Sunday, March 6, 2011

Non passaran! Ban the polls ...

As predictably as night follows day, each poll producing disagreeable results unleashes a torrent of complaints about French polling methods, manipulation of the polls by occult forces, the uselessness of polling many months before an election, etc. etc. Here is a recent example, triggered by the astonishing rise of Marine Le Pen in one recent poll by the Institut Harris Interactif. One can admit the justice of certain of the critiques in this article without accepting its overall conclusion, that polling serves nefarious purposes and ought to be strictly limited. For example, there is no doubt that the quota method used by Harris is inferior to random sampling. But even random sampling has its limitations and requires correctives and a whole art of polling beyond what is covered in elementary statistics textbooks. The pity in France is not that there are so many polls but that there are so few. If there were more, with a greater variety of polling houses and corrective techniques in use, then someone might do for France what Nate Silver has done for the United States: compare polling results systematically to attempt to correct for (and explain) "house effects," which do not necessarily reflect, as is so often assumed in France, malevolent manipulation.

To say that, 14 months before the election, one poll shows Marine Le Pen topping all other candidates is not a prediction that Marine Le Pen will be the next president of France. It is no less foolish, however, to dismiss the result out of hand as the result of bad polling, the use of the Internet, or the hidden hand of some Machiavellian strategist. It's interesting, for one thing, that it used to be argued just as predictably that FN poll results were lower than they ought to be because people were "ashamed" to admit that they were voting FN. If nothing else, the new poll suggests that this "French Bradley effect," if it ever existed, is now definitively dead. Indeed, the opposite may be true: respondents may want to threaten to vote FN in order to "send a message" to the major parties that voters are discontented. An "expressive" rather than an "instrumental" poll response, as political scientists like to say, is nevertheless a warning to be taken seriously by all political parties. Attacking the messenger won't help, and outlawing polls, as Jean-Pierre Sueur and Hugues Portelli have suggested, is a solution worthy of ostriches rather than owls.


Anonymous said...

(not sure why they don't put royal in the mix, she polls at 25% which isn't high, but is much better than Hollande's 7-10)

Anonymous said...

There'll be a new poll then, where someone else than Aubry faces Sarko and Le Pen.
As the NPA/CGT attests, the trend is there. Banning polls or changing them until they give the expected result is indeed a solution "worthy of ostriches".

Anonymous said...

And so they redid the polls, and guess what? Found the exact same result, regardless of the person facing Sarkozy on the left. I think they should have included Royal and Montebourg since they're official candidates (v. DSK, Aubry, and Hollande, who aren't) but nevertheless, it's hard now to say the poll was totally faulty, as some pundits have been saying.