Beyond the Numbers: What I Said and What Happened

A good, credible commentator continuously examines what he or she felt confident predicting in the past for both accuracy and learning.  It was the lack of this behavior on the part of right-wing commentators that I was screaming about yesterday.  Of course, introspection is made a little easier for me this time around, as my projections came through very solidly.  But, hey, what's wrong with a little victory lap?

Let's revisit the principle arguments made for a Romney victory and the counterpoints I made to those last Friday:

1) President Obama is still under 50% in virtually every national poll.  Undecideds will break late for the challenger and give Romney the narrow victory.

I wrote:

Recent history suggest no evidence of this rule of thumb.  Undecideds in 1980 surely did break for Ronald Reagan over Jimmy Carter late.  In 1984, they broke for Reagan again - this time as the incumbent.  In 1992, undecideds broke evenly.  In 1996, they broke for the challenger.  In 2004, they broke evenly.  There doesn't seem to be a pattern here to support the "rule of thumb" that an incumbent under 50% is in trouble - George W. Bush was under 50% in the polling and got 51% of the vote on election day. 

What happened:

Exit polling indicates that of those who decided in the final days of the election, 50% voted for Barack Obama, 44% for Mitt Romney.  This explains, to a certain extent, why my results had a slight Pro-Romney bias - I had assumed a 50/50 split (the difference is worth about a 0.5% shift in the popular vote.)  Certianly the notion that undecideds break late for the challenger is debunked.

(2) No incumbent President has ever been re-elected to a second term winning less states than he won the first time around and it is impossible to see a path to President Obama winning more states than in 1988.  It's win big or go home for incumbents and Obama cannot win big.

I wrote:

True, but irrelevant.  No one had ever won 49 states...until 1984 when Ronald Reagan did.  Candidates always win their home state - heck, even George McGovern and Walter Mondale did - until Al Gore lost Tennessee and the election with it.  The winner of Missouri always wins the election - until 2008 when Barack Obama won without it.  My point is that you can point to lots of things that are "always" true - until they aren't. 

What happened:

President Obama won re-election by a narrower margin than 4 years ago (approximately 2.5% in the popular vote versus 7.2% in 2008.)  Guess we can cross this one off the "no incumbent President has ever" list.

(3) The polls systemically overestimate Democratic turnout and the actual results will therefore differ from the polls by several percentage points.

I wrote (in a previous post):

Non-partisan pollsters only stock-in-trade, their only incentive is to get things right.  If you consistently get your polls wrong, you are out of the business.  Polls have been wrong before, but almost never because the pollster wanted to get it wrong.  The polls were very accurate in 2008.  Ditto in 2004.  In 2000, George W. Bush got less popular vote than the polls implied late, but that was principally because the release of his DWI conviction appeared to cause a late slide in his numbers that pollsters weren't able to capture in their final polls because it happened too late.

So forget the notion of a vast conspiracy.  The only reasonable way to believe the polls are systemically biased is if they are ACCIDENTALLY biased, that is, the majority of pollsters make an honest mistake in the turnout dynamics of the election.  Now clearly all the pollsters are reading and analyzing the claims on the right - and most are sticking by their guns.

What happened:

To the extent that there was any polling bias, it was Pro-Romney.  Obama outperformed 7 of the 10 polls in my final projection.  In the key swing states, Obama outperformed the projection in Florida, Virginia, Colorado, Pennsylvania, Nevada, Iowa, Michigan and Wisconsin.  Only in Ohio did the President underperform the projection and there only by 0.2%.

Clearly the polls did have a systemic bias - FOR Romney.  Pollsters generally estimated a turnout model between the 2004 and 2008 models when, in fact, turnout of demographics that favored Obama met or exceeded 2008 levels in 2012.

(4) The national polls show a tighter race than the state polls and the national polls are generally conducted by better-established, more reliable polling firms.  It is therefore reasonable to believe that swing states are actually in better shape for Romney than the state-level polling data would indicate.

I wrote:

Generally, the evidence doesn't support this theory.  On average, state-wide polls have been at least as accurate and often more so than national polls on election day...see 2000 for a great example of this.  Secondly, while there are some smaller firms polling in swing states, there are also a lot of large ones - Rasmussen, CNN/OR and Survey USA are all poling Ohio and their results are actually well in line with other polls from smaller firms.

What happened:

The polls largely converged by election day and both showed a slight pro-Romney bias, rendering the argument largely irrelevant.

If you like this site, tell your friends

The Big 2012 Map - No Change - Obama Still Predicted to Win, A Look at Early Voting and What It Tells Us About Systemic Poll Bias and Swing State Outcomes

First Election Day Polls Open In: 2 Days, 15 Hours

Projected Popular Vote Total: Obama +0.1% (down 0.3% from yesterday)

Projected Electoral Vote Total: Obama 290, Romney 248 (unchanged from yesterday)

Current Betting Odds: Obama 67%, Romney 33% (unchanged from yesterday)

Current Popular Vote Betting Odds: Obama 56%, Romney 39%, within 0.5% - 5%

Please note that for the listing of national polls, those highlighted in green were released either today or yesterday and include polling up through 1-2 days ago.  The other polls are older, but less than a week old.  I have (at least temporarily) dropped the Battleground poll from my list as it is older than a week.  We will see how the other firms that shut down during Sandy do at getting back up and running - it appears Gallup will have a final poll publish on Monday but will likely not publish on Sunday.

We see no states swing on the map today as the theme of stability following the third debate continues.  The candidates are campaigning vigorously, focusing the majority of their time on all-important Ohio, although Romney has fanned out to the marginal swing-states of Pennsylvania and Wisconsin as well (fortunately, they are close by) as well as Colorado this weekend.

Meanwhile, Paul Ryan hits up Florida, Nevada and Minnesota.

Barack Obama and Joe Biden, in addition to a strong focus on Ohio will be in Florida and Colorado this weekend.

As the polling doesn't really show much of a different picture than yesterday, I thought I'd focus my efforts on taking a look at available early voting data to see how it might inform us as to how the actual election is going.

Early Voting Data Is a Mixed Bag, But Largely Supports Our Story

What had been a dull murmur in GOP circles about polling bias has grown to a full-blown scream.  The GOP faithful believe strongly that Mitt Romney is going to win, although from the looks of Intrade, few are placing bets on it (but then, gambling is a fairly un-GOP-like activity.)

Mitt Romney will win 315 electoral votes!  Or so proclaims conservative columnist Michael Barone.

A very long night on Tuesday for Democrats says the National Review.

Good bye, President Obama scoffs the American Spectator.

And so on.

Meanwhile, the left, has been quietly saying that the President looks to be in good shape.  

Of course all of this sparks an obvious view of bias on both sides.  If any of these people are able to objectively analyze data, then how can it be that almost all of the Republicans believe that the data clearly point to a decisive Romney victory and vice versa?  Well, I guess you see what you want to see when you are emotionally involved in a Presidential race.

I set up my methodology for analyzing the data long before the election.  If the polling were exactly reversed, I would be showing exactly the reverse numbers.  The objective data from the polls all point to a victory for President Barack Obama.

The only way that this isn't true is if the data aren't truly objective.  Hence the full-blown scream from the right.  You cannot look at the polling data on face and say "yep, it looks like Mitt Romney is going to win" with a straight face.  The only way to look at the data and believe that Romney will win is to believe that the pollster's conclusions are wrong.

Some have quietly questioned the general accuracy of polling because of the low response rate pollsters get these days.  But when there is a consistent trend across many polls, it is hard to argue that this alone would cause polls to be wrong.

This is why the debate over likely voter models is so crucial and why it has become such a loud GOP talking point.  I've spewed a lot of digital ink on this subject and my conclusion has been that there is no reason to believe that there is any inherent bias in the polling.

Non-partisan pollsters only stock-in-trade, their only incentive is to get things right.  If you consistently get your polls wrong, you are out of the business.  Polls have been wrong before, but almost never because the pollster wanted to get it wrong.  The polls were very accurate in 2008.  Ditto in 2004.  In 2000, George W. Bush got less popular vote than the polls implied late, but that was principally because the release of his DWI conviction appeared to cause a late slide in his numbers that pollsters weren't able to capture in their final polls because it happened too late.

So forget the notion of a vast conspiracy.  The only reasonable way to believe the polls are systemically biased is if they are ACCIDENTALLY biased, that is, the majority of pollsters make an honest mistake in the turnout dynamics of the election.  Now clearly all the pollsters are reading and analyzing the claims on the right - and most are sticking by their guns.

So other than hot air, the only thing that could deter me from believing in the power of poll aggregation is if actual facts on the ground contradict the assumptions in the polls.

The one very reliable source of data that we have on this subject is the release of early voting data from the swing states that provide the data and have party registration.  I'm not immune to new data - if the early voting suggest an electorate vastly different from what the polls have been modeled on, they we should adjust for the possibility that those models are flawed.

So, let's dig in on the key states:

Colrado

Early Voting in 2008: Democrats +1.8%

Early Voting in 2012 through 11/2: Democrats +0.1%

Democratic Margin of Victory in 2008: +8.9%

Republicans have gained 1.7% in early voting in Ohio in 2008.  The problem is that they have an 8.9% gap to close.  Now clearly, they will gain among Independents versus 2008 - Obama won independents by about 5% in 2008 according to exit polls and recent polls have had him trailing Romney with indies anywhere from 2% to 10%, depending on the individual poll and how it assigned party affiliation.  But even in the worst case, a 15 point swing among independents, this is only worth a 4.2% swing in the general election margin.

Adding the 4.2% swing in indies to the 1.7% swing in Democratic advantage, you get a net 5.9% swing for Romney, or a 3.0% margin of victory for Obama on election day, consistent with recent polling.  If you assume the lower end of the independent swing (a 7% swing to Romney among indies), you get only a net 3.6% pick-up in total for Romney, or a 5.3% Obama win.

Conclusion: early voting in Colorado gives me no reason to believe that the recent polls there are wrong and, in fact, supports the case for a low-to-mid single digit victory for the President, slightly higher than his current standing in our aggregation.

Iowa

Early Voting in 2008: Democrats +18.0%

Early Voting in 2012: Democrats +11.8%

Democratic Margin of Victory in 2012: +9.5%

Assuming similar logic to Ohio - on the high end, a 6.2% pick-up for Republicans in voting demos, combined with a 3.7%, you would get a Romney win by 0.4%.  Assuming the low-end of the independent swing, you woud get a total swing of 7.9% or a 1.6% Obama victory.

Conclusion: Early voting in Iowa would imply that Obama is probably sightly ahead, but less decisively so than in Ohio.  The early voting statistics would seem to imply a slightly closer race than the current aggregated polling does.

Nevada

Early Voting in 2008: Democrats +6.4%

Early Voting in 2012: Democrats +6.4%

Democratic Margin of Victory in 2012: +12.2%

Romney hasn't gained at all in early voting.  Assuming just the independent swing would give him a pick-up of between 1.3% and 2.8%, not enough to put him anywhere close to victory in the state.

Conclusion: Early voting in Nevada would imply a less close race than the polling.  Obama is comfortably ahead using these data points.  Bear in mind that Obama significantly outperformed polling in 2008 in Nevada as well.

North Carolina

Early Voting in 2008: Democrats +28.7%

Early Voting in 2012: Democrats +16.2%

Democratic Margin of Victory in 2012: +0.3%

This would represent a huge swing, 12.5% from party shifts and 1.4% to 3.0% in independent shift, leading to a double-digit victory for Romney.

Conclusion: Early voting would imply that Mitt Romney is further ahead in North Carolina than the poll aggregation would imply.

Florida

Early Voting in 2008: Democrats +8.3%

Early Voting in 2012 through 11/2: Democrats +2.2%

Democratic Margin of Victory in 2008: +2.8%

This would represent a 6.1% swing to Republicans from early votes plus a 1.1% to 2.6% swing swing among independents, implying a Romney margin of victory of 4.4% to 5.9%, somewhat greater than our current projected margin.

Conclusion: Early voting validates a Florida lead for Romney and may indicate a wider margin than the polling currently indicates.

Note: Ohio, Virginia and Wisconsin do not have party registration systems and therefore do not release party identification information for early voters.  Pennsylvania, Michigan and New Hampshire do not have early voting (they only have absentee voting for valid, verifiable reasons.)

In total, in early voting, Romney appears to be outperforming the polls in Florida and North Carolina, very slightly outperforming in iowa and underperforming in Colorado and Nevada.

Of course, in all these cases, this is just a measure of the early voters and gives us no insight into how voters will show on election day.  In fact, in every case, Democratic turnout is on pace to exceed 2008 levels for early voting - it's just that Republican turnout is exceeding it by more.  This could represent heightened Republican enthusiasm or it could simply mean that early voting is growing more popular in general and Republicans are just later to the party in joining that trend.

There is some room for debate in a few states, but I see nothing in the early voting data to indicate a race dramatically different from the one that I have been projecting.

One clear thing that this exercises demonstrates is that for all the discussion about winning independents, getting turnout among your base is more critical to winning an election in a closely divided country.  Karl Rove knew this in 2004 when he successfully engineered very high GOP turnout amidst lackluster Democratic turnout for John Kerry.  Barack Obama's decisive victory in 2008 was built as much on unprecedented Democratic turnout as it was winning independents.

Both campaigns need to get their team to the polls, whether early or on election day.  And you should visit the polls too to make your voice heard.  But it still looks like Obama's election to lose.

Just a note on the publication schedule for the remainder of the election season (which isn't much):

I will be publishing a daily update on Sunday that will include the latest Presidential polling data an assessment of Obama's first term relative to the promises that he made.

I will then be publishing my final projections on Monday night, including my final popular vote projection, my final electoral vote projection and my final projection in the Senate and the House and a guide to watching Election 2012.

On Tuesday, I will be live-blogging, starting approximately when the polls close in the first states and will seek to keep you updated on how our projections are faring.

I hope you tune in and if you like this site, tell your friends.