Proving That Data Science Works

When I say the word “proofs,” most people think of their high school geometry classes and the pain that ensued. By and large, proofs are seen as some sort of academic exercise that has no place in the real world.

That couldn’t be farther from the truth. Let’s take a look…


In about 75 words, we showed that if you take any two even numbers – positive, negative, close to 0, close to a billion, whatever – and add them, you’ll get an even number.  That result might seem unimportant, but it illustrates the point that proofs are incredibly powerful.  There are an infinite number of even numbers and we literally took all of them into account in just four lines.

So how does this relate to politics?

We can use proofs to show lots of things.  For example, in my first week at Øptimus, I proved the best method for minimizing the margin of error when sampling people from different groups according to different specifications that our team had in mind. That proof was much longer than the one above; I marked up walls of whiteboards in the office with Lagrange multipliers from multivariable calculus and conservative variance estimation from survey sampling. But the point is, with a little bit of legwork, some EXPO markers, and a Keurig, we were able to prove the optimal way of doing something.

Gone are the days when political decisions are made by a group of individuals sitting in a conference room until a consensus was achieved or the loudest opinions prevailed. These days, politics is moving toward a sleeker, more efficient operation.

We don’t rely on sweeping generalizations like “We want to do better with older Republicans and many of them watch Bill O’Reilly’s show, so let’s pour our TV budget into commercials during that block.”

Øptimus runs the numbers and we might conclude that we get more impressions and that it’s cheaper to run ten commercials on other shows than one on Bill O’Reilly.  We can pinpoint exactly how to maximize our spending or minimize our losses. This is about disrupting the “traditional” way of doing things with data science.

Data science isn’t some silver bullet or one-size-fits-all solution where we just run pre-packaged computer code.  In order for this to work on the front lines, we let the political team decide what questions they need answered… then we answer them.  We develop a method, prove that our method is the optimal method, implement the method to find an answer, and then turn the results back to our clients where they use it to make data-driven, informed, optimal decisions.

The proof is in the pudding: data science works.