From Forbes Magazine Sept. 2, 2008
International
Early Election Polls Are Often Misleading
Oxford Analytica 09.02.08, 6:00 AM ET
Coverage of the presidential "horse race" invariably crowds out more substantive coverage of policy issues, and the heavy media coverage of polling can also be misleading.
While polling, though still imperfect, does have some predictive capacity in the days before an election, it is quite unreliable months before. These early polls are outperformed by political scientists' election forecasts based on the "fundamentals"--factors such as the state of the economy, the role of foreign wars and whether the incumbent president is running. At best, polls represent a snapshot of the electorate on a given day, rather than a forecast of the eventual outcome
Yet even this modest view has its problems. A snapshot may be seen as worthless if it is clear that the picture must eventually change. In four of the last five presidential elections, the eventual popular vote winner has trailed in the polls at some point, often by a substantial margin. Moreover, these changes are often predictable, as shifts usually move in the direction suggested by forecasts based on readings of "bread and peace": the state of the economy and U.S. standing abroad.
Yet even this modest view has its problems. A snapshot may be seen as worthless if it is clear that the picture must eventually change. In four of the last five presidential elections, the eventual popular vote winner has trailed in the polls at some point, often by a substantial margin. Moreover, these changes are often predictable, as shifts usually move in the direction suggested by forecasts based on readings of "bread and peace": the state of the economy and U.S. standing abroad.
Polling methodology problems. Pollsters also face statistical modeling challenges, which are far more acute months ahead of the election:
"Likely voters." Pollsters attempt to survey the subset of the population that will vote on election day, rather than the entire adult population. (This makes sense, given that these polls are geared toward an election and not an abstract representation of the public's preference, such as presidential job approval ratings.) Some polls survey only registered voters, while others will ask a series of questions designed to screen out those who are not deemed 'likely voters':
--Misapplied technique? These screening techniques help give a more accurate picture of the electorate in polls conducted shortly before election day. However, polls with "likely voter screens" portray a very volatile electorate, with levels of fluctuation that appear unrealistic. Crucially, it is more difficult to predict likely voters in July or September than it is in late October; there is some evidence to suggest that likely-voter screens, in particular, may actually make early polls less accurate.
--Turnout guesswork. Turnout is not easy to predict, even in principle. For example, anecdotal reports suggest that Barack Obama's campaign has a large advantage in the 'ground game' of door-to-door canvassing (in terms of measurable campaign activity such as money invested, volunteers enlisted, and field offices opened). This could conceivably translate into a higher Democratic turnout than models based on past elections predict.
Undecided voters. In an apparent paradox, polls move a great deal over the course of the campaign but usually claim that only a small segment of the population is "undecided." Yet if 85% to 95% of the electorate has already "decided," then early polls should track the November results closely--and they do not. If pollsters' estimates of undecided voters early in the election campaign were meaningful, then the race would essentially be over when one candidate's lead exceeds the remaining number of undecided voters. This paradox has two main causes:
--Pollsters' questions. Aiming for a momentary snapshot, pollsters almost always ask respondents whom they would vote for "if the election were held today." Many also use a follow-up probe to see if those who say they are unsure "lean" in a particular direction.
--Voter preferences change. This approach fails as a predictor because while a voter might have a preference when polled in August, he or she could easily become undecided or change his or her mind later in the campaign. Indeed, since early polls fail to predict November results, many voters clearly do change their minds--more than the 5% to 15% that polls generally report as undecided. Academic studies and exit polls suggest that over 20% of voters make up their minds during the final month of the campaign.
Polls become more accurate predictors of the November election result in October. To draw meaningful conclusions from polling data, it is crucial to dissect the partisan and demographic compositions of pollsters' samples--and to recognize that more voters are genuinely undecided than pollsters acknowledge.
Polling methodology problems. Pollsters also face statistical modeling challenges, which are far more acute months ahead of the election:
"Likely voters." Pollsters attempt to survey the subset of the population that will vote on election day, rather than the entire adult population. (This makes sense, given that these polls are geared toward an election and not an abstract representation of the public's preference, such as presidential job approval ratings.) Some polls survey only registered voters, while others will ask a series of questions designed to screen out those who are not deemed 'likely voters':
--Misapplied technique? These screening techniques help give a more accurate picture of the electorate in polls conducted shortly before election day. However, polls with "likely voter screens" portray a very volatile electorate, with levels of fluctuation that appear unrealistic. Crucially, it is more difficult to predict likely voters in July or September than it is in late October; there is some evidence to suggest that likely-voter screens, in particular, may actually make early polls less accurate.
--Turnout guesswork. Turnout is not easy to predict, even in principle. For example, anecdotal reports suggest that Barack Obama's campaign has a large advantage in the 'ground game' of door-to-door canvassing (in terms of measurable campaign activity such as money invested, volunteers enlisted, and field offices opened). This could conceivably translate into a higher Democratic turnout than models based on past elections predict.
Undecided voters. In an apparent paradox, polls move a great deal over the course of the campaign but usually claim that only a small segment of the population is "undecided." Yet if 85% to 95% of the electorate has already "decided," then early polls should track the November results closely--and they do not. If pollsters' estimates of undecided voters early in the election campaign were meaningful, then the race would essentially be over when one candidate's lead exceeds the remaining number of undecided voters. This paradox has two main causes:
--Pollsters' questions. Aiming for a momentary snapshot, pollsters almost always ask respondents whom they would vote for "if the election were held today." Many also use a follow-up probe to see if those who say they are unsure "lean" in a particular direction.
--Voter preferences change. This approach fails as a predictor because while a voter might have a preference when polled in August, he or she could easily become undecided or change his or her mind later in the campaign. Indeed, since early polls fail to predict November results, many voters clearly do change their minds--more than the 5% to 15% that polls generally report as undecided. Academic studies and exit polls suggest that over 20% of voters make up their minds during the final month of the campaign.
Polls become more accurate predictors of the November election result in October. To draw meaningful conclusions from polling data, it is crucial to dissect the partisan and demographic compositions of pollsters' samples--and to recognize that more voters are genuinely undecided than pollsters acknowledge.
To read an extended version of this article, log on to Oxford Analytica's Web site.
Oxford Analytica is an independent strategic-consulting firm drawing on a network of more than 1,000 scholar experts at Oxford and other leading universities and research institutions around the world. For more information, please visit oxan.com. To find out how to subscribe to the company's Daily Brief Service, click here.
You may also sign up to The World Next Week, Oxford Analytica's free weekly service providing depth and context to next week's important international issues.
*************************************************************
Go to this link for Forbes, enter the word Polls to read the article as written in Forbes-----see this interesting article about Apple from today's Forbes online---mj
Click the link below to read the full story:
forbes.com |
