How can we read a poll to determine when an incumbent is vulnerable to a primary challenge?
In Connecticut, a Quinnipiac University poll finds Democratic Sen. Christopher Dodd trailing Republican challenger Rep. Rob Simmons, 39 percent to 45 percent. The same poll shows Dodd leading primary challenger Merrick Alpert -- but with less than 50 percent of the vote (44 percent to 24 percent), even though 91 percent of Connecticut's registered Democrats say they don't know Alpert well enough to rate him.
Pondering these anemic numbers, Nate Silver declares Dodd "vulnerable to primary challenge." Given that Dodd's support is well under 50 percent -- a common rule of thumb denoting vulnerability -- Silver's assessment is reasonable.
But then what about Pennsylvania, where another Quinnipiac poll finds newly Democratic Sen. Arlen Specter running slightly ahead of Republican challenger Pat Toomey, 46 percent to 37 percent. The same survey shows Specter leading likely Democratic primary challenger Rep. Joe Sestak with slightly more than 50 percent of the vote (50 percent to 21 percent).
No rule of thumb I know of would have labeled Lieberman vulnerable in early 2006.
Contemplating the poll numbers, liberal blogger Chris Bowers describes Specter as "vulnerable" to Sestak's challenge, while conservative blogger Jay Cost discounts Specter's early lead as the result of voters' "lack of familiarity" with Sestak. Republican commentator/consultant Mike Murphy predicts that Specter would lose a primary against Sestak.
So just how low do poll numbers need to be do make an incumbent "vulnerable"?
It is tempting to try to reduce vulnerability to a simple formula, something many pollsters have done over the years (yours truly being no exception). The most frequently cited definition has been electoral support below 50 percent, although some pollsters look at horse-race questions featuring a named opponent, some look at "re-elect referendum" questions that ask only about the incumbent, some extend the exercise to job-approval ratings and some prefer thresholds lower than 50 percent.
The basic idea behind these heuristics is sound. Challenger candidates typically begin with low levels of name recognition (like Sestak and Alpert) that artificially inflate the incumbent's lead on early "horse-race" vote preference questions. Well-funded challengers typically rise in tracking polls as they gain recognition. So it makes sense to ignore the challenger's number in the early polling and focus instead on the level of support for the incumbent.
Unfortunately, the various rules of thumb about what number makes for a vulnerable incumbent are just rough guides. Each situation is different. Sometimes the incumbent's number on a horse-race or re-elect question provides a good measure of the support the incumbent ultimately receives; sometimes it doesn't.
Consider Joe Lieberman's poll standing three years ago. In January 2006, the Connecticut senator had a 55 percent approval rating among registered Democrats in the Quinnipiac poll (29 percent disapproved), and 59 percent of Democrats said Lieberman deserved to be re-elected (24 percent said he did not). A subsequent Quinnipiac poll conducted in May 2006 showed Lieberman leading Democratic primary challenger Ned Lamont by 46 points (65 percent to 19 percent).
On every measure, Lieberman was well over 50 percent. No rule of thumb I know of would have labeled Lieberman vulnerable in early 2006. Yet when the votes were counted in Connecticut's Democratic primary that August, Lamont defeated Lieberman by four points -- 52 percent to 48 percent. (Lieberman went on to win re-election as an independent.)
One reason for the misleading early numbers in 2006 may have been that Quinnipiac sampled all self-identified registered Democrats rather than a narrower subset of likely primary voters. Their May 2006 sample of 528 Democrats, for example, amounted to 34 percent of the full sample of 1,536 registered voters they interviewed. Yet the actual Democratic primary turnout amounted to just 15 percent of Connecticut's active registered voters.
The odds are good that the true "likely voters" felt more antipathy to Lieberman than other Democrats did. By focusing on the wider universe of registered Democrats, the Quinnipiac polls in early 2006 may have made Lieberman seem less vulnerable to a primary challenge than he turned out to be.
The Connecticut experience from 2006 is important in assessing the two polls last week in Connecticut and Pennsylvania, because Quinnipiac is again reporting on the preferences of all registered Democrats, a much larger slice of each electorate than is likely to participate in the 2010 primaries.
Quinnipiac's recent samples of registered Democrats, for example, amount to 39 percent of the full sample of registered voters in Connecticut and 47 percent in Pennsylvania. What are the odds that a Dodd-Alpert primary in Connecticut will match the 15 percent turnout of registered voters seen in the hotly contested 2006 Lieberman-Lamont primary? Low, if you ask me.
Voters may show more interest in a Specter-Sestak primary in Pennsylvania, but higher than the 2002 Ed Rendell-Bob Casey gubernatorial primary, when, again, only 15 percent of that state's registered voters turned out to vote?
The lesson is that we need to interpret these early primaries with a sharp eye on how narrowly each survey screens for likely primary voters. And given that the two Quinnipiac polls come nowhere near a true "likely voter" sample, here's a hunch guided more by intuition than data: Between Dodd and Specter, the more truly vulnerable to a primary challenge may be Specter, the one with the better numbers.
Why? Because even though Pennsylvania Democrats seem to be rallying to Specter, his record and statements give Joe Sestak much substance to work with in questioning Specter's Democratic bona fides, particularly among the hard-core partisans who tend to vote in primaries. In Connecticut, on the other hand, Dodd may be better positioned to rally the party faithful. But that's just a hunch.