2000115-Strengthening-Communities-with-Neighborhood-Data

370 Strengthening Communities with Neighborhood Data
blanket approaches that generally do not take spatial considerations
into account. As such, place-based approaches recognize so-called spatial
effects. That is, problems related to poverty, youth unemployment,
housing shortages, or school quality vary by region (spatial heterogeneity),
and even within regions, they cluster in particular neighborhoods
(spatial dependence).
Some Key Spatial Concepts
for Place-Based Evaluation
Modifiable Areal Units and Target Area Boundaries
Data analysis for place-based evaluations is usually based on data for
observations summarized at the neighborhood and target area levels,
often because Census data and other data are only available in aggregate
format or because it can be easier to visualize patterns for known areas.
This practice makes the evaluation of place-based initiatives vulnerable
to MAUP (Openshaw 1984). The boundaries of areal units are modifiable
for two reasons: multiple scales (it is often unclear how many zones
to use) and aggregation (it is equally unclear how to group the zones).
For instance, housing parcels could be grouped into zones at a large
number of scales such as (from presumably smallest to largest) census
blocks, block groups, tracts, Zip Codes, perceived neighborhood boundaries,
housing submarkets, and so on. How to group these zones into
target areas for place-based initiatives is also associated with ambiguity.
This uncertainty is problematic because descriptive and statistical
results (including spatial statistical results) are likely to vary significantly
depending on which scale and aggregation zones are used (gerrymandering
is a classic example of how election outcomes are influenced by how
households are aggregated). In other words, outcomes are likely to change
depending on which and how many zones households are assigned to. In
a classic quantitative analysis, Openshaw and Taylor (1979) concluded
that the size of a correlation coefficient expresses a relationship between
variables of interest that changes with scale and aggregation levels, resulting
in “a million or so correlation coefficients.” In other words, correlation
coefficients are modifiable with the areal unit. Specifically, their size tends
to be inversely related to the number of zones; that is, they often increase
as the number of geographical areas decreases (Openshaw 1984).