Copyrighted material


Presidential Poll Results Meaningless

by David Murray

Sloshing around within the margin of error
Great armies, faced with the confusion of battle, were told to "march to the sound of guns." In contrast, today's political armies often "steer to the sound of applause," in the deft words of Michael Gove of the Times of London. That is, when confronted with Matthew Arnold's "darkling plain, swept with confused alarms of struggle and flight, where ignorant armies clash by night," they turn to polls for guidance.

So long as polls are kept in their proper place, used as occasional check points on the road towards one's principals, they are useful servants. But when polls are used to determine destinations in the first place, we've entered a Humpty Dumpty world, only now upside down -- soothing words and pithy phrasings become the masters of political principal. This state of affairs creates two problems. Not only do voters detect this pusillanimous procedure and, seeking authenticity, turn away, there is an additional danger to political strategists themselves -- they might believe the polls, and change their tactics accordingly. Though polls have their scientific methods and their authoritative numerical window-dressing, they have a capacity to mislead. When that happens, more than one egghead is destined for a "great fall."

A case in point is the interpretation of the most recent (July 17) CNN/USA Today/Gallup presidential poll. Under the headline, "Poll: Gore Closes Gap Against Bush," CNN announced a sudden turn around in performance in a very short time. Scarcely a week ago in a comparable poll, candidate George W. Bush led Vice President Gore by a comfortable margin -- 50 percent to 41 percent. But a poll of 635 "likely voters" taken July 14-16 showed a considerable swing. Bush now stood at only 48 percent, to Gore's 46 percent, a scant 2 percentage point difference. These numbers no doubt are an encouraging sign for a formerly dispirited Democratic party.

But then things get a little chancy, as the electioneers interpret what we're seeing in the numbers. Gore is said to be "capitalizing on gains among women and Independents," and the numbers do appear to make this case. In the prior poll of July 6-9, Gore got only 43 percent of the female voters, while now he receives 51 percent. Only 32 percent of Independents went for Gore last week, but he now stands as 44 percent; that is, he picked up a whopping 12 percent of additional support.

The real problem emerges when the CNN analysis identifies the political factor that must have turned the trick. What happened during the week? The endorsement of Bill Bradley, whose "natural constituency during the Democratic primary" are those now said to be breaking for Gore. Leaving aside for the moment the common problems of all polls (it's only valid 'if the election were to be held today,' we don't know whether those polled were actually a representative sample of those who will vote on election day, we don't know the non-response rate or the effect of a telephone methodology, etc. etc.), take a gander at those margins of error. For the "constituencies" in question, they range between 4 and 6 percent. Because the original sample is only 635 individuals, the overall margin of error is 4 percentage points, plus or minus. Hence, a result that shows a lead of 48 percent to 46 percent is really only one outcome taking place between two "error bars" that are actually 8 percentage points apart -- 4 percentage points higher or lower than the actual result. Thus, the actual finding of the poll is a range of outcomes spread between a Bush lead of 52 percent over Gore's 42 percent (to take one possible outcome of the 8 percentage point swing) or a Gore lead over Bush (50 percent to 42 percent), a result also contained within an 8 percentage point margin of error swing. It gets even worse when it comes to interpreting numbers for males, females and Independents. Since these are sub-samples of the overall sample, and have roughly 320 individuals in each category, the margin of error swells to 6 points plus or minus -- a 12 percentage point overall swing. Hence, this week's result show that Gore may lead Bush for female likely voters by 51 percent to 44 percent, but the actual possible spread for the finding runs between a huge Gore lead over Bush (57 percent to 37 percent), or a Bush advantage over Gore (50 percent to 45 percent). The same argument pertains to the results for Independents.

The critical point for the political prophets is that the apparent swing from the week of July 6-9 to the week of July 14-16 is within the margin of error of either poll. That is, it could well be an exercise in self-delusion if Democratic (or Republican) strategists read the results of this poll and concluded anything at all about the impact of Bill Bradley's endorsement. It could have been meaningful. It could have been an anomaly (an interpretation strengthened by the odd finding that among 18 to 29-year-olds, Bush walloped Gore 64 percent to 30 percent -- a 15 percentage point decline from Gore's standing a week before. Be endorsed by Bradley and lose 15 points among the young?) Or it could just have been the fluctuation of statistical noise represented by the sloshing around within the margin of error.

The take home message? Findings such as these make great headlines, but as a guide for political generals, they are not a map to steer by.


Dr. David Murray, Director of the Statistical Assessment Service in Washington DC

Comments? Send a letter to the editor.

Albion Monitor July 24, 2000 (http://www.monitor.net/monitor)

All Rights Reserved.

Contact rights@monitor.net for permission to use in any format.