“So, Dave, the election is now over. How did you predict the presidential race with such accuracy, even getting Florida correct before Nate Silver and other experts? On the other hand, how did you get some races so wrong, like Joe Saunders’ win in Florida House District 49?”
Well, it is time for me to answer these questions that I have been getting and give you a glimpse into the method to my madness.
To examine this, I need to explain how I look at races. Currently, there are two polarizing methods regarding election predictions. On one side, we have Nate Silver’s “the polling numbers explain everything” method. On the other side, we have the Dick Morris “I am just pulling predictions out of my ass because I want my guy to win” method. Honestly, I don’t agree with either of these methods. As far as the Morris way of doing things, I think the reason why his method is pretty flawed is quite obvious.
As far Nate Silver, I like his model much better, but I feel he does have some gaps. The first gap is that polling information isn’t always based on human emotion, but instead on what the interviewee thinks the interviewer wants to hear. For example, if we were to poll certain regions of north Florida and ask “would you vote for an African-American candidate for President”, chances are the responses will be much different than their actual feelings. There sometimes seems to be an idea of “being right” instead of giving your opinion. Therefore, looking at polls could be flawed.
The second issue that Silver doesn’t put into his method is electoral behavior and results on the micro-political level. For example, he will take a state like New Hampshire and say that even if the state is considered a “toss up”, the history of that state’s Democratic leaning gives more weight to the Democrats winning. In the case of New Hampshire, which is quite small, that might be true. On the other hand, that doesn’t explain larger states, such as Florida, which can have strong Republican ties in one part of the state and strong Democratic ties in another part making the state a pure toss up.
The Silver method also doesn’t account for other time frames. Let’s look at the 1992 election. Polls in states like Kentucky, Pennsylvania, Michigan and New Jersey would probably have given Silver a different result than the actual result if he was around in 1992. But with elections becoming so polarized nowadays, it is easier to see a right-left split in this country. Between the 2008 and 2012 race, only two states switched parties. Between 2000 and 2012, Democrats have a lock on 19 states and DC, while the Republicans have a lock on 22 states. This means that only nine states have voted for both parties at one point over the last 12 years. So, basically, only nine states have to be predicted correctly. Thus, making predictions isn’t as difficult.
But back to looking at “only the polls” like Silver does. Another reason why this is flawed is because there are other factors in determining who is going to win the election. Some of it has to do with “hunches”, while others have to do with numbers. Let me use my projection of Florida as an example of connecting these two dots. One week before Nate Silver projected that Obama was going to win Florida (which was actually on Election Day), I had predicted that the president would win Florida. What brought me to this conclusion so early? First, I looked at the same information that Nate Silver had and saw that the polling, especially the Public Policy Polling poll, was trending in Obama’s direction. The momentum was purely on the president’s side. I’m sure Silver took this into account as well.
Now comes the part where we differ a little. As someone that knows quite a bit about Florida elections, I was under the impression (or had a hunch, if you like) that Romney would not be able to win Florida without winning Hillsborough County. I called this his “Ohio in Florida”. Therefore, when it came to this idea, it had nothing to do with numbers or polling, but just purely a thought. To try to back up this idea, I looked at the early voting numbers in Hillsborough County and saw that Democrats were not only outperforming the Republicans, but they were outperforming their own registration percentages as well. Basically, the early voting was more Democratic than it was in 2008. Adding two and two together, I projected that Obama would win because early voting in Hillsborough County favored the Democrats heavily. And since I stated that Romney needed to win Hillsborough to win Florida, the early voting numbers confirmed to me that Obama would win Florida. This method added with the fact that the more accurate polling firm was now showing the race as a dead heat, and even showing Obama slightly ahead, would give Obama the win at the end of the day.
These hunches and examination of the early voting numbers is not what Nate Silver does. What if the PPP polls done before the election were just one or two points in favor of Romney? Would Silver still have predicted the state for Obama? The fact that he waited until the last minute means he looked at momentum as a factor. But if that momentum wasn’t measured correctly, would he have gotten Florida right? The PPP poll was just a confirmation of my prediction, not the main factor of it. This is where Nate Silver and I differ on our methods.
When putting “hunches” into political predictions, one must leave their own mind and body and put themselves into the mind of the average voter. Prior to Election Day, we were accused by some of being “biased” in our picks. When I make my picks, I take every aspect of my personal bias out of it. Instead, the hunches are based on what I feel the electorate would do, and not by what I would do. For example, let’s look at two congressional races I predicted correctly, the Florida 18th and the Minnesota 6th. In the FL 18th, I predicted Patrick Murphy would win. Why? Well, in addition to looking at the numbers as far as registration and past voting history, I looked at the fact that Allen West was only a one term congressman and won in a highly Republican year. In a year in which Democratic turnout would be substantially more, I looked at this as an advantage for Murphy and picked him over West.
In the case of Michele Bachmann it was a little different. Unlike West, who only won once in 2010, Bachmann has constantly won her district, even by small margins. There had never been a rejection of her by her constituents. These election wins by Bachmann happened in strong Republican years, like 2010, and in strong Democratic years, like 2008 and 2006. Therefore, if she had survived these tests, she would survive 2012 as well. Bachmann winning was a clear decision, as was West losing to Murphy.
Tomorrow I will explain my examination of the votes in the Florida House and Senate.
I dont think its fair to classify Nate Silvers way of prediction as solely based on numbers. In his book he makes it very clear that qualitative as well as quantitative data must be taken in to get a full projection. He spends most of his book arguing that a purely numbers driven way of analyzing data is foolish, and that without the context of a smart person who knows what they’re looking at it means nothing. I do agree though that silvers method may be too simplistic for Florida, which as we know can be a pain in the ass to get your head around.