Same day different numbers
Just as no one would plunk down money on the ponies without checking the odds in the Daily Racing Form, political junkies hunger for those first pre-election polls to see how a new race is shaping up. But what’s a political handicapper to do when the first peeks at the pack send decidedly mixed messages? That’s just what happened when, on the same Sunday in early June, The Boston Globe and Boston Herald both came out with early polls on the race to succeed Joe Moakley in the 9th Congressional District.
According to the Globe survey of 400 likely Democratic voters in the district, the race was beginning as a three-way statistical dead heat between Max Kennedy, state Sen. Stephen Lynch of South Boston, and state Sen. Brian Joyce of Milton. The poll–published just one day before Kennedy abruptly pulled the plug on his wobbly candidacy–had the son of the late Robert F. Kennedy in the lead with 22 percent, followed by Lynch with 20 percent and Joyce with 18 percent.
But all three numbers were well within the poll’s margin of error of plus or minus 5 percentage points, making the nascent race a dead heat. (State Sen. Marc Pacheco of Taunton trailed at 9 percent, while 29 percent of those surveyed were undecided.) NO ROMP AHEAD FOR KENNEDY, POLL HINTS, read the Globe headline, capturing the surprise of some observers that his famous name hadn’t spotted Kennedy a big early lead.
How could two polls conducted at the same time produce such disparate results? Gerry Chervinksy, of KRC Communications Research, who conducted the Globe poll, says the numbers are not as different as they appear. Given the 5 percentage point margin of error in both polls, Chervinksy says the Herald number for Lynch, for example, could be as low as 23 percent, while the Globe poll leaves open the possibility that Lynch’s strength is as high as 25 percent.
Still, Chervinksy concedes that it is only at the edges of the margins of error that the two polls are statistically compatible with each other. It is far more likely, he says, that the polls sampled differing proportions of voters in different areas of the district. With his poll showing Joyce’s strength to be twice what it was in the Herald poll, Chervinksy says, “I can guarantee you right now that I have more people from [Joyce’s senate] district in my sample than the Herald does.”
There’s no way to know for sure. The Herald’s pollster, R. Kelly Myers, of RKM Research and Communications, says he gets his sample by selecting phone numbers at random from a list of 9th District households. Chervinksy does likewise, but then he compares the geographic distribution of respondents with voting returns from a previous election. If there are marked disparities, he’ll weight the findings in accordance with past turnout numbers. In this poll, though, he says he found no need to adjust the numbers, based on the 1996 state primary he used for comparison.
But perhaps he should have. The race that functioned as his benchmark happens to be one in which Milton’s share of the vote in the 9th District was three times greater than it was in the primary two years earlier. In 1996, Milton turnout swelled due to the bruising battle between Joyce and the longtime Milton state representative he defeated to win the Democratic primary for the House seat.
But if Chervinsky’s polling method might have boosted Joyce’s showing in the Globe poll, there is no ready explanation for Lynch’s nine-point lead in the Herald’s. Myers notes that his survey tested more prospective candidates, but that doesn’t explain why Lynch would poll higher in a wider field, especially when one of the alternative choices–Ray Flynn–hails from his own neighborhood.
Asking voters at the outset of a race about a large number of people “who may or may not run,” Myers says, “by the very nature of the poll, it is somewhat speculative.” Adds Chervinsky, “I think people get too horse-race crazy.” And that’s straight from the horse’s mouth.