Partitioning of Data - Spatial Mathematics: Theory and Practice through Mapping - page 129

Table 6.1 Classification of Data

Method Merits Limitations

Natural breaks Big jumps in the data appear at class

boundaries.
Extreme values are visually obvious.
The two merits taken together may produce a
“realistic” view of the data—hence, the
possible suitability for choosing this method
of data partition as a default.

Class intervals are difficult to read.
Clear replication of results may be difficult.
Merging or mosaicking maps will produce different
classifications.
Coloring may need to be adjusted so as not to give
undue visual importance to extreme values.

Quantile Well-suited to data that are linearly

distributed—data that do not have
disproportionate numbers of features with
similar values (“clusters”).
Easy to explain to others how it works.
Useful for making comparisons in relation to the
partition: For example, to show that a
commercial establishment is in the top quarter
of sales of all stores in the region.
Distinctions among intermediate values, grouped
in natural breaks, may be easier in quantile.

Features close in value to each other may lie in
different classes.
Features ranging widely in value may be included
within the same class.
Increasing the number of classes may help to
overcome these drawbacks but that act then adds
clutter to the map.
Clear replication of results is easier than with natural
breaks but still problematic when merging files.

Geometrical
interval

Polygons that are largest in area are in classes
by themselves.

Polygons that are smallest in area are grouped in
classes and distinctions among them may be
difficult to make.

Equal interval Familiarity: A natural legend in terms of ease

in reading (at least when the nature of the
entire range of possibilities is clear, as in
percentages, temperature, and so forth).

Hides variation between features with fairly similar
values.
When the data range does not already make natural
sense, a different classification scheme may be
better.

Emphasis on ranking in relation to the
partition: to show that a store is part of a
group of stores in the top quarter in sales.

Standard
deviation

Easy to visually assess which regions have
values above or below the mean for all data.

The data may skew class count and position in
relation to the mean: Many high values may cause
low values to be grouped in a single class below the
mean and produce multiple classes above the mean,
so that the mean class does not, itself, occupy a
visual central position on the map.

Generally, it helps to graph the data set in advance. Determine if it is lin-
early distributed. Determine if it has clusters near the top or the bottom of
the distribution. Different partitioning methods work well in some styles of
distribution but not in others. Choose a partition that is most appropriate for
your data (Paret, 2012).

6.2.1 Natural breaks

This method is the default partitioning scheme in many contemporary mapping
packages. It identifies breakpoints between ranges using a statistical formula