Assessing the Accuracy of Polls for the French Presidential Election: The 2007 Experience

Abstract

Suspicion towards polls in France has come to a maximum after their failure in predicting Le Pen's qualification to the second round of the 2002 presidential election. Despite general satisfaction with polls in 2007, we show that their performance is in fact not better than that in 2002 according to some measures of poll accuracy. Even if the final order of arrival of the top four candidates has been predicted exactly by most polls, this is largely due to the important differences of scores among them. Estimates are still based on quite untouched techniques, and therefore suffer important bias in addition to classical margins of error.

Sondages d’opinion : mesurer l’incertitude

“…On notera d'ailleurs que, bien souvent, les critiques les plus virulentes de sondages ne sont pas construites autour d'une discussion des fondations méthodologiques de ces études mais sur les usages qui sont faits de ces études (par exemple Champagne 2007). Le paradoxe pourrait en effet être inversé, en posant la question de savoir comment les sondages font « aussi bien » quand les considérations théoriques pointent vers autant de sources possibles d'erreur (Sauger 2008). Il faut rendre ici autant hommage à l'expertise des gens du métier qu'à la nature humaine, faite, au-delà des revendications de singularité de chacun, de biens des régularités.…”

Support for Le Pen in France: Two Elections in Trompe l'œil

“…Obstinately high unemployment, protests over government reforms, political corruption scandals, the failed European Union constitutional treaty referendum and a national state of emergency prompted by urban riots provided a backdrop of deepening political disaffection (Shields, 2006). Final polls in spring 2007 credited Le Pen with up to 16.5 per cent, a higher forecast than for previous presidential elections in which he had exceeded poll predictions by some margin; and as late as February 2007, popular support for his ideas was recorded at fully 32 per cent – one in three respondents, the highest level yet seen for agreement with Le Pen (Sauger, 2008, p. 122; Mayer, 2007, p. 429).…”