Going for Accuracy: Predicting Voter Turnout (With easier to read colors!)

After examining 50 Illinois House races from the 2014 General Election, I noticed a new pattern.

My algorithm produces a High Turnout Projection (the Green line) and a Low Turnout Projection (The Gray line).

In my previous post, I contend that during primary cycles, the Low Turnout Projection closely emulates the actual voter turnout.

After adding an additional operator to my algorithm, I found something that looks pretty special.

An average between the High and Low projections produces a number that more closely represents the actual turnout.

This is visible as the blue line (prediction) and yellow line (actual) meet at several points.

For example:

In the Illinois 31st House, 26,394 voters cast a ballot.  The algorithm predicts 26,848 ballots– a difference of only 1.7 percent.

In the Illinois 45th House, 32,001 voters cast a ballot.  The algorithm predicts 31,805 ballots.  This is a difference of 0.6 percent, barely 1 half of a percent off.

In the Illinois 5th House, 27,155 voters cast a ballot.  The algorithm predicts 27,093 ballots.  That is a difference of 0.2 percent.

I’ll keep working.  Let me know what you think!

PS> All data used in this post is public record.  Using data sets from proprietary voting files delivers even greater accuracy.  In my previous application during a live campaign, when campaign data was inserted into the algorithm, it outlined not only an accurate win number, but also purposed an accurate vote total for *Each candidate.  I would be very curious to continue this experimentation.