Bias is a devil... that lies in the detail.
"Did you buy Le Monde last night? Question that seems easy to answer... Most of the time. But take the case of this extremely regular reader of the daily newspaper who just yesterday didn't buy it: he picked it up on the plane! How will he respond?
"No" ? But in doing so, it will seem to him to be lying, to be distorting reality, because not only will he have had a copy in his hands, but he will have read it... and even taken it home.
"Yes" ? By answering "Yes"However, from a statistical point of view, it biases the results.
Consumers don't necessarily answer exactly the questions they are asked, but sometimes they do. what they feel best matches what is expected of them He's in a kind of role-playing game where he's guessing at the rules... which don't exist!
Another, more well-known bias is that the respondent does not dare admit to certain facts or opinions that the interviewer might judge negatively; for example, considering voting for an extremist party in the next elections. Or conversely, they may overestimate behaviour that will be perceived as positive, and say that they regularly give money to beggars even though this only happens very exceptionally.
The self-administration of questionnaires via the Internet minimises these latter biases to some extent, as the individual does not feel that he or she is being questioned. "judged" by the interviewer - but not entirely: by over-emphasising your generosity in the face of poverty, you show yourself to be not only better in the eyes of the interviewer, but also better in your own eyes - and this bias will never really disappear.
Rather than biases, we should be talking about discourse effects - or more precisely about speech contractin reference to Eliseo Verón which introduced the notion of reading contract in its analysis of the magazine press, following the work ofUmberto Eco on enunciation.
This implicit contract defines a common frame of reference between the authors and their readers: the authors express themselves at a specific level, and the readers at a specific level. model playerThey have their own behaviours, motivations and expectations, and to respond to these in the best possible way, they develop a range of skills. editorial line to which they will adhere.
Mutatis mutandis, the same type of contract binds interviewers and respondents:
- The institute adapts its questions to a model intervieweeThey presuppose attitudes and behaviour based on their own experience;
- The respondent imagines that if he or she asks this or that question, it is not innocently, but with a view to getting this or that answer - not about the content, but about the philosophy of the questionnaire.
It should be noted that the discourse contract works in the opposite way to the reading contract: whereas it is the reader who chooses the magazine he or she wishes to read - and therefore initiates the contractual process - it is the interviewer who approaches the consumer - physically or electronically, it doesn't matter.
Just as a magazine that denied the existence of the discourse contract - even if it remains implicit - would appeal to few readers, so a research institute that did not draw up its questionnaires with a model interviewee in mind would run the risk of producing completely erroneous results.
This obviously raises the question of the relevance of all the polling tools that Google or social media make available free of charge to their users - or rather, the relevance of the results obtained by neophyte users.
