Election Is Closer Than It Appears

THE polls in September suggest two views of the presidential contest between George Bush and Bill Clinton. They could be read as indicating that the Democrat is well ahead. Not one reputable national survey shows Clinton trailing. His lead is about 10 points.

But the polls also show a high level of voter indecision, with large segments of the electorate tugged in opposite directions. Bush suffers from a general dissatisfaction with the status quo, but he gets good marks for important elements in his handling of the presidency, especially his management of American foreign policy. Clinton, meanwhile, has failed to dispel doubts about his suitability for the nation's highest office.

The data make me feel that the 1992 contest is far from over. It's the toughest call I can remember. New tracking data for September confirm that feeling.

Election surveys are normally conducted in a span of two to five days. A "tracking poll," by contrast, is in the field more or less continuously. This year we have only one publicly available tracking poll, conducted by the Tarrance Group with analysis by Ed Goeas, a Republican pollster, and Celinda Lake, a Democrat. It's known as the "Battleground '92" tracking poll, and it's paid for and carried by a number of news media around the country. "Battleground '92" began collecting data on Sept. 1, conductin g 250 interviews a night, five days a week.

To reduce the day-to-day bounce, much of which might result from nothing more substantial than the small daily samples, the poll's directors follow the generally sound practice of releasing findings only for "rolling averages." The polling done Sept. 20 through 23 was combined and released as though it was a single survey of 1,000 registered voters. Hence, it's always a four-day "rolling" average that the reader sees.

At first glance, the tracking poll's findings over this past month don't appear to differ much from those of other surveys. For example, the results last week showed 49 percent saying they would vote for Clinton, 40 percent for Bush, the rest saying they were wholly unsure or for Ross Perot. That's the same picture that one gets from the composite of all national surveys this past month.

But when one lays out the results for each day individually, a different picture emerges. The poll officials decided not to poll Friday nights or Saturdays on the ground that, especially in good weather times of the summer and fall, people at home to take calls are not a representative cross-section of the adult population. Affluent people, for example, are more likely to be away. The wisdom of the judgment not to poll on Fridays and Saturdays is borne out when one sees how different the results are for the one weekend day - Sunday - that Battleground '92 includes from the results of middle-of-the-week days.

Individual daily findings, based on small samples, may reflect a random bounce. So there really is a good case for looking at rolling averages. Still, the pattern of Sunday versus mid-week responses for September is striking.

In the Sunday surveys, Clinton's margin over Bush was never less than 12 points and averaged 16 points. By contrast, on weekdays Clinton's margin averaged just over five points. He led by just three points between Sept. 15 and 17; and by four points between the 22nd and 24th.

The reader may view this latest piece of information on day-of-the-week interviewing with some consternation. "Do you mean," you may ask, "that not only should I discount poll findings on the grounds that a great many people haven't reached a decision a month before the election, and that many more people than usual are uncertain and drifting this year - but that now I must also pay attention to whether the poll was done on a weekend?"

Alas, the results of the September Battleground '92 tracking poll add to the evidence that the day of the week does matter. Each mid-week the race seems to tighten, only to widen again when weekend results are added. What's more, there is every reason to believe that it's the weekend findings which needed to be discounted - since the segment of the electorate at home at these times and thus potentially available to interviewers may be quite unrepresentative of the national electorate.