by Trevor Antley and Calvin Roberts.
[Update: The election has been called, and the winner of Ohio was President Barack Obama. Our forecast, as we noted was very possible, turned out to be incorrect. While the problem with the model can't be verified until after full election results come in, it is most likely that the margins assumed for election day voters in Ohio were much closer than we assumed (assumptions which were based on an aggregate of three of the most recent Ohio polls). Higher-than-expected turnout among African-Americans and youth voters evidenced in exit polls seem to have boosted Obama's election day margins.]
[Update 2: Our model was correct. The variables were that turnout was lower-than-expected (below 2008 turnout), and that the polls were slightly wrong regarding Romney's margin over Obama with election day voters (it was slightly less than the number we used for our projection). Had we aggregated all of the polls and not the final three, as other statisticians suggested, the model would have shown Obama's victory. Hats off to Nate Silver.]
Forecast: Mitt Romney Will Likely Win Ohio
Abstract: Actual reported early voting data requires that early voting will represent no more than 32% of total vote in Ohio, while virtually every poll was weighted for early voting to occupy ~35-40% of total votes cast. The smaller-than-expected number of early votes means one of two things: 1) 2012 will see historically low voter turnout in Ohio; or 2) Mitt Romney has a much better chance at Ohio than polls assumed.
Late Monday night the Ohio Secretary of State released the “final” early voting results from Ohio’s counties. The results got the attention — and slight consternation — of the New York Times’ Nate Silver. Dave Wasserman kindly put the data into a spreadsheet here, which tabulates early voting results by county and compares that data to early voting results from 2008. Wasserman’s spreadsheet also notes Kerry’s 2004 margins and Obama’s 2008 margins, allowing one to effectively deduce the partisan-leanings of each county.
In a discussion on Twitter, Silver and Wasserman focused largely on the surprise changes in turnout in many of Ohio’s counties. While total early voting in Ohio only increased by 2.44% from 2008, early voting in counties that voted heavily for Kerry/Obama declined 4.1% while counties that voted heavily Bush/McCain increased their early voting by a shocking 14.39%. Wasserman, while still predicting an Obama victory, suggested that trend meant a tighter race in Ohio than expected and suggested it might undercut Nate Silver’s famous forecast. Nate Silver’s response: “I’ll stick with the 538 forecast in OH. I disagree that the early voting data there provides much reason to doubt the polls.”
Seemingly overlooked by Silver, however, during the discussion of county-by-county results was the simple number of total reported early votes: a meager 1,787,346. As stated above, this number shows a 2.44% increase in early voting from 2008 — but the number is still surprisingly low. Virtually every Ohio poll this cycle was weighted on the basis that early voting would occupy a massive chunk of the total Ohio vote. Rasmussen’s final poll ceded 40% of the total vote to early voters (EVs). PPP gave EV’s a more reasonable 35%. The Columbus Dispatch calculated early voting to take up an astounding 47% of the total Ohio vote. Almost every other Ohio poll seems to have weighted early voting between 35% and 45% of the total vote.
The reported early voting numbers, however, show that virtually every single Ohio poll overestimated the amount of early votes cast. If early voting is calculated at 1,787,346, in order for total voter turnout to rival 2004 numbers, early voting cannot occupy more than 32% of the total votes cast — and even in that scenario, that high of a percentage means that total voter turnout will be lower than it was in 2008. In order for turnout to match 2008 levels, early voting can only account for 31% of total votes cast.
The next important piece of data is what the polls consistently report: Obama leads by huge margins among early voters but trails Romney among those who say they will vote on election day. This inverse in voting segments is why the proportion of early votes in the total votes — and that virtually every poll overestimated this proportion — is so tantamount. In most polls (which usually only have Obama leading by a small margin, although some give him a more comfortable ~+5%), lowering the percentage of early votes in the polling sample means lowering Obama’s lead drastically. And when Obama’s lead is only one or two percentage points, that can mean handing the election to Mitt Romney.
Our forecast is based largely on the reported margins between Romney and Obama among early voters and election day voters as reported by the Columbus Dispatch, Rasmussen, and other polls (all polling data considered is represented in the graphic below). The Columbus Dispatch gives Obama +15% among early voters; Rasmussen gives him a much wider 23%. Other polls for Ohio EVs: CNN/Opinion Research, Obama +28; Gravis Marketing, Obama +13; PPP, Obama +21. For our forecast we assumed a more conservative Obama +18 among EVs, averaging Rasmussen and the Columbus Dispatch.
In 2008 Obama won 58% of early voting against John McCain, who had virtually no get-out-the-vote infrastructure in Ohio; our model, giving Obama a 18% lead, again assumes he will win that 58% of early voters despite the fact that Mitt Romney is putting forth a much more competitive get-out-the-vote campaign and disregarding the GOP-leaning trend in early voting results of individual Ohio counties. When one considers the results from individual Ohio counties this cycle, Obama’s actual margin among EVs may actually be much lower (although without specific partisan data, it’s also possible that Obama’s margins have actually increased — although this seems extraordinarily more unlikely). But because this is impossible to determine without actual breakdowns of the early vote, which are not yet available, those implications are not included in this model.
In determining the margin among election day voters, the same polls were considered. For election day voters, Rasmussen has Romney +15; Columbus Dispatch, Romney +11; and CNN/Opinion Research, Romney +13. PPP and Gravis Marketing both had Romney’s election day margins at a much smaller +3. For our forecast, we assume Romney’s election day voter margins at 13%, an average of the first three polls. The consistency and disparity between the first three and the latter two polls made it difficult to average them since margins of error do not explain such a clear discrepancy between the two groups.
In this scenario — which seems to be supported by the majority of polls and early voting trends (but is notably not supported by all polls, as seen in the previous paragraph) — Romney should win Ohio. Based on these assumptions — which in turn are based on a combination of polling data and the state’s actual reported early vote — if early voting accounts for 32% of the vote (a very conservative number which would place total voter turnout slightly below that of 2004), Romney wins by a whopping 50.9% to Obama’s 47.8%. The higher voter turnout is — and therefore the lower the percentage of early votes in total votes — the higher Romney’s margin becomes.
In this scenario, even if we assume our model’s margins between Obama and Romney among early voters and election day voters are somehow skewed in Romney’s favor, Romney still has padding that those margins could be reduced and he still wins. If early voting is only 31% of the total vote — putting Ohio’s total vote at just above 2008 levels — Romney has incredibly more wiggle room.
The lower-than-anticipated turnout among early voters suggests the Obama campaign’s lead in Ohio was largely hot air. And this does not even seriously consider the county-by-county early voting results, which appear to be even more damaging to Obama.
Reasons Why This Projection May Turn Out to be Wrong
- In the case that the final early voting numbers reported by the Ohio secretary of state are incorrect and the final early voting results will include statistically significant additions, obviously this projection will have no meaning.
- As seen above, some of the polling data used in the projection (such as Romney’s margin among election day voters) is supported by several independent polling organizations but not by some others. If it turns out that the fewer polls’ results were right, then obviously our entire model is skewed too heavily towards Romney.
- Some have raised the possibility that effects from Hurricane Sandy stifled early voting in the final days and these early voters will simply vote for Obama on election day, increasing his election day margins beyond what polls indicated. In this scenario the polls are essentially still correct; Obama’s early voting margin was simply reallocated to his election day margin. There is no solid data to show that this is the case, but it is certainly possible.
- There is always the chance that the government and electorate will decide simply to defer to Nate Silver’s forecast and forget this whole voting nonsense. Since our forecast is based largely on actual votes, not subjectively weighted aggregates of polls, this would make our projection essentially meaningless.