Election Results Prediction With Presidential Plinko
You see the New York Times’ election needle and you start to worry. You are reading FiveThirtyEight and wondering how your candidate can lose by gaining a seemingly impressive number of “points” or polls before the elections.
Welcome to a world of uncertainty, which many people, including myself, tend to disregard when looking at statistics, probability, and unambiguous assumptions. A survey based forecast can be right and wrong at the same time; yes, in theory your candidate might be ahead in the survey, but still end up losing because the survey did not quite accurately capture the will of the people he intended to represent. And that little thing called the margin of error is also important, although we often pretend it doesn’t exist.
To help us understand this better, Matthew Kay, assistant professor of computer science at Northwestern University, has devised a rather clever way to show how uncertainty can affect survey data.
He uses the classic Price Is Right game, Plinko , to show how forecasting can still produce results that change in one direction or another. It all depends on how the Plinko chip or virtual ball falls. As Kay describes:
“In short, I approximate the predicted distribution of each forecaster with a skewed and scaled binomial distribution that ultimately determines the height of each board. I then determine the likely paths across the board that would lead to the final predicted distribution, which is displayed as a quantile scatter plot . Thus, although the output appears to be random, the final distribution is exactly the published forecast, up to the resolution of the scatter plot. Full details of the methodology and source code are in the Github repository . “
It’s all good, but I find myself throwing ball after ball in Plinko Kei’s game – which you can do for yourself on his court too. You have the option to throw one ball or just let them fall.
I think the first is the path you want to go if you generalize that your only virtual ball represents choices, and however it lands, is the result we will all get. Spoiler alert: Your candidate may not win, but that’s the whole point of this exercise. Probability is a science, not a time machine. Remember 2016?
As for the aforementioned Needle, which many of us are familiar with from The New York Times, it is a useful tool for gauging election sentiment when people understand what they are doing. Kay argues that the use of the needle, however, allowed people to more easily pervert the noble approach of the times.
“I think Igla did one thing right and the other didn’t. What was right was that this kind of animation could help people experience uncertainty. This makes the visualization more powerful and the uncertainty harder to ignore. Visualization made people anxious because they weren’t sure what they were concerned about. But if you are not sure about what worries you , you should be worried, ” he writes.
“However, I think the needle has also fallen prey to a deterministic design error : many people are more willing to associate the mechanism of the needle with some deterministic dimension rather than an indeterminate quantity. It is clear that these people thought that the rapid movement of the arrow reflects the same rapid change in the forecast itself. “
Now when can we get our election predictions modeled with Hole in One (or Two) ? Here’s what I want to know.