The result of all of this is response rates that hover around 50 percent. And, nationwide, only three-quarters of the voters can be surveyed on election day at their polling places. In 2004, biased responses had a serious effect on the exit polls. Unless the pollsters can find new solutions to the problems and lawyers take up the challenge, exit polls will be more error-prone in the United States.

There is an alternative model for projecting elections that I like, but it has not been extensively tested. It consists of combining two surveys of voters by telephone and statistically modeling the results. The first survey does a thorough job of reaching as many in the sample as possible. It also provides a baseline for measuring the change between the two surveys. The second survey is conducted as close to the election as is practical, perhaps even on election day. As the second survey will not succeed at reaching all respondents from the first survey its sole use is to measure change for key subgroups in the electorate. This approach has been used quite successfully in some Scandinavian countries and may be useful for overcoming the bias inherent in late telephone interviews as they are now done.

Some have touted internet polling, but I see no evidence of its reliability over a large number of elections. Harris Interactive did very well with its national poll in 2000 with a weighting scheme (propensity score adjustment) that it is trying to sell to market research clients as a panacea for selecting poll participants who volunteer rather than being selected at random. But an examination of Harris’s state polls for that same election showed as many states with poorer estimates as ones improving once their weighting technique was applied.

Exit polls in emerging democracies do not face some of the same problems they do in the United States. Usually response rates are very high. It is not uncommon to get 85 to 90 percent of a sample to respond. Also, there usually is no absentee voting. Although electoral commissions, which run elections in most countries, either prohibit exit polling completely or, at the other extreme, make no impediment to their conduct, their interference only affects the time of release of the results. Frequently, any restriction on releasing the results after polls are closed is ignored.

Exit polls are viewed differently in emerging nations. In the established democracies we expect the official result to confirm the exit polls. If it does not, the usual question is, “What went wrong with the exit poll?” In emerging democracies, the question is the other way around: “Is the official result correct?”

So what’s the problem in these countries? In many of them, it is the pollsters. They seem to harbor the naïve notion that I first heard from a government professor at Harvard: “They have a field force, so they can do the exit poll”—that is, they think that all it takes to do an exit poll is the deployment of interviewers to polling stations. The answers of voters on the questionnaires go into a central computer, and the number of respondents naming each candidate is the vote. In many cases, there are no ideas of scientific sampling. No attempt to make unbiased estimates. No quality control on the data. No method of evaluation. I hesitate to say no sampling error, as they have learned how to compute pq/n, even if it bears no relevance to what they did.

Take the 2004 runoff of the presidential election in Ukraine that was overturned in part because of the exit polls. There were many good reasons to overturn that election, such as documents and audio recordings of conversations that demonstrated intentional vote fraud, but the exit polls should not have been among them.

Unfortunately, while pollsters there started the sampling of counties and municipalities (PSUs) in a reasonable way, the sampling broke down after that. Probabilities were equal within PSUs when the units varied widely in size, quotas determined the number of precincts selected within PSUs, substitutions of precincts were allowed by the supervisors in the field, and respondents were selected by quota during only half the day in rural areas and slightly later in urban places. Weighting consisted of forcing those interviewed to the assigned quotas and to the official turnout, which was not available until the next day. To make matters worse, results were released election night, well before there were even preliminary vote counts.

In short, the weighting had nothing to do with the selection. Any resemblance between the vote cast and the result of these exit polls was purely coincidental.

Regrettably, bad sampling and no estimation are the rule and not the exception in emerging nations. It seems to be common practice to tabulate the results in SPSS with little or no weighting. There are other types of missteps, too. Interviewer selection and training in recent exit polls in Venezuela and Taiwan led to biased results. Recent partisan exit polls in Azerbaijan and Mexico led to deliberately distorted results.

So what is the future of exit polls in the United States and abroad, given the problems? In the United States, I expect projection models to change to overcome the current difficulties. In emerging democracies, I expect the influence of exit polls to change when the media start making necessary distinctions between well-done exit polls and those that do not deserve to be reported. So far they have not done that. The general assignment reporters who cover these elections have too little election reporting experience and know almost nothing about what makes for a reliable exit poll.

What is needed is a skeptical press and a place for them to get technical guidance on when to avoid the results of poorly done exit polling, and how to make effective use of good ones. Until this happens, expect to be misled by many exit polls.

Warren Mitofsky is president of Mitofsky International, which joined with Edison Media Research to conduct the 2004 exit polls for the National Election Pool (NEP).