The ecological fallacy: what electoral maps do not tell us about voters

An electoral map relates proportions calculated by commune, by département or by polling station. It shows that a result is higher in areas where a characteristic is frequent, but it says nothing about the behaviour of the people who have that characteristic.

Moving from one to the other is an inference, not an observation. The reasoning seems natural because the map is easy to read and the link seems obvious, whereas it can be misleading, or even reversed, without anything on the map signalling it.

William Robinson formalised the problem in 1950, in an American Sociological Review article devoted to ecological correlations. In it he compares correlations calculated across American states with the same correlations calculated across individuals.

For the relationship between being foreign-born and literacy, the two levels gave results pointing in opposite directions. Immigrants had settled in states where the rest of the population was more literate, which was enough to reverse the aggregate relationship.

The first reason is composition. An area brings together different groups, and the overall result may come from any one of them. A high vote in a working-class commune may come from the residents who are not working-class.

The second is context. Living in a certain environment can influence everyone, whatever their personal situation. This mechanism is real and interesting, but it is different from the one we think we are reading when we attribute a vote to a group.

Drawing conclusions about areas from individual data raises a symmetrical problem. A sample survey describes people, but it may miss contextual effects, those that stem from place and surroundings rather than from each person's characteristics.

Aggregate data therefore retain a value of their own. They describe areas well and make it possible to follow trends over time. Methods of ecological inference attempt to reconstruct individual behaviour, at the cost of assumptions that must always be made explicit.

The reflex is to rephrase. Instead of saying that a group votes in a certain way, one says that this vote is more frequent in the areas where that group is numerous. The sentence is less striking, but it is accurate.

To know how a group votes, one needs individual data, drawn from surveys, with their own limits. When the two sources agree, the conclusion is strengthened; when they diverge, it is often the sign of a composition effect.

The same caution applies when placing oneself. The political profile of a place says nothing about that of each of its inhabitants, and one can perfectly well think against the grain of one's neighbourhood or region.