It's something of a tradition for newspapers to release polls on Sundays late in a presidential campaign. This Sunday, there were polls out in Ohio, Florida and Pennsylvania - but as it's Nebraska day at FiveThirtyEight, let's start with one from the Cornhusker State.
The Nebraska poll, from Wiese Research Associates for the Omaha World-Herald, put Mr. Romney 14 points ahead among likely voters statewide. Nebraska, however, awards three of its five electoral votes by Congressional district; Mr. Obama narrowly won the vote of the Nebraska's Second Congressional District, which is largely coterminous with the city of Omaha, in 2008. The Wiese Research Associates poll showed a tied race in the Second Distric t, suggesting that such an outcome is possible again this year.
The FiveThirtyEight forecast, which accounts for district-level polls in Nebraska and in Maine (the other state which splits its electoral votes in this way), is more skeptical about this possibility. It gives Mr. Obama a 20 percent chance of winning the district (although that's up from about a 10 percent chance before the poll came out).
Part of this is for a technical reason: the district-level results that the poll published were among registered voters, rather than likely voters, and a tie for Mr. Obama among registered voters would normally translate into a narrow deficit for him among likely ones.
The nonpolling factors that the model uses, like fund-raising data, also suggest that Mr. Obama will have a difficult time repeating his performance in Omaha. This year, Mr. Obama has raised just $261,000 in Omaha zip codes, as compared to $638,000 for Mr. Romney. (This contrasts with 2008, whe n Mr. Obama slightly outraised John McCain in Omaha.) Also, as Micah Cohen explained, the district's boundaries were redrawn in a way that makes them ever-so-slightly more favorable to Republicans. And Mr. Obama had overperformed there in 2008, even relative to his strong performance nationally, partly because of the element of surprise.
Still, a credible poll showing a tie this late in the race should obviously not be dismissed completely. So let's proceed to the next question: Are there cases where Omaha's lone electoral vote could matter?
Actually, there are fewer than you might think. In order for the vote to make a difference, the Electoral College would either have to be 268-270 against Mr. Obama without it, meaning that it would bring the race to an exact 269-269 tie, or it would have to be 269-269 already, with the extra vote giving Mr. Obama a 270-268 win.
The cases were the Omaha vote brings Mr. Obama to 269 electoral votes from 268 are probably n ot all that relevant, since a majority of delegations in the incoming House are likely to be controlled by Republicans, meaning that Mr. Romney would win a tied election under the 12th Amendment. So what we're really looking for is cases where the vote pushes Mr. Obama to 270 electoral votes, averting the tie.
The forecast model assigns only a 0.3 percent chance to a 269-269 tie occurring, however. There is no especially sexy reason for this; 269 is just not a number that happens to come up all that often when the model considers the relationships between the vote in the different states and simulates all the plausible Electoral College outcomes.
But one case where a 269-269 tie could occur is if Mr. Obama wins Ohio, Wisconsin and New Hampshire, but loses most of the other swing states. This case, depicted in the map below, is the most likely 269-269 tie scenario, according to the simulations.
Notice, however, that in this instance, Iowa is one of the state s that Mr. Obama loses. The city of Omaha shares a river border with Iowa and has a fair amount of cultural continuity with it, even if they disagree vehemently on college football.
Mr. Obama is several points ahead in the Iowa polls now, whereas he is tied, at best, in Omaha. It's not very likely that Mr. Obama wins Omaha conditional upon losing Iowa, as this outcome would require.
In fact, you have to be pretty creative to come up with 269-269 ties in which Mr. Obama wins Iowa; these instances occurred only twice out of the 25,001 simulations that we ran on Sunday. Both were odd-looking maps. The instance depicted below is the more feasible-looking of the two; it would require Mr. Obama to lose ground among working-class voters (losing Ohio, Wisconsin and Nevada as a result) while holding up well enough in the suburbs to win states with high education levels like Virginia, Colorado and Iowa.
Still, the model would give roughly 1,000-to-1 odds against the Omaha district being the tipping point in a close election. More likely, it would be somewhere that Mr. Obama might win if he were already having a strong night over all, as he did in 2008.
Remember The Maine Electoral Split
In fact, it's the electoral vote split in Maine that is more likely to make a difference, according to the model. Mr. Romney is running well behind in Maine statewide, but he's a bit closer in the state's rural Second Congressional District, which encompasses most of Maine outside of the Portland metropolitan area.
A recent poll of Maine put Mr. Romney seven points behind in Maine's Second Congressional District. That isn't extraordinarily close, and the forecast model gives Mr. Romney only about an 8 percent chance of winning it. However, it also isn't that far from the national average, since Mr. Romney appears to trail by four or five points nationally right now. In fact, Maine's Second District is slightly closer to the electoral ti pping point than Nebraska's, according to our analysis.
The most frequent instance where the one electoral vote in Maine made the difference for Mr. Romney is shown in the map below. This scenario has Mr. Obama losing New Hampshire - which seems sensible if he's also underperforming in Maine - along with Wisconsin, Iowa and Virginia. With all four of Maine's electoral votes, this would be a winning map for Mr. Obama. But with just three of them, it would be a 269-269 tie, likely to be resolved for Mr. Romney in the House.
There was even one simulation in which Mr. Romney won the Maine Congressional district to offset a win for Mr. Obama in the Nebraska one. But let's not get too carried away with this. Mr. Romney trails Mr. Obama in the national race and in most of the swing states and will need to rebound for any of the Electoral College math to matter.
The Rest of Sunday's Polls
The most problematic poll of the day for Mr. Romney was conducted by th e University of Cincinnati for a consortium of Ohio newspapers. It put Barack Obama five points ahead in Ohio, up from three points in a poll they conducted last month.
A survey of Florida, however, from the firm Mason-Dixon, had more mixed news, giving Mr. Obama a one-point lead there, the same margin as in their July poll.
Mr. Obama's performance in post-convention polls of Florida has not been as consistent as in other states like Ohio, Virginia or Wisconsin. As a result, Florida has fallen slightly on our tipping-point list. It now looks like Mr. Obama would have a slightly easier time winning Ohio, winning Virginia, or winning some combination of mid-sized states like Colorado, Iowa, and Nevada.
This is certainly not to suggest that Florida should be ignored, since it's close to inconceivable that Mr. Romney could win the election without it.
But if you're strategizing for Mr. Romney's campaign in Boston, I think you can have a pretty reasonable hope that Florida will end up back into Mr. Romney's column if the overall race moves a couple of points toward him. It's less clear if that's true for Ohio, where Mr. Obama's polling has been quite robust, and to a lesser extent in Virginia, which is harder to get a read on because of the inconsistent polls there.
One state we haven't discussed much lately is Pennsylvania, since Mr. Obama's lead has appeared quite safe there. But a poll there on Sunday, by Susquehanna Polling and Research for the Pittsburgh Tribune-Review, had a different take on the state, putting Mr. Obama just two points ahead of Mr. Romney.
If this result came from another pollster, it might merit headline treatment, since the electoral map would look a heck of a lot different if Pennsylvania is really this close.
However, Susquehanna has consistently shown much more favorable results for Mr. Romney in Pennsylvania than any other polling organization in the state. A poll they published last week also showed a very small lead for Mr. Obama, just one percentage point. And they are the only polling firm at any point in the year to have shown Mr. Obama trailing in Pennsylvania, as they did in a survey in February. The average of recent polls in Pennsylvania shows Mr. Obama about seven points ahead instead. (Susquehanna conducts polling for Republican clients in addition to its work for news organizations.)
The FiveThirtyEight model applies a house effects adjustment to polls that are consistently more Democratic- or Republican-leaning than the consensus. So it treats this poll as showing the equivalent of more like a five-point lead for Mr. Obama, meaning that it did not affect the forecast all that much.
One thought, however: although it's easy to detect these house effects statistically, it's not always so clear as to why they occur, especially when the polling firm is not very transparent about its methods.
Sometimes you can dig into a po ll and detect the reason - but I've found myself less and less inclined to do that as time has gone on. An awful lot of time can be wasted arguing about individual polls, when it's really the consensus of the data that we should be focused upon. The house effects adjustment usually suffices to give us a clear-enough view about which polls are within the consensus and which aren't, without having to comb through the cross-tabs of dozens of individual polls.
But - and here's the thought - shouldn't polling firms, and the news organizations that publish their results, have a responsibility to explain to their readers why they consistently diverge from the consensus view? Any one survey can and occasionally will be an outlier; I sometimes think that polling firms get too much grief over the occasional odd result. But when a firm is consistently out-of-consensus, they ought to be able to offer a plain-English explanation for what they're seeing that everyone else is missin g. Polling firms very rarely do that, however; instead, they very often act as though their polls exist in a vacuum.
In any event, the Pennsylvania poll is not quite as good for Mr. Romney as it looks, and the Nebraska poll is not very important, as fun as it might be to speculate about cases where Omaha could make the difference.
That leaves the relatively favorable poll for Mr. Obama in Ohio and the less-favorable one for him in Florida. The FiveThirtyEight forecast was essentially unchanged from Saturday, with Mr. Obama having a 77.6 percent chance of winning the Electoral College.
No comments:
Post a Comment