The trap … of the approximate numbers
The voting intentions of a sample are not those of the entire population, despite all efforts to make it as representative as possible. Therefore, it has overly accurate figures, overly detailed analysis, and unprecedented voting configurations.
The illusion of precision
Doing a survey involves questioning a fraction of the population and extrapolating their response to the entire population. Distrust, there are many possibilities that the answers obtained from a randomly selected population sample differ from the opinion of the general population. According to the laws of statistics, this is even more likely the smaller the sample.
Thus, with a random sample of 1,000 people – the typical panel of political polls – the actual proportion of voters who want to vote for a candidate is within a range of about 3 points around the percentage obtained, with a 95% probability. If a poll gives 42% of the voting intentions to a candidate, the actual value is probably between 39 and 45%. Therefore, a change in voting intentions of less than 3 points between two consecutive polls means nothing. Suffice it to say that the half-point “won” by Nicolas Sarkozy in the Ifop-Fiducial barometer for Europe 1, Party of Paris and Public Sénat, after his televised speech on January 29, as well as the 3 points “earned” by François Hollande in the same sequence (bottom left), are on the verge of statistical insignificance.
The limits of the quota method
For a poll to better fit the moods of the general population, there is nothing like a sample, as large as possible, randomly selected from the tab that groups all the voter lists. However, this file, managed by INSEE, is confidential. Therefore, the companies surveyed manufacture samples from scratch, the sociological structure of which is supposed to be the same as that of the general population. That is to say, panels respecting the quotas of gender, age and profession, in addition to a stratification in size of the agglomeration and by regions.
From a methodological point of view, nothing to complain about. Except that these quota samples appear to differ from the general population. Indeed, the small number of categories to which the latter falls, for simplicity, does not reflect the real sociological complexity. For example, graduates are generally overrepresented, while the inactive (about a third of the population) form a very heterogeneous category, due to the diversity of their situations. For Jean Chiche, of the Po Science Research Center, “these differences have little effect on voting intentions”. However, it has been found that they indefinitely skew the detailed analyzes that can be extracted from a survey. In short, if a poll “predicts” who will vote for this or that category of the population, distrust is in order.
The inaccuracy of the “repair
“ Pollsters know that extremist candidates’ voting intentions are sometimes barely noticeable. In contrast, undecided respondents tend to position themselves in favor of a candidate main current, thus artificially increasing its rating. Reasons why the results of raw surveys should be “addressed” before they can be published.
To do this, during a poll, the pollsters find out about the previous votes of the respondents. They then compare the answers with the known results of previous polls, in order to correct the answers about a future voting intention. This does not apply, of course, when comparison with previous elections is not possible. As in the case of the Socialist primaries, for which the pollsters greatly underestimated the score of Arnaud Montebourg, in the same way that they did not foresee the collapse of Ségolène Royal. Under these conditions, how can we trust the “announced” scores of Jean-Luc Mélenchon, a candidate from the Left Front, whose political party did not exist in 2007?
The uncertainty of the human factor
Statistical error, quotas, adjustment … There are even more problems for those who want to poll the voting intentions of a population: can we trust the respondents? Thus, we consider the Ifop survey conducted between February 6 and 9, 2012. To the question “If next Sunday the first round of the presidential election takes place …”, the answers were as follows: François Hollande, 31%; Nicolas Sarkozy, 24.5%; Marine Le Pen, 19.5%; Francois Bayrou, 10%; Jean-Luc Melenchon, 8%; Eva Joly, 3%; other candidates, 4%. Magnificent accuracy of the figures! Except that 38% of respondents say they are still likely to change their minds. What bothered the result of the elections compared to those of the polls and seriously relativizes its realism …