Wolraich: Obama at the Gates of... Gates
Dr. C: In Praise of Writing Binges
Maiello: Gatsby Doesn't Grate
With twenty-four days to go until the final votes are cast, it is clear that the Obama-Romney contest has narrowed drastically. The number of states that are “in play” remains small, and the number of states that could realistically decide this election is smaller still. Taking a close look at data from the last week and reading it against what we know from 2010 gives us a window into how narrow this contest is, where it will be decided, and likely how. As I will explain, I am more bullish on President Obama’s likelihood of winning re-election than others caught up in the national trackers, Nate Silver’s plummeting assessments of the President’s prospects, or the gloomy post-debate narratives (Andrew Sullivan being the undisputed champion of that particular raindance). So here – with close attention to polls and what they do and do not show – is why I see President Obama’s chance of winning re-election at roughly 80% today.
1. Parsing Romney’s Post-Debate Bounce
Even the President’s most ardent supporters now readily concede that Mitt Romney won the October 3 debate in Denver. (Indeed, some of the President’s most ardent supporters can speak of little else, as this election approaches its finish.) The measure of the bounce is found in a comparison of data drawn from mid-September (when the President’s convention bounce, lengthened by focus on Romney’s Libya and 47% comments) and from the period October 4-11, comprising the week following Romney’s strong showing in Denver.
This bounce was substantial, but not uniformly distributed across all states. While I first thought it would be worth two points nationally, given the rigidity of polling in this cycle, my prediction understated the force of Romney’s gains. In Gallup, the registered voter bounce was from Obama +6 to Obama +3 (though the fact that Gallup uses a seven day rolling average implies that the bounce was at least four points, given that the long period softens peaks and valleys). In Rasmussen, Obama swung from +2 to -2 – a four point decline. In the RAND Corporation tracker, mysteriously omitted from the RealClearPolitics average, but which Nate Silver utilizes in his data set, Obama fell from +8 to +2 (a six point decline), and in the IBD/TIPP tracker, which began on October 1, Obama fell briefly to -5, likely representing at least a six point decline from Obama’s peak standing to that lowest point (though IBD bracketed that five point deficit with days showing Obama down 2). Accordingly, the national decline in Obama’s status, picking his highest peak and lowest valley, was more like 4-6 points nationally.
While Florida Bounces the Most, Other Swing States Simply Do Not Experience These Changes.
While Romney showed gains almost everywhere, he did not experience gains in swing state commensurate with his national bounce. Grouping these is helpful.
The Full Monty: Romney’s Six Point Gain in the Sunshine State. There is no question that post-debate polls of Florida have been very good for Romney. ARG and Mason-Dixon showed 8 point declines for Obama, with Romney now up 3 and 7, respectively. Rasmussen showed a six point gain for Romney, from down 2 to up 6. Finally, Marist showed no decline, with Obama still up 1. The picture is one of Obama moving from up 3 to down 3, consistent with the national bounce.
The In-Betweeners: Colorado, Virginia, New Hampshire, Wisconsin, Nevada Move Two to Three Points. While scoring strong gains in Florida, Romney seems to have moved ahead narrowly in Colorado, but by 2-3, and not 6, points. ARG showed Obama falling from +2 to -4, consistent with GOP-leaning Gravis Marketing showing Obama falling from +4 to -3. In the same period, however, both Quinnipiac and SurveyUSA showed Obama falling from +1 to -1, and Rasmussen showed Obama gaining from -2 to +1 after the debate. The picture in Colorado is thus much closer to a 2-3 point swing toward Romney, with the race there showing Romney up by 1 or so. Virginia has moved about like Colorado. There, Marist shows Obama moving from +2 to -1, PPP showed a move from Obama +5 to +3, Rasmusssen showed Obama moving from +2 to -1. These 2-3 point moves are bracketed by symmetric outliers: Quinnipiac showed Obama gaining a point (+4 to +5) while We Ask America showed him losing 6 (+3 to -3), which amount to the same average of 2-3 points of motion toward Romney. Additionally, lightly polled New Hampshire showed Romney up 4 and tied in post-debate polling, which meant either a 7 point gain for Romney (ARG showing Obama falling from +3 in the average of its September surveys to -4) or a 3 point gain for Obama, from -3 to tied in Rasmussen. This averages to a 2 point gain for Romney, in a state which (excluding high and low results was +4.5% for Obama in September). Obama is likely slightly ahead there. Traditionally narrowly Democratic Wisconsin has moved roughly like these other states in the three post-debate polls. PPP shows the biggest Obama decline (+7 to +2), while Quinnipiac shows a 3 point drop from +6 to +3. Rasmussen shows a one point drop from Obama +3 to +2, thus showing a 3 point average drop into a 2-3 point Obama lead, with essentially no undecideds. Nevada is harder to assess because of light polling, but we have Obama falling 5 in PPP (from +9 to +4), falling 2 in Rasmussen (from +2 to even), and other surveys showing Obama up 2, 1, and 1, which can be compared roughly (not apples to apples, like the rest of this piece) with other one-off surveys averaging to a 5 point Obama lead. In sum, Nevada has moved about 3 points toward Romney, but he has not led in a single poll there since early 2011.
The Hardened Electorate: Ohio Is Barely Affected. Romney’s 6 point national bounce has not showed up in Ohio. PPP shows Obama up 1 (from +4 to +5), Rasmussen has shown Obama up 1 in its last three polls across September and October, Marist shows Obama down 2 (from +8 to +6), as have GOP-leaning Gravis Marketing and ARG (both showing Obama moving from +1 to -1). The polls without counterparts on both sides of October 3 are more equivocal, with Obama up 4 in the post-debate CNN/Opinion Research poll, while other polls before the debate by pollsters not again in the field showed Obama up 3, 4, 5, 7, 8, 9, and 10. Based on these data, we can see that the average of change in apples-to-apples polls at less than one point toward Romney, the smallest among any swing state. The polls in the last week tell a story not much changed from Nate Silver’s pre-debate model of Ohio at roughly 51-47.5. Until other pollsters return to the field and show an apples-to-apples swing from mid-September to mid-October, it is hard to argue that Ohio has swung much toward Romney.
2. The Boundaries of This Race’s Likely Outcomes Favor President Obama.
As I have argued before, this election exists in a space bounded by the best possible Obama outcome (a six point or so win, consistent with his highest average national tracking poll status to date), and the best possible Romney outcome (a one point or so win, consistent with his highest average national tracking poll status to date). In that sense, President Obama was overperforming as a candidate throughout mid-September, and Mitt Romney has overperformed as a candidate likewise in the last week. This leads us to different points that lead to the same conclusion: structural factors in this race, however considered, favor President Obama.
First, there is the obvious point that if the band of possible outcomes is consistent with an Obama +6 to Romney -1 world, there are simply far more possible outcomes in favor of an Obama victory. In a broader sense, this is why Nate’s model has not yet gone below 61% in favor of an Obama victory, and why it, flush with a September bounce, rose to a high of 87% in its synthesis of polling data. But this is a more superficial way of looking at things. You can parse strands in the data, and they lead to the same place.
Second, put another way, Obama has a greater likelihood of making marginal gains by simply performing adequately as a candidate, while Romney has a strong need to continue his overperformance even to maintain his status as within a point or so. You can see this in the nascent data set concerning the Biden-Ryan debate. Vice-Presidential debates have little effect on self-reported voting decisions, as Gallup hastened to point out. True enough. But the Vice-Presidential debate turned the first Obama-Romney debate into old news, and old news cycles. And more than that, it provided a feisty, widely-viewed statement of the Obama-Biden case that was preponderantly viewed as Biden victory. Despite a poll of more-Republican-than-the-electorate debate watchers by CNN favoring Ryan 48-44, the rest of the data ran the other way: a CBS News poll of undecided showed Biden winning 50-31, Reuters-Ipsos showed Biden winning 42-35, and the new PPP poll of Ohio shows a Biden victory by that margin, with even a narrow plurality of Romney-Ryan voters thinking Biden won. And with Biden modestly outperforming Ryan, and the prior debate relegated to past-event status, the RAND tracker, which reports daily results, has shown a 2 point uptick (from +2 to +4) in its wake, while IBD has shown Obama up 1 over the last three days.
Conceding that the Vice-Presidential debate was an event of modest significance (though viewed by 51 million people), the fact that it has tended to move the data to the midpoint of the range of plausible outcomes reinforces the idea that it is easier for Obama to gain than for Romney to do so. If Obama by conventional wisdom roughly “ties” Romney on Tuesday, the effect will be the same – tending to center outcomes in the middle of the -1/+6 range that defines these two candidates’ best days. Not only is the media primed for a comeback narrative that would tend to ratify such a move (or settling) in the data, but Obama himself by all reports plans more assertively (well, any assertively would be more) to challenge Romney on his core vulnerabilities and points of clash, making that outcome more likely as well.
Third, and finally in this regard, another way of looking at the concept of Obama performing at the lower portion of his range of outcomes is that certain concrete parts of his performance are amenable to improvement. In the polls showing Romney up modestly on Obama after the debate, Romney moved to a position of leading among independents by close to 20 points. Currently, IBD shows Romney’s advantage moderating to something more like 8 or 9 points. Yet the massive advantage – likely Romney’s best day with persuadable voters – is necessary for him to lead by 1-2 in national tracking polls. Put another way, Obama is leading in IBD with only 36% of white voters in his corner. That would be far lower than Obama’s own or Kerry’s performance among white voters. One could far more readily anticipate Obama gaining a point or two there than losing even more. Again, the concept of a range and limits comes into play, and Romney cannot rely on winning 65% of the white vote or winning independents by 20% or more. Romney approached these marks with a 72-20 margin on whether he won a debate, perfect storm figures he is unlikely to replicate and which still have not put him in an Electoral College lead.
3. Either Obama Has an Electoral College Advantage, or the National Polls Are Wrong, And Whichever Is the Case Makes His Victory More Likely Than It Presently Appears.
In our highly polarized nation, there have been wars in the media about what the facts concerning voter behavior are. There are Republicans who argue that national polls oversample Democrats and that Mitt Romney is really ahead by 5 or 7 points nationally, and who argue that some pollsters, like PPP, are Democratic happy talk. There are Democrats who argue that national tracking polls by Gallup and Rasmussen oversample Republicans because they do not sufficiently describe the voting intentions of younger voters, who are far more likely not to have land lines, of lower-income voters, who are modestly less likely to answer surveys, and Latinos, who seem to have been undersampled in recent elections. A related point of conflict – grounded in your view of the polling data – is whether President Obama has an advantage in the Electoral College, such that you would expect him to win an election tied in the national popular vote. Democrats tend to argue yes, Republicans tend to argue no.
Synthesizing the data, I think the neutral and intellectually honest conclusion is one that Nate Silver has not thus far stated, though you would expect him to say it first: either President Obama really has an Electoral College advantage, or the national trackers are modestly off. Here is why.
Consider first the bases each side begins with. Consider that President Obama is firmly ahead in the Kerry states plus New Mexico but minus New Hampshire (where he leads, but by 2 or so). That is 247 EVs. In that status quo, we are told today by Gallup and Rasmussen that Romney leads by 2, we are told by IBD that Obama leads by .75, and by RAND that Obama is up 3.75. The average is even. Yet when you take the hard Obama base of 247, and add to it Ohio (19 EVs), where Obama leads, and Nevada (6 EVs), where he has never trailed, Obama gets to 272. This ignores lightly-polled Iowa (6 EVs), where Obama led by 2 per Rasmussen after the Denver debate, New Hampshire’s 4 EVs, Virginia (13 EVs), where Obama narrowly leads in post-debate polling, and Colorado’s 9 EVs sitting in a jump ball. Even ceding North Carolina’s 15 EVs to Romney, and ignoring Florida’s 29 EVs, this “even” status in the trackers is not “even.”
More to the point, Gallup and Rasmussen show Romney up 2, a position I which he should be more likely than not to win. Yet the most recent polling in Ohio, Nevada, Wisconsin, Virginia, and Iowa alone would give Obama 293. Indeed, there is little argument that Romney trails in Wisconsin and Ohio, and has literally never led in 2012 polls of Nevada. The trackers would have to move at least two, and likely three more points (and we have to assume that motion would translate equivalently to motion in OH, WI, and NV) for Romney to appear more likely than not to win these states. In other words, Romney needs whatever Gallup and Rasmussen measure to be +5 for him to be truly “even.” This suggests that those trackers are not put together well.
In trying to understand what is wrong with this picture, consider the polling data heading into the 2010 Senate elections – importantly, from many of the states now studied by the same pollsters – and how in every close election in 2010, the polls overpredicted Republican performance. The average of polls in California showed Barbara Boxer up 5; she won by 10 (a +5 swing). Polling averages showed Harry Reid losing in Nevada by 3 when he won by 5 (a +8 swing). Polling averages showed Michael Bennet losing his Colorado Senate seat by 3; he won by 1 (a +4 swing). Polling averages showed Mark Kirk winning an Illinois Senate seat by 3.3; he won by 1.9 (a +1.4 swing). In these four states with significant Hispanic populations, the Democrat overperformed polls by an average of almost 5 points. Consider that Barack Obama does not even need to overperform the recent average of polls to win Nevada, and that a 1 point overperformance would win him Colorado. Nate Silver approached this a different way yesterday, when trying to explain a survey showing President Obama ahead in Arizona by 2 points as a function of that poll’s use of Spanish-language surveys, and President Obama’s whopping 77-10 margin among Arizona Latinos. He tweaked his model to include Latino voting margins collected by surveys specializing in Latino voters. Thus tweaked, President Obama’s likelihood of winning Nevada leapt from 62 to 77%, and his chances of winning Colorado moved from 43% to 55%.
Putting aside the inference I draw that Latino voters are undersampled in a way that suggests that President Obama will win Nevada and Colorado, the 2010 polls also underestimated Democratic performance in a variety of states with closely contested elections and without large Latino populations. This suggests to me that cellphone v. landline issues and other cultural differences in survey response by age or lifestyle have skewed recent political surveys modestly toward overpredicting Republican advantages. Thus, the Democratic candidates also overperformed final polling averages by 2.5% in Pennsylvania, 3.5% in Washington, 5.6% in West Virginia, and by 2.8% in Wisconsin. Here, the average is 3.5%.
In sum, if Gallup and Rasmussen are right that Romney leads by 2, then it is obvious that Obama has an Electoral College advantage and should be expected to win if he is within 3 or so points. If on the other hand Obama has no Electoral College advantage, then the tracking polls are off – by about 4 points in Gallup and Rasmussen’s cases. This happens to coincide closely with the 2010 margin of overprediction of Republican performance in all eight Senate races that were within 10 points – NV, CA, CO, IL, PA, WA, WV, WI. My strong belief based on 2010 is that the continuing disconnect between state and national polling is a function of presently-unexplained systemic defects in the national polls. This inference is reinforced by the very strong evidence in 2010 – all eight polls having a systematic and consistent bias. This inference is also reinforced by the wide range of states (OH, VA, NV, IA, WI, NH, FL, CO) not showing a consistency with the supposed national trend.
Shorter version: the underconsidered RAND tracker showing Obama up 3.75% nationally is much closer to reality than either Gallup or Rasmussen. Otherwise, the state polling is systematically incorrect, despite the wide disparity among states and pollsters in the filed in those states. Systematic error more logically resides in the two famed national pollsters.
4. Obama’s Powerful Ground Game, Demonstrated By Early Voter Behavior in Ohio, Iowa, and Nationally, Makes It Very Likely He Will Win.
Obama’s ground game is very potent and will likely decide this election. PPP reports that 19% of likely voters in Ohio have voted, and by the extreme margin of 76-24 for Obama. That showcases the difference in muscle between the Obama and Romney organizations. When last I saw, Obama’s campaign had contacted 7% or so more voters than Romney’s at two different points the subject was studied. The strong implication is that Obama’s voter contact is connected to a concrete outcome – getting and delivering early ballots – and that Romney’s is not.
There are other data less striking but which underscore the difference between the two campaigns and their efficacy in promoting early voting. Reuters/Ipsos is reporting that Obama leads Romney 59-31 among those reporting early voting (presuming that the rest are not stating a preference to the pollster, one can infer that close to two-thirds of all early votes are being cast for the President). Other sources are indicating that Democrats are early-voting in greater numbers in Iowa and North Carolina than did so in 2008, while Republicans are early-voting less in North Carolina.
In 2008, President Obama overperformed his poll numbers more in Nevada than in any other state, polling +7 and winning by 12 (an overperformance of +5). Senator Reid in 2010 overperformed his expected performance by 8, turning polls of -3 into a win of +5. Given what we have seen of the Democratic and Obama ground game in Nevada, it is very hard to see how President Obama is suddenly going to perform at or worse than polls there – and Romney has not led in one this year. President Obama is very likely to win Nevada, and that same ground game is why.
Applying the methodology of comparing final poll averages to outcomes, one sees that President Obama also overperformed his polls by 2 points in winning Virginia by 6. The Obama ground game there – where he leads very narrowly in the average of post-debate polls – is likely to make the difference in another jump ball election.
Again, one has to overlay the indications that Democrats are far more organized and getting out the vote in Ohio, Nevada, and Iowa, for example, with an awareness that adding them and safe New Mexico to the Kerry states would give President Obama 281 EVs – and no need even to hold Wisconsin – putting aside the good likelihood he will win jump-ball Colorado and Virginia with a ground game when polls show Romney up 1 and down 1, respectively, in those two states.
Conclusion: 80% Likelihood of an Obama Win Seems About Right.
From my review of the foregoing, I am confident that today, President Obama would win the election, and is winning it by a fair and neutral understanding of the data. Mitt Romney has a clear path to victory, but it is narrow – he must continue to thump a now-more-motivated President in debates that play more to his strengths. (It is harder to thump an engaged incumbent President in a townhall session of expressing empathy for voters while answering their concerns, and harder to thump an incumbent who enjoys more credibility on foreign affairs with only a harder line on Iran and Afghanistan, the Benghazi attack, and a great deal of agreement, than it was to win decisively in Denver.) And even if Mitt Romney wins debates, there are far fewer independent voters available to him now than before, precisely because his status as roughly tied depends on already having them.
Base-revving is not going to win Nevada for Mitt Romney, given the superior performance of the Democratic base in 2008 and 2010, and judging from the PPP survey of Ohio, the auto bailout and Romney’s “Let Detroit Go Bankrupt” response are holding him down there with white independents, as the President leads Romney by 5 in an Ohio sample that reports preferring Republicans in Congress by 2. And to win, Romney needs Ohio, Florida, and North Carolina, and could still lose by losing WI, NV, IA, CO (272 EVs), or WI, NV, VA (270 EVs). Based on all of this, Romney can win, and has done well to stay in the hunt. But far from being “ahead” as his wishful throng of trailing media would have it (Gallup, Rasmussen, RealClearPolitics, FOX), he must still win debates and thread a needle. The far greater likelihood is that he will not.