Monday, July 11, 2016

2016 Election

Let’s go!
Four years ago I launched this blog as a place to publish my 2012 presidential election model. It went well. I nailed all 50 states in the presidential election and several of the key races around Washington State.

Two months ago, I dusted off the model. Between feeling horrified by my 2012 Excel workbook design skills and impressed by my 2012 math skills, I’ve cleaned up the file and improved the methodology. I’m ready for 2016, and I'm excited election math is back! 

For now, I have the brief write-up below, the election map below (now off to the left), and the state data off to the left.

Why you should follow my model
There are many people publishing election models, so why should you read mine? A few reasons. My math is better, my math is more transparent, and most of all, it’s likely you know me personally.

How the model works
At a basic level, the model is polls plus math.

At a slightly more complex level, there are a few steps to “polls plus math.” I will be writing at least one detailed post on each step:
  • It aggregates polls for each state into an average, taking each of the following poll attributes into account
    • Sample Size - larger samples are weighted more heavily
    • Age - Newer polls are weighted more heavily
    • Pollster - each poll is adjusted to account for the 538-measured bias
    • Likely vs. Registered voters - Likely voter polls are weighted more and more heavily the closer we get to election day
  • It aggregates national polls using the same methodology as the state polls
  • It combine state and national polling to estimate a lead and a probability of winning in each state (the technique for doing this is new and improved in 2016)
  • It simulates the election (over and over), captures the results of each siimulation, and retursn lots and lots of statistical analysis
    • For example, Clinton won 80.7% of the time in today's simulation (see map)
    • This process allows for correlation in state outcomes, they are't treated independently
Here's what's next
  1. How national polling is combined with state polling
  2. How national variance is captured
  3. More/new data, analysis, graphs, writing


  1. Excited about the new model.

    I'd like to see a couple stats you could probably get pretty easily off your simulations - expected electoral votes, expected electoral votes in simulations where Clinton wins and expected electoral votes in simulations where Trump wins. It seems like the election is going to be somewhere in the range of close Trump win to Hillary crushing, and these numbers would speak to that sort of thing.

    1. Sure thing. Maybe I'll add that as an output / graphic but for now it's

      Expected Electoral Votes

      Clinton 335.0 / Trump 203.0

      If Clinton Wins
      Clinton 367.8 / Trump 170.2

      If Trump Wins
      Clinton 218.8 / 319.2

  2. Interesting model you are running here. I was curious as to what you meant by "It simulates the election (over and over), captures the results of each siimulation, and retursn lots and lots of statistical analysis". What exactly does an election simulation consist of for you ?

    I run a blog over at . Essentially I am making my calls on discretionary basis by aggregating the aggregators (538, Real Clear Politics, Predictwise, etc) and looking at historical results. More or less meshes up with most of your trends. However there are some significant differences. I have states like MT and IN for example at an almost certainty for Trump while you have them at a toss up.