Last year, we ran a project during the 2012 presidential election that successfully predicted that Barack Obama would win his bid for election over Mitt Romney. This year, we decided to see if we could repeat the win by predicting voting behavior in the elections for governor in Virginia & New Jersey.
In 2012, we broke down all of our models for you piece by piece so you could see what we were up to. This, year, we’re getting even more transparent.
Data, just like photos, can be manipulated to look better. But just like the trend of publishing the raw photos from photo shoots, we wanted to give you a look under the hood at our raw data. No gimmicks. No weighting. No controls. You see what we see.
We have made only 2 minor adjustments to our sample:
1. We have eliminated people who are not actually registered to vote.
Why? The goal is to estimate opinions of voters, not opinions of non-voters.
2. We have eliminated people who are not actually planning on voting.
Why? Again, the goal is to estimate opinions of voters, not opinions of non-voters.
The only mathematical manipulation we’ve used is the computation of a 5-day trailing sum.
Wait, what is a 5-day trailing sum?
Data is volatile and we’re drawing our sample from people who take SurveyMonkey surveys. You take a survey on our site, and when you’re done you see a link to our political poll. These people can vary greatly from day to day. To make sure an odd day doesn’t swing our prediction, we use a running 5-day trailing sum. That means that any given data point will reflect a changing snapshot of attitudes that is spread out over the past 5 days.
So, remember, this is our raw data you’re about to see. We’re going to display it using data taken from RealClearPolitics (RCP) polls. RCP keeps a running average of all of the available political polls out there (the majority of which are heavily weighted phone polls).
Both our data and the data from RCP indicate that Terry McAuliffe and Chris Christie will be victorious tomorrow. However, there are discrepancies on any given day of polling between our estimates and those of RCP, sometimes by as much as 5%.
Why would the phone polls be saying something different?
Well, we might not be talking to the same people. In both Virginia and New Jersey, roughly 60% of all of our respondents say that they have not taken a single phone poll on any subject in the past 6 months. Think about that. That means out of every 10 people we talk to, six of them are not being represented in phone polls. That means that phone polling just doesn’t have the reach it once was.
Why is it losing its reach?
Well, think about the last time you answered the phone during dinner. The fact is, people just don’t talk to telephone pollsters anymore. Also, think about the people you know under the age of 35. How many of them even have a landline? Our country is moving to a cell phone-only culture of people who screen their calls–and the era of representative data through phone polling is coming to an end.
Okay, by now we’ve convinced you that telephone polling is so 1990’s, but what gives us the idea that we’re doing it the right way? As we discussed above, we seem to be reaching a broader spectrum of voters than phone polls. And, as part of our experiment, we added a question to quality check our sample. We asked our respondents what candidate they voted for in the 2012 presidential election. Our results are within a 2% margin of the ACTUAL votes in those states. Think about that. Most polls give you a 3% margin of error. We’re at 2%. This means that our sample is indeed a great estimate of the general population we’re interested in–voters who show up to the polls.
As the data above shows, we’re betting on Christie and McAuliffe to win in the polls tomorrow–but the outcome is really up to the voters in Virginia and New Jersey. So, get out to the polls and vote!
And then, tune in on Wednesday to see how we did…