The process is illustrated in Figure 7. The first panel shows a number of block
groups plotted with respect to two component scores. The second panel shows a
non-dominated set of block groups (note that there are no other block groups above
and to the right of each of these block groups), which are the most vulnerable
locations. The top-left-most of these block groups has a lower score on component
1 than all other block groups. In some weighting-based schemes where component
1 was weighted sufficiently heavily relative to component 2, this condition may
result in this block group not being judged among the most vulnerable. In the
present scheme, its high score on component 2 means that it is not dominated by
any other block group and ends up in the first Pareto rank of most vulnerable block
groups. With the first Pareto rank of block groups removed from consideration,
a new set of non-dominated block groups is identified in panel 3. This process
continues with each rank being ‘peeled away’ like the layers of an onion until all
block groups have been assigned a vulnerability ranking, as shown by the lines in
the final panel.
While these illustrations are in two dimensions for clarity, precisely the same
logic and procedure can be applied to higher-dimensional data. Note, however,
that as the dimensionality increases (i.e., the number of component scores used to
determine vulnerability increases), the number of cases in each rank will decrease
until, in the most extreme case, all block groups are in the first rank. This situation
should only occur when a large number of component scores were used to assess
relative vulnerability for a small dataset.
In this study, with 1027 block groups and 3 component scores, block groups were
sorted into 19 ranks. Block-group rank membership showed a normal bell-curve
distribution. The middle ranks each contained approximately 100 block groups,
whereas the very highest and very lowest ranks contain only a dozen block groups
or less.
To assess overall social vulnerability, the 19 Pareto ranks were reassigned such
that the most vulnerable block groups had a score of 19 and the least vulnerable
block groups had a score of 1. The social vulnerability score of each block group
was then defined as its Pareto rank. To increase interpretability, the results were
rescaled from 0 to 1 and overall vulnerability zones were established by sorting the
scores into four equal-interval classes.