutrition in Andhra Pradesh

“…Over the past two decades a number of studies have attempted to comprehend the myriad of factors that affect child nutritional status in developing countries 4 -11 . Their results show that child nutrition is associated with both family background 8,12 and the wider environment in which the child lives 13,14 . In India 62 million children are malnourished, corresponding to half of the country's child population 15,16 .…”