UPDATE–11/6, 3:30AM PST
As an addition to the data provided below, results from our presidential poll tracker from Monday, 11/5 indicates that Virginia has gone blue in all three of our models. Also, our raw model now matches our model that is weighted on “how” we sampled. Our “who” model is still bluer than both the others.
Updated expected electoral totals and popular votes are below along with the remainder of results.
At some point, there isn’t any more data to collect before an election—it just needs to happen.
Estimating election results is a complicated business—as you can probably tell if you’ve watched the pundits squirm and speculate over the past few weeks. They throw around terms like “margin of error” and “dynamic weighting” and disagree with each other vehemently. But at the end of the day, there is only one right answer.
Unfortunately, there’s no simple, clear-cut, agreed upon formula to get to that one right answer. And we’re the new kids on the block at this polling game.
First, a few disclaimers about what might bias an online poll of election results…
- VIRTUAL REALITY: People might be giving different answers on an online survey than they do on the phone, or, more importantly than they do in the polls! This might be because they aren’t taking the online survey seriously, or simply because it makes people respond differently than they would otherwise.
- SEEING THE FUTURE: Our survey asks people to predict their future attitudes (Who do you want to vote for?) and behaviors (Will you actually vote?), things that psychological research has shown people to be notoriously bad at.
- SAMPLING THE FUTURE: Not everyone who answered the survey will actually show up to vote. And, despite the fact that over 1.2 million people have filled it out survey so far, not everyone who shows up to vote will have taken our survey. We are relying on an internet sample here which is fairly new ground, and may contain systematic biases about who it is reaching.
So, newbies that we are, we have hedged our bets a little bit, and created three different models to estimate what will happen on Election Day.
Each model looks at the world a little differently, but all of them are fairly similar to each other and to the phone polls. In fact, 48 of the 51 states are similar across our three models. And only 4 states ever differ from Real Clear Politics (RCP) or Nate Silver’s (FiveThirtyEight blog) predictions. All 6 of our tossup states are included in the set of 11 tossup states identified by RCP. Moreover, our popular vote percentage totals are well within the 5% margin of error of both of these “polls of polls.”
But even though we’ve got three horses in the race here—we do have to pick a winner or it’s not a fair fight.
Because our polls are much more sensitive to topical events, like Chris Christie praising Obama on Fox News or Ann Romney talking about women’s issues on The View, we believe that we are able to pick up on electoral changes before the telephone polls are.
Why? Well, because we get our results immediately–much faster than telephone polls that have to collect data and enter it before they can analyze anything. For example, up until November 1st, Iowa, Virginia, North Carolina, and Nevada all looked strikingly different in different models. A week ago, our predictions, and potentially our choice of model might have looked very different. Ultimately we want to go with the model that is most sensitive to the “pulse” of the nation. The model that is ahead of the curve, not behind it.
So, (drumroll please) with all of these disclaimers and goals in mind, we have picked the model that we believe exhibits sensitivity to media events, conformity to conventional pollster wisdom, and existing theories of internet usage.
We pick…Model #3: OBAMA, 318; Romney, 220
How does our estimation differ from everyone else’s predictions?
VERSUS RCP: VA will go to Romney, while North Carolina will go to Obama.
VERSUS FiveThirtyEight BLOG: Same as above, and also Florida will go to Romney.
We hope you’ll visit us here tomorrow as we launch one more grand experiment—exit polling—SurveyMonkey style, of course.
But, most importantly…GO VOTE
Still craving more SurveyMonkey election coverage? Stay tuned tomorrow for a special focus on swing state trends over the past month…and later on the day check back to see how our exit polls are going, and how good our guess was!