Conferences

Attrition is a regular challenge and feature of panel survey. Belong the inference problem attrition raises substantial question about the evolutions of ordinary citizens regarding electoral choice. The aim of the study is to analyze the process of attrition of panel data, by comparing four French electoral surveys between 1958 and 2007 (conducted by phone and face to face over national elections in France):

* the electoral panel of 1958: referendum and parliamentary elections (Association française de science politique, 1960); * the “French National Election Panel Study, 1967-1969”: parliamentary election, referendum, presidential election (Converse, Pierce, 1986); * The “Panel électoral français (2002) CEVIPOF-Ministère de l’Intérieur” : presidential and parliamentary elections (Cautrès, Mayer, 2004) ; * The “Panel electoral français (2007) CEVIPOF-Ministère de l’Intérieur” : presidential and parliamentary elections (Cautrès, Muxel, 2009).

This comparison is unique and helps to assess the impact of elections types and survey design on panel attrition, but also the impact of the transformation of the general public in term of level of education and generational renewal. The comparison of these four surveys shows two important results which underlines the dialectics of “changes in permanence”:

1. The permanence of the explanation by socio-political characteristics between 1958 and 2007 in terms of political interest, age, level of education. Our results demonstrate that the “usual suspect” of political awareness explain very well those who refuse to integrate the panel (the “Refusers”). But we also discover that level of political sophistication does not play a role for the “Leavers” (i.e. the respondents quitting the panel). 2. The scope of attrition varies a lot between surveys: for example, “Leavers” are much more numerous in 2002 than in 2007. The political climate of the surveys seems a strong hypothesis: invariably since 1958, a current favorable political mood is accompanied by a lower attrition of panel surveys.

Our comparison draws major results and open new solutions for lessening attrition. Along with context and survey design effects, the 2007 survey allows a assessment on the interview effect on recruitment of panelist.