Ecological Fallacy

The ecological fallacy arises when inferences are made about the characteristics of individuals in a population based on analyses using aggregate statistics of those same individuals. The general usage of the term applies to inferences made about individuals that are based solely on the group characteristics of that population. Such would be the case with stereotyping or geographical profiling, where individuals are characterized by the general traits of the area in which they live. In a more specific manner, the ecological fallacy occurs when an ecological correlation between variables based on statistical constants such as group rates or means is assumed to characterize individual members of that population. William Robinson first demonstrated that statistical correlations of variables aggregated by group or geographic area can differ markedly from correlations based on individual-level variables of that same population. The possibility of false or misleading associations has significant implications for geographical studies since relational inferences about members of a population are often based on data aggregated by spatial units.

In addition to the ecological fallacy, Hayward Alker identified other forms of fallacy that prevent valid generalizations from being made across different levels of analysis. The individualistic fallacy occurs when coarse-level aggregate relationships are inferred from individual or fine-level relationships. Cross-level fallacy exists when inferences are drawn from one subpopulation to another subpopulation at the same level of analysis. The universal fallacy applies when inferring generalized relationships based on nonrandom subsamples of a population. The selective fallacy typically occurs when a few cases are chosen to prove a universal point. Cross-sectional and longitudinal fallacies infer generalized relationships across many time frames based only on one point in time. Applying to all the above are contextual fallacies, where context or social structure may alter the strength or form of the causal or statistical relationship.

At the root of these fallacies is the concern that inferring generalized causal relationships based on spatially aggregated data can generate erroneous conclusions. Part of the problem is that errors arising from ecological inference are unquantifiable because of the lack of access to individual-level data. The spatial aggregation of data into defined areal units is common in geography and especially so with the increasing availability of digital data sets, digital boundary geographies, and geographic information systems. The Modifiable Areal Unit Problem is a derivative of ecological fallacy. There are many possible combinations of modifiable spatial units, such as voting districts, that can be used to spatially aggregate nonmodifiable units, such as the individual voter. As Openshaw demonstrated, aggregating data into arbitrary spatial units, which often have no meaningful geography, can greatly influence the outcomes of a study. Solving the ecological problem as to how to make inferences about individual behavior based on aggregate data continues to be an active focus of research in the social sciences.