January 11, 1996

Chance Error and Bill Clinton’s Political Fortunes

The big difference this week between the CNN/USA/Gallup poll and ABC/WP poll is the first major polling disparity of the ’96 campaign. Gallup’s results indicate that support for the President has tumbled sharply, while the GOP leadership is viewed more favorably than a month ago. In sharp contrast, the ABC/WP poll found no trend away from the President, and no increase in support for GOP leaders, or their policies. Both polls were taken at about the same time, use identical or comparable question wordings, and employ similar sampling and interviewing approaches.

So why the difference … Two things may be happening. First, public attitudes toward the dispute between the President and his GOP adversaries may be softening. Volatility in response is often one consequence of people being, or becoming, unsure about something. Very small differences in method can make for large differences in results when survey organizations ask the public about issues on which their thinking is unclear, or when the climate of opinion is destabilizing.

The second reason may be more difficult for the political and media communities to accept. The difference in the survey results may reflect sampling error. Poll reports always describe their margin of error … +/- 2 or 3%, 4 or 5% points, but usually little attention is paid to this advisory. What this means is that the chances are 95% that the true percentage is within that margin. The true percentage is what would have been found, if the surveyors had taken a census of the entire population, instead of sampling it.

But as every poker player knows, from time to time an outcome will seem to buck the odds of 1 in 20 (or a 5% chance).

An analysis of national media poll results over the last six months finds very stable survey to survey measures of the President’s approval rating. Since late June national media surveys have made 29 trend comparisons of the percentage of Americans who say they approve of the way the President is doing his job.

Typically, a survey organization will find a 2% or 3% point difference between its current approval percentage and its preceding measurement. In 7 out of 29 cases the comparison has been in the 4% to 5% point range. In each of these instances, polls of other news organizations showed a trend in the same direction, or a comparable percentage.

In just 1 out of 29 instances, the difference has been greater than 5% points. And that is the current controversial Gallup survey (-9% points). This well may be a case of the odds catching up with the surveyors and chance error breaking against Bill Clinton.