Why You Shouldn’t Let Data Scientists in Your March Madness Pool

Using statistics to attempt to predict future outcomes about otherwise unknown events is nothing new. Humans have improved their methods in recent years with computerized factual gathering, and tools for sorting and crunching huge sets of data. Such is the discipline we call predictive analytics—big data, if you will—and it’s ubiquitous across the professions. But it’s perhaps best illustrated in things we do for fun. Namely sports, and the annual college hoops bracketed bacchanal we call March Madness is a great example.

Big data bracketology

One of the foremost voices in data science and predictive analytics, Nate Silver of FiveThirtyEight.com applies algorithm-based methods in his NCAA Men’s Basketball Tournament predictions. Silver is perhaps best known for his uncanny data-driven political predictions, but he’s been using predictive analytics in his bracketology, in some form or another, since 2011.

Silver refines his approach each year for picking winners during tournament time. Since FiveThirtyEight is a publication first, his centerpiece is an attractive interactive map that viewers engage online when making their own picks, following along as the tournament progresses. FiveThiryEight’s 2017 NCAA tournament forecast rolls in data feeds from several different indexes sporting experts use to rank teams: They use sportswriters’ computer analyses taken from different points in the season, plus human-generated ratings like the “S-Curve” system by the NCAA, and a composite of subjective media ranking analyses. They make their own Elo rating system in-house, a ranking method traditionally used for chess player ratings.

Spinning data science into sports journalism

In years past, FiveThirtyEight data predicts the winner of the tournament matchups upwards of 70 percent of the time—consistently besting Las Vegas odds-makers, especially in the opening rounds of the tournament. Since Silver’s goal here is to entertain and engage his viewers more so than win gambling proceeds, he processes even more data for his real-time online map, which identifies potential upsets and gauges excitement in each matchup as it happens.

The “excitement index,” for example, measures the movement of teams’ probability for winning during their games. The more dramatic the movement indicates higher levels of drama on the court.

His visualizations feature real-time stats showing the win probability for each team competing at the moment, with computer systems crunching data on the fly. Time remaining in the game, score differential, possession of the ball, special consideration of a team shooting free throws at the moment—all of this is added into the matchup data as teams play and progress through the tournament.

Understanding the toolsets of data science through data journalism

Silver’s foray into March Madness provides a timely way to understand data science as a profession. Specifically, the emerging field of data journalism. Data journalism seeks to tell tangible stories with numbers—lots of them—the same way someone in a data science career would for insights in healthcare, election politics, marketing, computer science, and nearly every other industry vertical.

A computational journalist wears several hats. Their primary responsibility is piping in the data needed for processing real-time predictions and more refined data points. Theirs is a skillset akin to a full-stack programmer. That means experience with back-end programming languages like Python, object-oriented languages such as Ruby, and some front-end expertise typically in Javascript. Specific web frameworks handy for computational journalism might fit along the lines of Rails, Django, and node.js. On the database side, data professionals want to be familiar with document-based databases queried with MySQL, Postgres, and MongoDB.

A visual journalist has a web design background and knows HTML, CSS, and Javascript, naturally. There are several programming languages and frameworks specific to data storytelling. D3.js stands for ‘Data-Driven Documents,' and it’s a JavaScript library used for creating graphic documentation manipulated by data inputs. You have vector graphics scripting libraries also—Raphael, BonsaiJS, and PaperJS to name some popular tools.