We can no longer count the polls already published, six months before the presidential election. Cascading hypotheses, candidates sometimes not yet declared, different samples... How to navigate them when they are regularly relayed and commented on? A few reflexes allow us to take a step back from this deluge of figures and curves. Nicolas Hubé, teacher at the Center for Research on Mediations (Crem), at the University of Lorraine, discusses the main elements that should alert you.
1 Identify the margin of error
“There is always a fairly large margin of error, of the order of 3 to 4 points, depending on the number of people questioned,” recalls Nicolas Hubé. Differences or variations of a point or two from one survey to another, for example, lose their relevance because they will be included in the margin of error. The risk is therefore to overinterpret the results. “We should even practically classify personalities according to ranges,” estimates the teacher, who agrees with the difficulty of commenting on them. Such a candidate is “between 12 and 15 points, and his rival between 10 and 14, it is much less simple to explain”.
This figure is always specified in the original studies of polling institutes, but much less often in the press articles which relay them.
2 Check sample size
To reduce the margin of error, you need to increase the sample size, that is, interview more people. “We often use the number of 1,000 people surveyed” because the margin of error is “fairly small”, it is “a good compromise, we have a fairly precise figure and it is not too complicated to obtain, nor too expensive”, explains the professor of information and communication sciences. If the survey was carried out with around a thousand people, and we do not know the margin of error, it may be more prudent "to add or subtract three points" to a personality tested, he believes.
In a survey carried out on 300 to 500 people, this margin will logically be greater. “Except, for example, if we have 300 farmers interviewed. Given the population of farmers, we can say that this is starting to be interesting,” notes the specialist.
The sample size is also necessarily specified in the official documents of the polling institutes and appears in principle in the articles.
3 Pay attention to the questions asked
Many polls do not limit their questions to “Who would you vote for?”, but explore certain themes: “Who is the candidate who has the most legitimacy, the most charisma, etc.” Questions that are a bit "tricky", according to Nicolas Hubé, because "we don't know how someone judges someone's charisma. What is charisma for the young people, for the old people, for the women, for the men" questioned? They may understand the question very differently. Same thing on competence: “Competent because he/she has diplomas? Because he/she understands the world?”
What should attract attention, believes the teacher, is not so much the question asked as the way in which political comments will address it: "The figure is not very interesting, it is rather how we will construct the probability for a given candidate to be good or not good, to qualify him or to disqualify him." According to him, it is the commentator who will project, in a way, his reading of the political game. Nicolas Hubé sees a striking example of this when “completely different” comments concern “two candidates who are very close on the same question”.
4 Test survey aggregators
Survey aggregators cannot get rid of certain biases, such as the three points already raised. They also do not allow us to know why certain personalities are tested and not others. But they can help you see things more clearly in the long term. They “allow us to see the dynamics” and “to flatten microscopic variations”, which can be due in particular to margins of error, believes the teacher.
In this case, six months before the presidential election, aggregators are not yet relevant. “We are not at the stage of the election where it is decided by half a point, but the closer we get to it, the more we can say that perhaps the choices [of voters] are starting to be made,” analyzes Nicolas Hubé. A way to get out of this “horse race to the nearest comma”.
