Ecological Fallacy

Definition

The failure to distinguish between a correlation where the unit of analysis is indivisible (for example, an individual person) and one where the statistical object is a group or area.

Distinctive Features

An ecological correlation exists where the researcher is interested in the prevalence of a particular phenomenon in a group or area. Ecological correlations are often used because data about individuals are not available, and mounting a study to collect such data would be costly. However, the researcher should not then make inferences about the behaviour of individuals based on the ecological data. For example, there may be a positive correlation between levels of unemployment in police force areas and the crime rate in those areas. This is not, however, the same as comparing unemployment and offending amongst a number of individuals. To confuse the two types of correlation is an ecological fallacy. (For further comment on this example see Farrington et al., 1986). While a relationship may be found between the levels of unemployment and the levels of crime in police force areas, it does not follow that if individuals in those areas were sampled a correlation would necessarily be found between levels of unemployment and levels of offending for the individuals in the samples.

Evaluation

The dangers of drawing conclusions about individuals based on ecological correlations were highlighted by a number of studies during the 1930s in the United States. These have been considered in an article in the American Sociological Review which demonstrated the mathematical relationship between ecological and individual correlations (Robinson, 1950). Robinson explains that while individual correlation and ecological correlation both depend on the withinareas individual correlations, they do so in different ways:

The size of ecological correlations is affected by the number of sub-areas. For example, the size of an ecological correlation can depend on whether the unit involved is a county, a city, a district and so on. The size of the ecological correlation between two variables for a county may be affected by the fact that the correlation for the county is based on correlations for local authorities within those counties.

The danger of confusing individual and aggregate data is noted (again by Farrington, 1992) in discussing trends in juvenile delinquency. It is pointed out that whereas a great deal is known about correlates of juvenile offending based on studies of individuals, it is aggregate variables relating to neighbourhoods, such as measures of social disorganization, that are most useful in explaining variations in regional or national delinquency rates. The two should not, however, be confused. Another example can be found in a study in which two American researchers reported that the greater the airtime devoted to country music, the greater the white suicide rate (Stack and Gundlach, 1992). The study was based on data for 49 metropolitan areas in the United States. Although further analysis of the data showed that the association was independent of divorce rates, poverty and gun availability, because the data were based on metropolitan area rates, it does not necessarily follow that individual people who listen to country music more than others are more likely to commit suicide than those who listen to it less. Even were such an individual level correlation to be found, this does not mean that listening to country music causes people to commit suicide; consideration also needs to be given to the possibility that the suicidally inclined find solace in listening to country music.

...