The week before this past Tuesday's elections, I had a running disagreement with Cook Political Report House editor David Wasserman, whose office sits right next to mine, about public polling on the special election in New York's 23rd Congressional District.
Two weeks ago, the Club for Growth released a survey it had sponsored on behalf of Conservative Party candidate Doug Hoffman. This survey had been the object of criticism, as it was the first to show Hoffman narrowly ahead. I posted commentary on Pollster.com defending the poll's methodology as reasonable and its results plausible, but David sent an e-mail raising questions about its likely voter screen. A friendly debate ensued.
Then on Saturday, the Siena Research Institute released a new survey conducted earlier in the week, confirming Hoffman's rise and showing him running just a single percentage point behind Democrat Bill Owens (35 percent to 36 percent). Within a few hours, Republican nominee Dede Scozzafava, whose support had plummeted to just 20 percent in the Siena poll, announced that she was suspending her campaign. The next day, she further rocked the political world by announcing an endorsement of Owens.
The news caught everyone by surprise, but by Monday we had three new polls to consider, all showing Hoffman surging to a lead. As the table below shows, the new polls put Hoffman ahead by margins of 5 to 17 percentage points. When plotted on a chart, the polls showed Hoffman's support rising sharply.
I looked at the strong, consistent trend to Hoffman, the large number of Republicans who had been supporting Scozzafava and the relatively small number of newly undecided voters who knew her well and confidently forecast a "comfortable victory" for Hoffman.
Wasserman saw it differently. He noticed that Scozzafava's backers disliked Hoffman (57 percent) slightly more than they disliked Owens (50 percent), while two-thirds rated President Obama favorably (64 percent). "Scozzafava's withdrawal," he concluded, "was no guarantor of a huge Hoffman victory."
Then on Tuesday, Owens defeated Hoffman by an unofficial margin of 49 percent to 46 percent. Wasserman was right, and I was wrong.
In retrospect, the final Siena poll accurately captured a surge in uncertainty. Their horse race question informed voters that Scozzafava had quit the race but made no mention of her endorsing Owens. As such, it showed a huge increase in undecideds (from 9 to 18 percent) and a 5-point Hoffman lead that was not quite statistically significant.
But what explains the narrow Hoffman lead so close to Election Day?
The Siena pollsters looked closely at their results and offered a theory. When they tabulated their Sunday night numbers, they found that their margins were essentially correct in the District's eastern section and plausibly correct in the central counties that made up Scozzafava's base. If they assume that those in Scozzafava's base who moved away from her between Sunday and Tuesday took her advice to support Owens, their margins there would essentially match the election results.
No such luck, however, for the three counties in the Syracuse media market -- Oswego, Oneida and Madison. In that region, Hoffman led Owens in Siena's Sunday poll by 23 points (51 percent to 28 percent) but won by only 6 points (51 percent to 44). Via e-mail, the Siena pollsters attribute the discrepancy to random sampling error.
I'm not so sure. I checked with the pollsters who did a survey on Sunday and Monday night sponsored by the National Organization for Marriage, and they too showed Hoffman leading in those three counties by roughly the same margin as the Siena poll.
Democratic pollster Jef Pollock also tells me that his final survey for the Owens campaign had shown Hoffman with a big lead in the same three counties a week earlier. Pollack says that Hoffman's momentum in the Syracuse market convinced the Owens campaign to shift its final advertising message there to an attack directed solely at Hoffman.
Why might these three counties show so much more late volatility? It is the one part of the district, as Wasserman wrote, "which is not home to any of the three candidates and where the race has played out in fierce television ads." While voters in the central portion of the district learned about the race from frequent front-page stories in the widely-read Watertown Daily Times, news about the race was far less prominent for voters in the Syracuse market.
The unprecedented events of the campaign's final weekend, however, probably changed that. Experience tells us that when a candidate enters a contest's final weekend with both a lead and as much apparent momentum as Hoffman, they almost always win. By the end of a long campaign, voters have spent weeks or months acquiring information and pondering their choice, and most have made up their minds by the final weekend.
In this case, however, the voters experienced "man bites dog" news twice: First, the Republican nominee dropped out, then she endorsed the Democrat. I am guessing that the unprecedented news made a significant number of these habitual voters sit up and take special notice, especially those who had until that moment experienced the campaign mostly through television advertising.
So here is my hunch: When confronted by a pollster's call over the weekend, many were simply not ready to make a final decision. If pollsters pushed hard for a choice, some voters may have fallen back on an initial preference that they were now in the process of reconsidering. For those who shifted to Owens that weekend, however, the campaign had started anew. Their final decisions were probably not made until they cast a ballot on Tuesday.
This episode teaches two lessons. First, when it comes to House races, listen to David Wasserman. Second, when something truly unprecedented occurs in the final weekend of an election campaign, expect the unexpected.