Investigating the magnitude and scale of spatial
variations in outcomes
To investigate the spatial distribution of disorders, we
initially sought precise cartographic information, independent
of neighborhood boundaries. Our second objective was
to make inferences about the magnitude and scale of spatial
variations. Beyond knowing whether the magnitude of
neighborhood variations justifies including a contextual dimension
in public health programs (6), it is relevant to assess
the spatial scale on which programs should be
coordinated. To obtain this information, we compared three
modeling approaches that, building on different notions of
space, provided different insights into the spatial distribution
of mental disorders.
A flexible way to obtain precise cartographic information
was to fit a geoadditive model (24, 25). Working with continuous
space, this model was able to process the spatial
coordinates of individuals to produce a smoothed map
of prevalence independent of neighborhood boundaries
(26, 27)—a result far more precise than maps obtained by
using poor locational information at the neighborhood level.
However, this approach provides only visual information