of the prevalence of disorders independent of administrative
boundaries. The map in figure 2 shows increased prevalence
in a large area of northern Malmo¨, including two local subareas
of particularly high prevalence. Based on the spatial
smooth term for the 65,830 individuals, the interquartile
spatial odds ratio was 3.96, which approximately quantifies
the odds ratio between individuals in the lowest and highest
quartiles of spatial risk.
The empty multilevel model confirmed that there were
important neighborhood variations in the prevalence of
substance-related disorders (table 1). The Moran’s I statistics
computed from the neighborhood residuals indicated
that spatial correlation between neighborhoods decreased
with increasing distance between them (figure 3). Positive