Ecological Fallacy

Ecological fallacy can be defined simply as incorrectly inferring the behavior or condition of individual observations based upon aggregated data or information representing a group or a geographical region. These data are often referred to as ecological data, not in the biological sense, but because the data are used to describe the aggregated or overall condition of a region or a community. When the inference on the individuals drawn from the aggregated data is erroneous, the problem is known as ecological fallacy.

This is an important methodological problem among several social science disciplines, including economics, geography, political science, and sociology. These disciplines frequently rely on data collected as individual observations that are aggregated into geographical units of different sizes or scales, such as census blocks, block groups, and tracts, in order to represent the condition within the regions. Several physical science disciplines, such as ecology and environmental science, also rely on these aggregated data, also known as ecological data. It is also a common practice in geographic information science to use aggregate-level data as attributes of polygon features for representation, thematic mapping, and spatial analysis. If individual-level data are used, these data should be able to reflect individual behavior or situations reasonably well. When aggregated data are used to infer individual behavior or conditions, however, there is a great chance in general that the individual situation will deviate from the overall regional situation. Because ecological fallacy is partly about how individual-level data are aggregated to regional data, it is therefore related to the Modifiable Areal Unit Problem (MAUP), which has to do with how different ways of drawing boundaries of geographical units can give different analytical results.

The problem of ecological fallacy is composed of two parts: how the data are aggregated to regional or group level and how the data are used. Most socioeconomic data are gathered from individuals. Due to many reasons, however, such as privacy and security issues, individual-level data are usually not released, but are aggregated or summarized through various ways to represent the overall situation of a group of individuals. The group of individuals can be defined according to socioeconomic-demographic criteria, such that individuals within the group share certain characteristics. The group may also be defined geographically (e.g., within a county or a census tract), such that individuals are in the vicinity of a given location. GIS are often used to handle and analyze such data.

There are many ways to aggregate or summarize individual-level data. One of the most common methods is to report the summary statistics of central tendency, such as mean or median. Examples of these statistics in the U.S. census data include median house value and per capita income of given census areal units. When summary statistics are used to present the entire group, however, most individuals, if not all, have values that are to some extent different from the statistics. Therefore, data at the aggregated level cannot precisely describe individual situations. In GIS, and especially in thematic mapping, summary statistics of central tendency or some other summary measures are used to create maps. Often, these values [Page 120]are assumed to be applicable to all individuals within the areal units, and thus ecological fallacy is committed. Also, how much variation appears within the unit is often not reported or is not a concern for those making or using the maps. If individuals within each group are very similar, however, or the group is relatively homogeneous, then the summary statistics could possibly reflect individual situations quite well.

...