During the primaries, there were no mistakes in projections that could be traced to exit poll problems - the only projection error, the Associated Press mis-call of Hillary Clinton in Missouri on Super Tuesday, had nothing to do with the exit poll. But doing good exit polls sometimes requires more traditional polling methods that go beyond simply sampling voters at polling places. That's because, in a growing number of states, more and more people aren't waiting for Election Day to cast their ballots.
The percentage of voters who DON'T vote at the polling place has been consistently growing. As many as one in five voters may have voted before Election Day in 2004, and even more are likely to do so this year.
Some states have a history of making it easy for voters to cast their ballots before Election Day. In 1992, nearly a third of the vote was cast early in Texas, which (like some other states) opens the doors at a limited number of sites as early as 17 days before an election. In this year's primary, held on March 4th, that meant that while all Texans cast ballots knowing the results on Super Tuesday, early voters might not have known about Barack Obama's big victories in Wisconsin and Hawaii on February 19th.
Florida adopted this type of early voting after the contested 2000 election. What Floridians call "in person absentee voting" begins 15 days before a primary or general election. This year, Florida primary voters could cast ballots as early as January 14th. Those voters would know the results in Iowa and New Hampshire, but Obama's win and John McCain's defeat of Mike Huckabee in South Carolina on the next two Saturdays had yet to happen.
It may be that the increased use of exit polls by news organizations for analysis and projections has fueled the increase in absentee voting. As early as 1964, there were complaints about journalists' early projections of a Lyndon Johnson victory based on results in states that closed their doors and reported results before the polls had closed in Western states. There were even Congressional investigations of the potential impact of reporting those early results (no exit polls were taken back then).
After the 1980 election, when exit polls were available, these complaints grew louder, so news organizations promised not to make projections or characterize races in a state before the majority of its polls closed. But as it turned out, in each of the next four presidential elections (1984, 1988, 1992, and 1996), the winning candidate garnered an electoral vote majority well before 11 p.m. Eastern time. In theory a candidate could capture a majority of electoral votes by 8 p.m. ET, although it's almost impossible for a candidate to do so well as to be assured of victory by then. But about 80% of all electoral votes are available at 9 p.m. Members of Congress have occasionally promised to enact a nationwide uniform poll-closing time to prevent early projections, but have never actually done so.
Oregon, California, Washington, Idaho, and Hawaii close their polls at 11 p.m. ET; Alaska at 1 a.m. In Oregon in 1998, absentee voting soared to 61 percent of the total vote, and after that election, Oregon switched to a system in which votes are cast by U.S. Mail. People who want to can still drop off ballots on Election Day, but most Oregon voters don't. Ballots must be received by Election Day, so most voting takes place beforehand.
Washington State is in the process of adopting a similar all vote-by-mail system. In-person voting is permitted there only in the two largest counties (King and Pierce); yet even in those counties, just a minority of voters takes their ballots to the polls on Election Day. And while Oregon and most other states require absentee ballots to be received by Election Day, Washington only requires that they be postmarked by Election Day. That has delayed counting - and knowing the results - for days afterwards! Alaska and Hawaii also have significant number of absentee voters; and in California, one can register as a "permanent" absentee voter, and automatically receive an absentee ballot in the mail, for all elections, no matter how minor.
So, "traditional" exit polling is getting harder; and pollsters have to find new ways to sample these voters' choices. Sometimes those new ways are just like the old ways. So far in 2008, the news consortium National Election Pool (NEP) has conducted pre-election polls of voters who have voted (or definitely will vote) absentee or early in nine states. Several states that usually have high absentee voting, like Colorado and Washington, held caucuses for delegate selection this year, and there was no need for absentee polls there: Caucus rules usually require attendance in person.
The telephone surveys are typically conducted during the week before Election Day, using traditional polling methods. Phone numbers are generated by computer using random-digit-dialing procedures; respondents are selected at random within every household reached; and (although all respondents are asked a few necessary demographic information) only registered voters who have voted already or who definitely are voting absentee or early are interviewed fully. Those results - for questions about issues, opinions of candidates and voter demographics - have to be merged with the exit poll data collected on Election Day to reflect the correct proportion of voters who are casting absentee or early ballots. As with any survey, researchers must take account of sampling error and other polling issues, so the estimate has to account for them, too. Historically, absentee poll estimates have been very good.
Sometimes early voters have different preferences as well as different demographics from those who vote on Election Day itself. This year, when Obama came on strong at the end, those voting on Election Day were more likely to favor him than those voting absentee. Other times Clinton fared better with those who came to the polls on Election Day. Voting early sometimes means missing out on the information available to those who wait!
By Kathy Frankovic