In the United States pollsters refer to a so-called Bradley effect to explain why polls are (allegedly) often wrong in contests that pit a white candidate against a black. The theory is that people are reluctant to say that they're voting against the black, for fear of being tagged "racist," so they say that they are undecided or even that they will vote for the black candidate. If there ever was a Bradley effect, it seems to have diminished in recent years. But now Le Monde is adapting the theory to the French case to explain why pollsters underestimated the Front National vote in Sunday's election. But as Le Monde also explains, pollsters are aware of all sorts of biases in their surveys and do not predict results based on raw data. They always apply correctives based on past experience, and not just to correct for Bradley-type effects.
So the real question here is why the corrections applied in this case were wrong. One possibility is that a certain portion of the FN vote varies from election to election precisely because it is a protest vote rather than a pro-FN vote. The FN has its core supporters, to be sure, but it is also a convenient place to express a general ras-le-bol. Voters of this type may have preferences that swing dramatically from election to election. For example, suppose you have 3 percent of the electorate that is really disgusted with all the parties. One time they might vote for the LCR because they caught Olivier Besancenot on TV and liked what he was saying (and LCR's vote share rises sharply, leading to all sorts of wild speculation and even, perhaps, to a restructuring of the party). Then they hear that the NPA (LCR's successor) is running a veiled candidate, so they switch from Trotskyism to anti-Islamism and vote FN because they've seen the FN minaret poster. But maybe they don't want to say they're voting FN. When the pollster tries to correct for this by asking for opinions about a series of personalities, the answers don't necessarily offer the portrait of a confirmed Lepenist: they like Besancenot and Mélenchon, say, and are indifferent to Marine Le Pen. Or maybe, because the whole political process fills them with disgust, they just hang up on the pollster. The numbers involved here are relatively small, and the people involved are motivated by volatile political emotions rather than fixed preferences.