In Polls, 1 Plus 1 Doesn’t Always Equal 2

The latest round of presidential pre-election surveys has produced a flurry of seemingly contradictory findings, with Bill Clinton holding anywhere from a 2-point to a 10-point lead. A lot of that difference may be due not to shifting voter allegiances, but to the educated guesses of the polling mandarins about who is likely to vote.

The U.S. Census estimated that about two-thirds of Americans were registered to vote in 1988, but just about half of all adults actually voted in the presidential election. The challenge is to identify those who are actually registered--and those who will really cast ballots.

Pollsters generally identify registered voters by asking a respondent if he or she is enrolled to vote. That is only an estimate, because some people will say they are registered even if they are not.

Finding the elusive, smaller group of “likely” voters is tougher. It takes an additional series of questions that attempt to gauge people’s interest in and knowledge of the election, their commitment to their candidate and whether they have voted in the past. Based on the answers to these inquiries, participants are graded as “likely” or “not likely” to vote.

Pollsters often check the demographics of their “likely voter” samples against exit polling statistics from past elections, to see if they are correctly representing the shares of men and women, whites and blacks and young and older people who have traditionally turned out. Some may even verify whether the people in their sample are registered.

This is not an exact science. There is no industry standard for determining likely voters; different sets of questions and different scoring techniques are used for deciding who will and who won’t vote. And all this is based on what people say they do and feel, rather than on solid facts. Finally, though pollsters are fairly skilled at finding those who have voted in the past, they do less well at estimating the potential effects of younger, newer members of the electorate.

That’s what is complicating the pollsters’ job this year. Registration data and absentee voter requests from across the country indicate heightened interest in next week’s presidential vote, suggesting a turnout well above 1988 levels. Poll data supports that conclusion: a mid-October Los Angeles Times survey four years ago found 45% of voters who said they were very interested in the presidential election. In the latest poll that figure is 64%. Just 36% of young adults were following the Bush-Dukakis contest closely in 1988. Now, 54% express such sentiments.

All this is critical because the size of Bill Clinton’s lead seems to depend largely on how many of those untested but enthusiastic younger people pollsters allow through their “likely voter” window. Among the bigger, unscreened pool of registered voters, Clinton has a generally wider and more consistent lead.

Imposing a very tight screen on potential voters narrows the gap. Will those people turn out? While pollsters argue and point spreads open and close, only next week’s vote count will tell.

Another factor is the overemphasis on minor differences in the point spread.

A spread of 7 points between Clinton and Bush may look dramatically different than an 11-point margin, but from a statistical standpoint they are virtually the same. Take, for example, The Times Poll’s survey of likely voters, released Wednesday. Clinton was at 44%, Bush 34%, and the sampling error plus or minus 3 percentage points. So Clinton’s number actually could be anywhere from 41% to 47%, and Bush’s somewhere between 31% and 37%.