AR-Sen: New Polling Numbers

    53

    A new poll from Arkansas News/Stephens Media (MOE ~5%) shows Blanche Lincoln with a 44-32 lead over Bill Halter.  D.C. Morrison showed up with 7%, and 17% remained undecided.

    LITTLE ROCK — A new poll commissioned by the Arkansas News Bureau/Stephens Media shows U.S. Sen. Blanche Lincoln, D-Ark., with a 12-point lead over Lt. Gov. Bill Halter, the chief Democratic challenger to Lincoln’s re-election bid.

    [***]

    The sample included voters in all four of the state’s congressional districts, with quotas assigned to reflect voter turnout by county. The poll has a margin of error of plus or minus 4 percentage points except on questions about the primary races, where the margin of error is plus or minus 5 percentage points.

    When asked which Democratic U.S. Senate candidate they will vote for — or already did in early voting — 44 percent of likely Democratic primary voters picked Lincoln, 32 percent picked Halter, 7 percent picked D.C. Morrison and 17 percent were undecided.

    The gap between Lincoln and Halter has narrowed since January, when a poll conducted by the same firm, with a 303-person sample and a 6 percentage-point margin of error, showed 52 percent of likely Democratic primary voters supported Lincoln and 34 percent supported Halter. Morrison was not included in that poll.

    With Morrison now in the mix and Lincoln’s numbers sliding, a Democratic primary runoff appears “very plausible,” said J. Brad Coker, managing partner of Mason-Dixon.

    “She was at 52 in the last poll, and I would have said no way, she’ll be all right, but she’s come down into the 40s and she may not be able to get up over 50 percent,” Coker said.

    The gap of 12 points is larger than what has been recently reported elsewhere.  If you recall, DailyKos/Research 2000 had a poll about two weeks ago that showed an 8-point race between the two candidates (MOE 5%). So, what gives?

    The best I can tell, based on each poll’s description of its methods, the difference comes from one slight difference in methodology.  In the Arkansas News poll, as mentioned above

    The sample included voters in all four of the state’s congressional districts, with quotas assigned to reflect voter turnout by county.

    In the DK/R2000 poll, however

    Quotas were assigned to reflect the voter registration of distribution by county.

    That difference — voter turnout (based on previous elections, I assume) vs. voter registration — is subtle, but important.  Ideally, a poll wouldn’t force the responses in each county to be proportionate to either population, the reason being that it eliminates part of the randomness.1 However, if you are going to force quotas, doing so based on the raw total of potential voters in this election in each county is preferable to doing it based on how previous elections have turned out.

    Why?  Two reasons, actually.  First, the pool of potential voters gives you a better sample from which to work because it is a broader cross-section.  (i.e. In each district, you have a population; within that population, you have registered voters; and within those registered voters, you have voter turnout.)  Whereas the Arkansas News poll tried to be random within the smaller subset, the DK/R2000 poll is trying to be completely random within the population of registered voters.  To the extent you are going to force some kind of sampling quota, I don’t have a problem with this approach simply because those not in this population are irrelevant to what we’re trying to estimate.  Further, by not artificially forcing the voter turnout numbers, you’re letting the population of registered voters decide whether or not to answer, which more closely reflects the voting process.

    [Side note: The Arkansas News poll doesn’t say whether or not the voter turnout sample was selected from the group of registered voters or from the general population.  I would hope it’s from the registered voters, but surely you can see the inherent problem if that wasn’t the case.]

    Another way to look at it is this:  The whole point of polling is trying to estimate how people are going to vote.  The most accurate way to do this would be to hold an election, but, for obvious reasons, you can’t do that.  Instead, you estimate that by organizing a poll.  Wouldn’t you want this poll to reflect that election process as closely as possible?  DK/R2000 gets closer to this ideal.

    The second reason that the difference in methodology is important is that a quota based on how previous votes turned out carries with it unnecessary inherent flaws.  By which I mean, such a quota is subject to all the externalities of those previous elections, none of which will be exactly the same this time around.  Maybe, as in 2004 when the Presidential primary was included in the preferential primary, one candidate in a different race was highly popular (or highly unpopular), which skewed the turnout for some counties.  Or maybe, as in 2008, all of the statewide candidates were running unopposed, so the results are skewed based on which district or county races showed the largest turnouts.

    A quota based on the potential voters in 2010, however, removes previous exogenous variables and is subject only to the potential skew related to removing part of the random nature of the sample by forcing totals within each district (which, in addition to the above inherent flaw, a quota based on turnout would also have).

    With the results being so similar, why does all of this matter?  Because the Arkansas News poll has the race slightly outside the margin of error, while the DK/R2000 poll has the race within the margin.  As discussed here, the latter poll would suggest that it really is almost a pick’em at this point, while the former would suggest that voters actually prefer Lincoln.  Given the methodology of both, not to mention the underlying approval numbers for both candidates, I tend to believe that the DK/R2000 poll is the more accurate of the two.

    Of course, all of this discussion is more or less moot if we go to a runoff, so the only number that really matters in all this is Lincoln’s as the “leader;” if she doesn’t show 50% +1, second place is as good as first place on May 18th.  As both polls show her under 45%, a runoff appears to be all but certain.  My best guess (as of this writing) is that Lincoln “wins” the primary by ~4.5 and we are still talking about the campaign on May 19.

    Update: Pointed out to me after I posted this was that the Arkansas News poll was commissioned by Stephens Media.  The Stephens Group is Lincoln’s third-largest campaign backer.  Given the flaws inherent in the poll, I don’t think it is at all unreasonable to suggest that the “voter turnout” subset was a deliberate, results-oriented choice.

    1 I know that sounds kind of contradictory, but if your sample size is large enough, you are going to get a pretty accurate proportion on each county anyway. Generally speaking, you always want to err on the side of randomness.