Abstract
The samples of older people produced for medical research are affected by significant biases: under-representation of the oldest sector of the population (over 80-year-olds), individuals who are isolated, sick, and those who are not very active or affected by a decline in their cognitive performance. The analysis of the three main surveys conducted in France during the 2002, 2007, and 2012 presidential elections makes it possible to establish that the samples used in political science have similar biases. Our study demonstrates that the older people are, the stronger the selection bias affecting the composition of samples. As a result, the over-80s segment is not only significantly under-represented overall but also has all the characteristics of a statistical artifact. The samples from that age band tend to be made up of individuals who are more educated, who come from the higher socioeconomic categories, and who almost always vote, whereas this segment is in reality by far the most likely to abstain from voting. These sampling problems have become a major issue for the study of political behavior, since the over 80-year-olds now represent 10% of registered voters.