Monday, November 01, 2010

5 Reasons Democrats Could Beat the Polls and Hold the House - NYTimes.com

FiveThirtyEight - Nate Silver\'s Political Calculus

November 1, 2010, 8:20 pm — Updated: 8:20 pm -->

5 Reasons Democrats Could Beat the Polls and Hold the House

It was hard to pinpoint exactly when in the night things started to go wrong. But at some point, a trash can was knocked over in John Boehner’s office in the Rayburn House Office Building. A half-hour later on the other side of town, a hole was punched in the wall at the Republican National Committee’s headquarters.

Republicans didn’t really have much reason to be upset. They were going to pick up somewhere between 29 and 34 House seats from Democrats, pending the outcome of a recount or two and the receipt of mail ballots in some Western states. They gained five Senate seats from Democrats, and won the governorships in Michigan, Wisconsin, Pennsylvania, Texas, and Florida, among many other states. It had been a wave election, indeed — but a wave about on the magnitude of 2006, rather than 1994.

For most of the evening, Republicans had still seemed quite likely to pick up the House, perhaps by some margin. Exit polls that (erroneously, it turned out) suggested a nine-point generic ballot win for the party colored the early coverage. So, when Baron Hill, the vulnerable Democrat in Indiana’s 9th district, held onto win his seat by a surprisingly robust 9-point margin, it was mostly ignored. Instead, coverage was focused on the dozen or so Democratic incumbents who lost their races early in the evening — some of them expectedly so (like Alan Grayson and John M. Spratt Jr.), but others of which (like Gerry Connolly of Virginia and Chellie Pingree of Maine) were more surprising.

In states like Michigan, Ohio and Pennsylvania, however, Democrats held up surprisingly well. Mary Jo Kilroy, who had been all but written off, held her seat in Columbus, as did a trio of Democrats, Christopher Carney, Bryan Lentz and Patrick J. Murphy, in Pennsylvania. Ted Strickland won the gubernatorial race in Ohio, and Joe Manchin III was elected to the Senate in neighboring West Virginia (by double digits, in fact). Joe Sestak appeared to have upset Pat Toomey in the Senate race Pennsylvania, although the Associated Press had yet to call the race because of alleged irregularities in Philadelphia.

New York was another problematic state for Republicans: their gubernatorial nominee, Carl P. Paladino, was defeated by almost 40 points, and of the six or seven House seats they had hoped to win there, they had instead picked up just one, while another — the upstate 20th district — remained too close to call.

Still, the gains came steadily, if not quite steadily enough. Michael Bennet lost his Senate race in Colorado — taking Representatives John Salazar and Betsy Markey with him — even as more vulnerable-seeming Democrats, like Alexi Giannoulias of Illinois and Harry Reid of Nevada, held on. The Dakota Democrats, Earl Pomeroy of North Dakota and Stephanie Herseth-Sandlin of South Dakota, were both defeated, converting 148,000 square miles of territory from blue to red.

But Republican gains not only stalled out but reversed themselves by the time that the West Coast began to report its results. Not only had vulnerable Democrats like Jim Costa and Kurt Schrader held on, but Democrats had defeated two Republican incumbents — Dan Lungren of the California 3rd district and Dave Reichert of the Washington 8th — while also narrowly winning the Arizona 3rd district, where the G.O.P. nominee Ben Quayle had proven too difficult a sell. Overall on the night, Democrats won 8 seats formerly held by Republicans, about twice what most analysts had expected.

The Senate race in Alaska, meanwhile, as some had feared, appeared headed toward a prolonged legal battle concerning Lisa Murkowski’s write-in votes. But the plaintiff would Scott McAdams, a Democrat, and not Joe Miller, a Republican.

*-*

A scenario like this one is possible tomorrow — not particularly likely, but possible, just as a 77-seat Republican gain is possible. It’s probably a somewhat greater possibility than people realize. Here are five reasons Democrats could outperform their polls and beat consensus expectations.

1. The cellphone effect. This one is pretty simple, really: a lot of American adults (now about one-quarter of them) have ditched their landlines and rely exclusively on their mobile phones, and a lot of pollsters don’t call mobile phones. Cellphone-only voters tend to be younger, more urban, and less white — all Democratic demographics — and a study by Pew Research suggests that the failure to include them might bias the polls by about 4 points against Democrats, even after demographic weighting is applied.

There is also some indirect evidence for the cellphone effect. What follows is a list of each firm’s final generic ballot poll, arranged from the best result for Democrats to the worst. You can see that there is a rather strong relationship between whether a company included cellphones in its sample or not and the sort of result they showed. The polls that were conducted without cellphones showed Republicans ahead by an average of 9.3 points; those with them showed a smaller, 4.8-point advantage. That’s a difference of 4 or 5 points (and one which is statistically significant at the 95 percent confidence threshold), which is about of the same magnitude that Pew identified.

Now, this probably does not mean that Democrats are bound to overperform their polls by four or five points. A fair number of polls do include cellphones, so at best it might be half that. And the effects probably aren’t so uniform from company to company. Still, this is a theory that has a fair amount of evidence behind it.

2. The ‘robopoll’ effect. Unlike in past years, there are significant differences between the results shown by automated surveys and those which use live human interviewers — the ‘robopolls’ being 3 or 4 points more favorable to Republicans over all, although the effects vary a lot from firm to firm.

Automated surveys, while they have performed fairly well in the past (although in the past, importantly, they did not show these systematic differences from regular surveys), have a number of potential problems that essentially boil down to extremely low response rates, which could potentially bias the samples. For instance, it may be that only adults who are extremely engaged by politics (who are more likely to be Republican, especially this year) bother to respond to them.

Most automated surveys also do not call cellphones — although that’s the bad news for Democrats in some sense: the ‘robopoll’ effect, if there is one, may have a lot of overlap with the cellphone effect, or they may even be one and the same thing.

3. Some likely voter models, particularly Gallup’s, may “crowd out” Democratic voters. Gallup’s traditional likely voter model has consistently shown terrible results for Democrats this year, having them down by around 15 points on the generic ballot, which could translate into a loss of 70 to 80 House seats, or maybe even more. The Gallup poll and the Gallup poll alone is probably responsible for much of the sense of impending doom that Democrats feel and the (premature for at least 24 more hours) sense of triumphalism that Republicans are experiencing.

There is quite a bit of room to critique the poll, however. The basic potential issue is that Gallup uses fixed turnout targets. For instance, they estimate that 40 percent of the electorate will vote, and then let their respondents fight it out to see who the 40 percent most likely to vote are.

So, for instance, if you have a lot of Democrats whose likelihood of voting is a 9 on a 10-point scale — people who might ordinarily be quite likely to vote — they’ll be excluded from Gallup’s likely voter sample if too many slots are occupied by perfect 10’s (who in this cycle, no doubt, tend to be Republicans). This is why, when Gallup digs just a little bit deeper into its voter universe — for instance, with its “higher turnout” model, which assumes that turnout is 50 percent rather than 40 percent — its results quickly shift from being something of an outlier to being fairly consistent with the other generic ballot polls. There are a lot of Democratic 9’s just beneath the surface of those Republican 10’s.

I don’t know that this is very realistic portrayal of how voting takes place in the real world. Except, I suppose, in the case of extremely long lines at the polling place, voters are not competing with one another to vote in any real way: the only-somewhat-enthusiastic Democrat is not denied her vote because the Republican in line ahead of her is even more psyched up.

And Gallup’s model can produce some strange effects. Robert Erikson of Columbia University has found, for instance, that the preferences of “unlikely voters” in their sample move in an opposite direction from those of its likely voters, which does not seem sound.

The counterargument would be that midterm election turnout has indeed been extremely stable at about 40 percent of the population, so Gallup is on solid footing in assuming that it will be somewhere in that range again.

But there is some evidence that turnout might be unusually high in this election: there is no doubt that Republican engagement is likely to be extraordinary, but Democratic involvement, also, in fact appears to be about average or slightly above.

Most polls will also have some bias in who they reach and who they don’t. Specifically, they’ll have trouble reaching people who are less likely to vote, who usually won’t have much interest in completing a political survey either. Perhaps it’s the case that 40 percent of the general American population will vote tomorrow — but, because of this response bias, perhaps the rate would instead be something like 60 percent of the adults that Gallup actually got on the phone.

Unless you have a really good idea of exactly what effect this response bias might have (and it would be hard to have a really good idea, because this type of bias is probably increasing as response rates to surveys decrease), using fixed as opposed to fluid turnout targets can be dangerous.

Gallup gets a lot of deference, because it is one of the best polling organizations in the world, and because its likely voter model has done very, very well in predicting the outcome of past midterm elections.

The fact is, though, if you took the various likely voter models, and put them in a lineup without their brand names attached, Gallup’s is probably not the one you’d pick out of that lineup as having the most robust design. And while Gallup’s likely voter model has done very, very well in Congressional elections in midterm years, it has been only about average in Congressional elections in Presidential years, and in other types of elections.

*-*

So the rose-colored glasses scenario for Democrats looks something like this: throw out the generic ballot polls that don’t include cellphones. And then throw out Gallup, because there is something weird about the way its likely voter model is behaving this cycle.

The average of the nine remaining generic ballot polls in the table above is a Republican advantage of 3.9 points.

Democrats could possibly hold the House with a number like that one, although they’d be underdogs to do so. Most likely, they’d need another point or two. Where could they get it?

4. Democrats probably have better turnout operations. This is always what a party says when it’s about to lose an election: our amazing turnout operation will save us!

Still, Democrats probably do have an edge in this department with the voter lists and infrastructure they built up during Barack Obama’s campaign, and which have been perpetuated to some extent by Organizing For America. John McCain, by contrast, eschewed his ground game, devoting almost all of his money to advertising.

Now, Republicans may not need a terrific turnout operation — their voters are charged up enough, and probably don’t need a lot of glossy fliers and door-knocks.

Nevertheless, Democrats might be able to coax an extra percentage point or two of their vote to the polls, especially in states like Ohio and Pennsylvania where they’ve invested a ton of resources over the years. And in the event where Democratic turnout equaled that of Republicans (it won’t; the point is they might be able to get it a bit closer), they would probably hold the House, even with most independents breaking against them.

5. The consensus view of Democratic doom is not on such sound footing as it seems. When a party is likely to sustain fairly significant losses in a midterm election — and Democrats are going to sustain fairly significant losses tomorrow — there are a lot of things you might expect to see.

First, you would expect to see that the party’s generic ballot polls were bad. Some of them might be really bad. A handful of them might be not-quite-as-bad-as-the-others, and might provide some hope to the faithful. Most of them would be pretty bad, though.

Second, that party would get lots of bad results in polls of individual House districts.

Third, that party’s problems are going to manifest themselves among most, and possibly virtually all, demographic groups. So you can write a story about how the party is struggling among women. And among men. And young voters. And old voters. And rich voters. And working-class voters. And Hispanics. And whites.

Fourth, that party’s numbers are going to be especially poor among independent voters. Except in places where party identification is extremely lopsided, the party which is losing in an election — and Democrats are losing this election — will usually be losing by a larger lead among independents. That’s just the way the math works.

Okay, let’s stop there. So, what’s my point?

Each of the indicators that I mentioned above are direct manifestations of polling data. The message in the polls this year is unambiguous: bad things are going to happen to Democrats. The polls are probably going to be right.

It seems like the evidence that Republicans will win the House is very rich, redundant, robust. Look at this generic ballot poll. Look at this other generic ballot poll! Look at how badly Democrats are doing among whites. Look at how they’re doing among independents!

But all of these indicators are, in fact, highly correlated with one another. They’re all rooted in the polling, and they’re all dependent on the polling basically being accurate. There’s not much diversity at all: it’s just different manifestations of the same thing.

Our Congressional forecasting models are based on an intensive study of six political cycles: 1998, 2000, 2002, 2004, 2006, 2008. In five of those six years, the polls were quite good — they missed a few races, but were very strong overall.

In another year, however –1998 — they were quite poor. Democrats overperformed their polls by about four points in a great number of races around the country. What was supposed to be an echo to the Republican boom year of 1994 basically flopped, eventually costing Newt Gingirch his job as majority leader.

Consensus expectations also considerably underestimated the Republican wave year of 1994, although a few indicators (like Gallup’s generic ballot poll) got it about right.

If we wanted to be generous to Democrats (which is, of course, the purpose of this article), we could say that the consensus basically failed in two out of the last four midterm elections. Of course, that the consensus view could fail does not mean that it will fail in the Democrats’ direction: instead Republican gains could be much larger than expected.

But in my view, it doesn’t make sense to say, for instance — and a lot of people are saying things like this — that Republicans should gain between 50 and 60 seats, and the number could be higher, but it shouldn’t be lower. If that’s what you think, you should project their gains to between 60 and 70 seats (or whatever) instead.

The thing is, the upside case for Republicans is pretty easy to see. Most of the news in this election, after all, is favorable to them. You see the Gallup generic ballot number, you see incumbents like Raul M. Grijalva and Jim Obertstar in trouble, you see the president’s approval rating at 44 percent, you see the big crowds at Tea Party rallies, you see Scott Brown winning in Massachusetts, and it’s easy to connect the dots. It’s an easy case to make, and it’s a pretty good one.

The case that Democrats could do better than expected — not well, by any means, merely better than expected — rests a little more in the realm of what artists call negative space: not what there is, but in what there isn’t. There aren’t 50, or even more than about 25, districts in which Republican candidates are unambiguous favorites. There isn’t agreement among pollsters about how the enthusiasm gap is liable to manifest itself. There isn’t any one poll or one forecasting method that is clairvoyant, or which that hasn’t made some pretty significant errors in the past.

Instead, the case for Democrats is basically: yes, the news is bad, it just isn’t exactly as bad as you think, or at least we can’t be sure that it is. This isn’t a sexy argument to make.

Nor, probably, will it turn out to be the correct one; more likely than not, Republicans will indeed win the House, and will do so by a significant margin. But just as Republicans could beat the consensus, Democrats could too, and nobody should be particularly shocked if they do.

DCSIMG

No comments:

Post a Comment