Here a Poll, There a Poll ...
AS I travel around the country, I frequently get asked about polls. Many wonder whether they should believe polls at all, some ask whether they are valid measurements of public opinion, and others fret over the possibility that the results can be manipulated to report what the pollster wants to say.
Having been a pollster once, and seeing the results of countless surveys each week, I explain to these people that when taken in a methodologically valid way, polls can be very accurate, with 19 out of every 20 coming within an established margin of error. Yes, their results can be tilted to get a desired response, through question wording or sequence, but pollsters with good reputations don't usually risk them to do that. Most surveys sponsored by national news organizations, i.e. the polls featured on the news, can generally be relied upon.
Having said that, I am pretty much at a loss to explain what's going on with the national polls on the 1996 presidential race. If you believe the ABC News/Washington Post poll taken Jan. 6-7 of 852 adults nationwide, President Clinton is running 16 points ahead of Senate majority leader Bob Dole, the GOP front-runner, 53 percent to 37 percent. But at the other extreme, a Gallup poll taken Jan. 5-7 for CNN and USA Today puts Mr. Clinton ahead of Senator Dole by only 1 point among the 1,000 adults interviewed, 47 percent to 46 percent. When Gallup pollsters ''screened down'' to include only respondents who said they were registered to vote, Dole actually moved into the lead by 3 points, 49 percent to 46 percent. A newer Gallup poll, also taken for CNN and USA Today Jan. 12-15, pegged Dole as being up by one point among registered voters, 49 percent to 48 percent.
But the confusion isn't just between those respective pollsters and their news organization clients. Two polls have Mr. Clinton leading Dole by 6 percentage points: one conducted by Yankelovich Partners (Jan. 10-11 among 1,000 adults), which put the horse race at 47 percent to 41 percent; and another by CBS News (Jan. 2-3 among 961 adults), showing the president ahead by 48 percent to 42 percent. Yet another poll showed Clinton up by 8 points: a U.S. News & World Report-sponsored survey taken by Democratic pollster Celinda Lake and GOP pollster Ed Goeas, which showed Clinton beating Dole 47 percent to 39 percent.
So, what gives? Sometimes it's the timing of the interviewing, with intervening events changing public attitudes, but no such pattern emerges among the seven polls examined. The questions asked were essentially the same - none were loaded to favor either candidate.
While it's impossible to say for sure what the cause of the variation is, one factor may be that five of the seven surveys were taken entirely or partially over weekends, when many pollsters over the years have found it very difficult to find representative cross-sections of the public at home and willing to sit through a lengthy telephone interview. Saturdays in particular are risky, many pollsters say. Some admit they go to great lengths to avoid polling on Saturdays, though for scheduling purposes, clients often demand it. Four of the seven polls were taken, in part, on Saturdays.
The two polls examined that were not taken over weekends had exactly the same margin, a 6 point Clinton lead. That was just six-tenths of a point off the average of all seven (using the registered-voters number when results from both all adults and registered voters were reported).
When faced with conflicting polls, however, the best thing to do is to average them or, as some suggest, try the Olympic scoring method: Throw out the highest and lowest and average the rest. It isn't terribly scientific, but it works.