Résumé
Conducting surveys on older adults, and even more so on the oldest old (over 80) raises specific problems of representativeness. It is thus likely that our knowledge of very old-age adults is usually based on seriously biased samples, thereby masking some of the consequences of aging on their behavior, even as the proportion of oldest old (over 80) is fast growing in the French electorate: from 5% in 2002 to 10% in 2012. In this paper, we will present and explain these selection biases as well as reporting biases that are specific to the oldest old age group.To do so, we draw on the analysis of very large sample bases produced in the framework of surveys pertaining to the last three presidential elections in France between 2002 and 2012. We will perform secondary analyses of the 5 panel studies that were conducted in the field between 2002 and 2012, plus a ‘barometer’ tracking opinions in the 2012 election. These are the French equivalents of the American National Election Study in the United States and the British Election Study in the United Kingdom. We will show that such studies have a high risk, both in terms of representativeness and of self-reporting biases, when they try to survey the oldest segments of the electorate.We will examine the sampling strategy of all the major electoral surveys that were carried out in France over the last 10 years. How do they take into account voters who are over 65 years old and, among them, those who are over 80? In order to do so, the paper will examine the composition of the respondents in the “French Electoral Panel Studies” of 2002 and 2007 (managed by CEVIPOF-Sciences Po, Paris) and in 2012 by the TRIELEC study. To identify biases in terms of representativeness, these samples are compared to the INSEE (national statistics institute) “Participation Studies,” which include 118,308 registered voters between 2002 and 2012 and which monitor their actual voting (or non-voting) behavior, thanks to the official electoral registers. As the INSEE survey is based directly on the electoral registers, it carries no risk of selection or reporting bias. Its exceptional size makes it possible to define a representative subsample of 8350 individuals aged over 80. Using this database, it is possible to identify precisely the biases that affect the over 80 age group in major election surveys.This assessment will focus on the classic category of the “65+” in French surveys as a means to study older adults’ opinions. In this paper we show that this broad overarching category is not a relevant way to assess the diversity of old people and does not correctly represent this segment whose weight is becoming more and more important in our aging societies. For the three samples (2002, 2007, 2012), we check the representativeness of age distribution focusing on people over 65 and over 80 years old, level of education, gender crossed by ages, former professions of retired people, marital status and turnout at the first round of the presidential election. First, we demonstrate that people over 65 are under-represented in the samples compared to our control data (age distribution of the list of electors). Moreover, within this broad “over 65” age group, the “65-74” are too numerous at the expense of the “over 80” who are dramatically underrepresented. As a consequence, not distinguishing between 65-79 and over 80 leads to over-representing the most regularly voting part of the French electorate (65-74) and to under-representing the largest non-voting segment of the French electorate (over 80). Second, we show that graduates are over-represented in the samples, whereas people below high school diploma level are under-represented. But this bias is rising with aging to reach a peak among people over 80. In the same way, the group of retirees is affected by important bias of composition with an over-representation of upper classes and a dramatic under-representation of blue collars and employees. Last but not least, we establish that the self-reported rate of participation in the 3 polling surveys is significantly overestimated and that these selection and reporting biases reach their peak among the oldest olds. This result suggests that the (very) old display a higher propensity to refuse to disclose altogether or to misreport their opinions, political preferences and actual behavior. It confirms that survey samples considerably over-represent the most socially and politically active fraction of the oldest olds.Overall this paper, assessing the 3 largest quantitative studies undertaken in France over the last 10 years, we demonstrate that our understanding of elderly political behavior suffers from an in-built artefactual dimension. Consequently, surveys construct an incomplete and partially erroneous image of the political behavior of the oldest old adults at a time when they have become a key component of our democracies.