Europe is in the midst of a cluster of important elections. In the coming days and months, voters will go to the polls in Latvia, Bosnia-Herzegovina, Ireland, Turkey, and Austria. These follow elections in the last month in Slovakia, Serbia, Germany, Sweden, and Macedonia. Surveys on voter preference are a familiar part of the process, but their record of accuracy is mixed, leading on occasion to major surprises on polling day.
Prague, 2 October 2002 (RFE/RL) -- Voters go to the polls this weekend in Bosnia-Herzegovina and Latvia, the latest in a cluster of important European elections whose results will be key to the future makeup of the European Union and NATO.
Next up is a referendum on the Nice treaty in Ireland, followed by elections in Turkey and Austria later this year. In the past few weeks, there have been polls in Slovakia, Serbia, Germany, Sweden, and Macedonia.
In the run-up to these elections, media attention always focuses heavily on the parties' or candidates' ratings in opinion surveys. But the accuracy of these surveys has been mixed, leading on occasion to major surprises on polling day. In Serbia on 29 September, for example, nationalist candidate Vojislav Seselj won nearly twice the share of the vote indicated by pre-election surveys.
In Slovakia last month, the party of incumbent Prime Minister Mikulas Dzurinda polled much more strongly than expected, and the Slovak Communist Party got into parliament for the first time since the fall of communism.
True, the Slovak and Serbian results barely come close to the shock of, say, the 1992 elections in Britain, where John Major's Conservatives defied poll predictions of a Labour victory -- or a hung parliament -- to cling to power, but the examples show how pollsters can still get things wrong.
To judge by the number of areas where inaccuracy or error can creep in, it's a wonder polls come close to getting it right at all. Pollsters say respondents can lie or be reluctant to say who they're really voting for, particularly if the candidate or party is on the political fringe.
Some agencies use quota sampling, where respondents are chosen to resemble the local population based on census data on sex, age, class, employment, and so on. But such statistics can be out-of-date. It has been suggested that polling organizations were working with old statistics when they made their 1992 predictions of a Labour victory in the United Kingdom.
Roger Mortimer is an analyst with MORI, which has conducted market and opinion research in Britain for more than 30 years and has offices worldwide. He said even random samplings -- where interviewers call telephone numbers at random -- require a reasonably reliable picture of the population. "The biggest problem is that if one party's supporters are more prepared to answer questions than others, or are more likely to be in when you're doing the interview, then if you're not aware of that, and if you're not able to correct for that, obviously, you'll get a biased result," Mortimer said.
Then there's the question of what to do with the large chunk of the electorate -- sometimes 20 percent -- still undecided about a candidate until the last minute. There may also be a blackout of a few days or even, in Slovakia's case, two weeks before polling day in which opinion surveys cannot be published for fear their results could influence voter turnout and preference. In Slovakia, surveys were unable to pick up on a swing in favor of Dzurinda's party late in the campaign.
Wolfgang Donsbach is a professor at Dresden University and an expert on public-opinion research. He said the main German pollsters got their predictions accurate to within a fraction of the actual result, which gave the Social Democrats and the Greens a victory over the Christian Democrats and Liberals on 22 September. But he said the polling agencies would have had a harder time working within a blackout restriction. "Imagine that in Germany, if the survey organizations had polled a week and a half before the election, they would not have picked up the downswing that the Liberals had got because of a scandal around one of their leading members," Donsbach said.
Then there's the specific local dimension to factor in.
Marko Blagojevic is spokesman for the Center for Free Elections and Democracy in Belgrade. He said polling agencies there apparently forgot to take into account the blessing that nationalist candidate Seselj got from former Yugoslav President Slobodan Milosevic from his cell in The Hague. "There are some estimates that some 9 percent of [Milosevic's Socialist Party of Serbia] SPS voters voted for Seselj solely because of the fact that Slobodan Milosevic...suggested [they do] so. And that could be the difference between what was published before the election and what the elections...actually showed," Blagojevic said.
In some countries, agencies have devised ways to "correct" their results to indicate the likely voter preferences of those unwilling to respond to surveys. But Blagojevic said pollsters in Serbia have not yet been able to gauge Seselj's support accurately. And he said agencies there often come up with results that are completely at odds with each other.
Asked if there are any lessons to be learned from past mistakes, Blagojevic had one suggestion that's not likely to go down well with survey organizations. "Yes and no. Because the only suggestion that I could give would be not do any polls anymore," Blagojevic said.
Inaccuracy also raises another issue related to reliability: how credible the polls are.
Vadim Tudor, the nationalist who came in a strong second in Romania's presidential elections two years ago, accused some agencies of deliberately downplaying his support level in pre-election polls.
The World Association of Opinion and Marketing Research Professionals, known as ESOMAR, and the World Association for Public Opinion Research set global guidelines on public-opinion research, but these are voluntary.
Mortimer and Donsbach said that, theoretically, polls can be skewed to favor one party or candidate over another. Donsbach said he can recall one such incident uncovered recently in Mexico. But in general, he and Mortimer agreed that polling organizations would stand to gain little by doing this, as they would lose customers and damage their reputations.
In any case, Donsbach said his research shows that pre-election polls have little, if any influence, on voter behavior.