Another week, another outlying Gallup generic ballot result. Gallup finds the Republicans with either an 11- or 17-point lead, depending upon the likely voter screen. Compare that to a Republican advantage of a little under 7 points in the RealClearPolitics average.

What is going on here? I think there are two factors at play.

First. As Mark Blumenthal points out, the traditional Gallup likely voter model imposes a relatively tight screen on who gets included as a likely voter and who doesn’t. Imagine a sinking ship with 100 passengers on it, and one lifeboat that will take 40 people. Who gets on? If the passengers throw out the old "women and children first!" maxim, and a mad scramble ensues, the answer is: the most industrious and committed! That’s basically Gallup’s “traditional” model. It assumes a 40% turnout rate of the voting age population, and includes the 40% of adults who appear to be the most committed to voting. The higher turnout model sets the cap at 55%.

What to make of this? Gallup has history on its side when it caps turnout at 40% of the voting age population. According to Walter Dean Burnham (and when it comes to electoral statistics, there is no higher authority), turnout in the midterms since the ratification of the 26th Amendment has been amazingly consistent:

Michael McDonald at George Washington University has compiled slightly different turnout results, which tend to show lower numbers than Burnham's, but they are basically the same.

Why isn’t it an open-and-shut case vis-à-vis turnout? Cap it at 40% and call it a day. What’s wrong with that?

The chief problem is that, unlike our lifeboat metaphor, turnout could go above 40%. And just because it hasn’t happened in 35 years does not mean that it will not.

I’d say there are some good reasons to expect a greater-than-average turnout this cycle. Note that in the last three midterms, we’ve seen a steady increase in turnout, so another increase this cycle would follow the recent trend. But more than this, we already know there is an amped up Republican base that is going to turn out a higher than normal rate. Meanwhile, the Democrats remain a big question mark. If they turn out at a correspondingly lower-than-normal rate, we should not see any change in turnout. But if the Democratic decline is less than the Republican increase, then we will see an increase in turnout.

Time will tell, and my guess is that Gallup and the other polls will converge at some point before Election Day, though it remains to be seen which side will do most of the movement.

This actually underscores the importance of keeping an eye on the party spreads in the polling data. Gallup’s traditional model might be overestimating Republican turnout, but I think many statewide polls (and some national ones) are underestimating it. In particular, it is appropriate to be cautious of polls that mimic the 2008 party spreads in any state (or, for that matter, nationwide). That was a year for which Republican turnout was low and Democratic turnout was high. We are probably not going to see that this year.

And remember to always be an informed consumer of polls! Bookmark the 2008 (a great Democratic year), 2006 (a good Democratic year), and 2004 (a slightly good Republican year) exit poll sites -- and before you accept any poll, cross-reference what it predicts the electorate will be with what it was in those past years. This is so important this year because I think polls are increasingly being used to move public opinion rather than to inform us about it.

Second. While Gallup shows a larger Republican lead, it also has fewer undecided voters. This could help explain why it’s showing such a bigger than normal Republican result. Compare the two latest Gallup numbers to the other polls currently in the RealClearPolitics average.

While Gallup has the highest number of Republican supporters, it also has the lowest number of undecided voters (other than CNN). These two factors could be related. If the undecided voters in the other polls are actually leaning softly to the GOP, and Gallup is pushing them to name a preference, then the outsized Republican advantage might not be that far from the norm.

To appreciate this, let’s do two things. Let’s drop the GOP’s best and worst polls from the average – the Gallup high turnout model poll and the Bloomberg poll. This will just give us the seven central polls. Also, let’s allocate the undecideds proportionally, i.e. let’s assume that the voters yet to decide will break the way the pollster thinks the decided voters broke. In that scenario, we get this.

In this scenario, Gallup is not far from the average result. It’s on the high end, sure, but it’s not any higher than Reuters/Ipsos is low.

Conclusion. Let’s put aside the wonky considerations, and note some pragmatic points. I think there are two. The first, as I mentioned, is that the partisan composition of the electorate remains the critical unresolved issue of this cycle. Every pollster is making a guess as to what the electorate will look like, and these guesses are at least as important as their final numbers. Be aware of this.

The second point is that this could be a cycle that ends up confounding the pollsters. The fact that Gallup – the oldest and most respected public polling firm in the country – is offering two likely voter models (with just two weeks to go, no less!) is a sign that there is as yet no consensus on what the final electorate will look like. So, it is especially appropriate this year to be a cautious and judicious consumer of public opinion polling.

Next Page