|Love and Hatred of Electoral Surveys
Erika Casajoana (USA)
|(Translated from the Spanish original)
Many politicians despise surveys in public, arguing that are not good for anything. In a free market economy, an
instrument that is not good for anything −and does not come cheap at all− would be quickly eliminated from the list of
expenses that our debt-ridden parties accumulate during electoral campaigns.
Politicians tend to make ostentation of surveys when they are favorable to them and to dismiss them when they are not.
The fact is that all of them are avid consumers of surveys.
Surveys, done well, offer an accurate picture of the electorate at a precise point in time. They are a tool to understand
the dynamics in play in elections, and can contribute to make fraud in vote counting more difficult, as we saw last
November in the Ukraine. A survey predicts the winners and helps understand the sense of citizens’ behavior in the
exercise of their right to vote.
A survey consists of reuniting a representative sample of possible voters, and analyzing its answers to a series of
questions on electoral conduct, positions with regard to certain subjects and demographic profile.
There are considerable technical challenges in the accomplishment of a good survey of public opinion. For a starter,
the sample must reproduce on small scale the totality of possible voters. The sample’s random selection ought to, in
theory, give all citizens in the census an identical probability to be contacted by the interviewers. The ideal sample
would be the one that includes all the members of the group to study, known as the "universe", with a margin of error of
zero. But if we could do that, we no longer would need a survey and therefore this article would be superfluous as well.
There is the well-known anecdote of Thomas Dewey, Republican governor of New York and favorite to beat Harry
Truman in the 1948 presidential election. During election night, he asked his wife: "How would it feel to you to sleep with
the president of the United States?" She answered that it would be a great honor and that she was looking forward to it.
However, against all expectations, Truman won the elections. The next morning, during breakfast, Mrs. Dewey joked:
"So, tell me, Tom, must I go to Washington or will Harry come to see me here?"
Among other technical shortcomings, the surveys that anticipated that Dewey would win were done by telephone, then
a luxury outside the reach of many Americans. The sample thus selected suffered a bias towards the wealthy classes,
and consequently underestimated the Democratic vote.
It is not enough to begin with a representative sample. Also, the people who complete the survey must reflect the total
population who will vote. The set of respondents is corrected to try to make it agree with the set of voters in traits such
as proportion of men and women, geographic distribution and of ages, levels of education and rent, and political
affections, for example.
It is very important to guess right when calculating the probability of the respondent to exercise his or her right to vote.
No polling firm wants to interview people who will not vote, since although they are citizens just like any other, the
opinions of the non-participants do not count politically and therefore distort the reliability of the collected data.
Other difficulties arise as a result of respondents who refuse to participate, lie or otherwise do not pay enough attention
or do not understand what they are being asked. All of these incidences diminish the survey’s capability to predict the
electoral result, unless the pollster is skillfull enough to calculate those distortions and correct them.
The corrections, weights and manipulations that with more or less ability and more or less rigor are applied to the
survey’s gross results are known in jargon as "the kitchen". After all, like with chefs, "the kitchen" is what distinguishes
good professionals from charlatans, and artists from craftsmen.
During the 2002 French presidential campaign, the leading company in polling, Ipsos, conducted surveys in which 4
percent of the interviewed said that they would vote for the extreme right-wing candidate Jean-Marie Le Pen. Aware of
the fact that some respondents were probably ashamed to admit they favored a racist and a xenophobe −even to a
stranger speaking on the phone−, Ipsos analysts calculated that the actual proportion of Le Pen voters would be 4
times greater, that is, 16 percent. Le Pen unleashed a political earthquake when he obtained in the first round more
than 17 percent of the votes, thus qualifying for a one-on-one confrontation with president Jacques Chirac in the
second and final round. Ipsos was the company that came closest to the actual Le Pen vote, and even they
Surveys can fail for many reasons. Aside from the challenges with the sample and respondents, it is also indispensable
that the questions be scrupulously neutral, that the answers be correctly taken and processed and that the data
analysis be objective. And yet, the best of surveys is subject to the vagaries of human behavior. A survey reflects the
state of public opinion at a determined day and hour, which is never the moment when we citizens face the final,
intimate decision on our vote. After all, politicians are right: the only survey that counts is the scrutiny.