During the past decade, there has been growing research
interest in the impact of neighborhood of residence on
health (1). Most studies have used the multilevel modeling
technique (2). With this approach, standard errors for the
measures of association between neighborhood factors and
health are corrected for the nonindependence of individuals
within neighborhoods (3, 4), and measures of variation
based on random effects (e.g., neighborhood-level variance)
allow quantifying the magnitude of variations in outcomes
among neighborhoods (5–8). However, our hypothesis was
that the multilevel approach may provide only limited information
on the spatial distribution of outcomes, both when
modeling variations and investigating associations, since it
fragments space into administrative neighborhoods and
ignores spatial associations between them.
This methodological question was motivated by an epidemiologic
investigation of spatial variations in mental disorders
in Malmo¨, Sweden, using data on all 65,830 residents
aged 40–59 years in 2001 geocoded at their exact place of
residence. Several previous studies that investigated neighborhood
variations in mental health as a general category
reported only weak variations between neighborhoods (9–
12). Such variations were usually explained by differences
in neighborhood composition (9–11), but some analyses