top of page


A general explanation of how we make our projections

Data Used

The Beech270 model uses two main sources of data to make predictions about the 2016 General Election. These sources are:

  • State polls taken for the 2016 General Election

  • State polls/results in General Elections from 1916-2012

And so the Beech270 model is, quite simply, a model that makes projections based on a combination of current and past data.

How we use Current Polls

We collect state polls whenever they are published, and evaluate these polls on three main criteria: historical accuracy, sample size, and recency. Based off of these criteria, we assign a rating for every poll, and this rating determines how heavily we weigh the poll in our model. A few things to consider:

  • Polls from firms that have historically been more accurate receive higher weight. Historical accuracy is the most important of the three criteria, by far.

  • Polls with larger sample sizes tend to have higher scores.

  • Scores are adjusted over time for recency. This means a poll's score drops over time, as it becomes less relevant to the current state of the election.

  • While we look at a lot of polls, we do not use all these polls in the model. Polls that have historically been very inaccurate, for example, are dismissed, instead of just having less weight than stronger polls. Quality > quantity.

How we use Past Elections

The component of this model that makes our projections so unique is our emphasis on past data. The model incorporates the results of every state in General Elections in every contest since 1916. We follow a similar logic to the recency criterion of how we evaluate current polls, in that we place more emphasis on recent elections. Still, this model uses data from 100 years ago.


Besides the two main components of this model, there is not much else. It really is that simple. Here are some additional notes about how simple this model is:

  • We do not use national polls. Our calculations for national popular vote are based solely on state polls. There are zero national polls used in this model.

  • We do not explicitly factor in demographics. We believe demographics are accounted for implicitly in the polls.

  • We do not look at the state of the economy. Again, our philosophy is that these factors are implied in the polls.


If you have any questions about the model, feel free to shoot an email to

bottom of page