Choropleth maps, like the one above, use the visual variable of value (aka shade or lightness), sometimes in concert with hue or saturation, to present data about featues. Academic cartographers teach that this symbology should only be used in certain circumstances.
- The phenomenon being mapped must be thought to vary only between enumeration units, and not within them. That is, the phenomenon is observed to change abruptly at enumeration unit borders (abrupt).
- There is little variation in the shape/size of enumeration units (uniformity).
- The data itself must be standardized (divided) by another attribute value of the feature, typically land area or a related variable. (standardized).
- The phenomenon being mapped is continuous, that is, present everywhere (continuous).
Though I think there’s reason to talk about/problematize all four, I’m most interested in the third criterion: standardization. It is taken as gospel in academic cartography that for a choropleth to “work”, the underlying data must be standardized. The need for standardization, as stated by Dent (1996), is because “the varying size of areas and their mapped values will alter the impression of the distribution.” Slocum (2005) backs up Dent, writing that standardization is necessary to “account for varying sizes of enumeration units”.
These same cartographers, though, stress that standardization need not be by land area. Slocum, for example, lists three methods of standardization other than land area (for example, crimes per capita, or really any per capita variable). Using non-area variables is justified by Slocum because “this approach indirectly standardizes for area because larger areas tend to have larger values for both attributes”.
This seems questionable, especially since small enumeration units with high values are a problem standardization was designed to address. My point: dividing a variable (say, elderly population) by a non-area variable (say, total population) does nothing to address the issue of varying unit size on the resultant choropleth map.
It seems to me we’re left with two options.
- We must standardize our data, and it must be by land area, to correct for the varying size of areas on the map.
- We can standardize by any other variable, and indeed abandon standardization altogether by mapping raw totals.
I won’t weigh in just yet, as I could really go either way. Are there other justifications for standardization that don’t require standardization by land area?