Money Puck: Booth Professors Publish Paper on Hockey Analytics
Chicago is a city that loves its hockey. The University of Chicago Booth School of Business is a school that loves sound business decisions.
When the National Hockey League changed their salary restrictions after the 2012-13 lockout, a new metric created by Booth School of Professors Robert Gramacy and Matt Taddy, together with Professor Shane Jensen of Wharton, could help general managers figure out which players are worth their time on the ice in efforts to best shape their teams cost effectively.“Estimating Player Contribution in Hockey with Regularized Logistic Regression,” which was published in a recent issue of the Journal of Quantitative Analysis and Sports, discusses a more precise assessment of a player’s involvement in a goal by calculating the partial effect of each player.
Aside from goals and assists, the most important player statistic in hockey is a player’s plus-minus value—players are given a plus if they are on the ice when their own team scores, and a minus if the opposing team scores. The issue is that players can have positive values even if they’re never part of a key play or they can hold negative values despite impressive contributions on otherwise fruitless lines.
Because the current system doesn’t account for the effect of other players on the ice, special teams and other factors, individual players could be receiving inflated or deflated salaries that may not be in line with their abilities.
When the authors analyzed NHL players from four regular seasons (2007 to 2011), they found decidedly different data on player performance than traditional figures, deducing that a team made up of the league’s most expensive players wouldn’t compete much better than a low-budget team. They also suggest that well-praised and well-paid players like Sidney Crosby, Jonathan Towes, and Zdeno Chara should even be re-evaluated.
To back up their controversial findings, Gramacy and Taddy began this blog where they link to weekly, updated cumulative and current-season player ability estimates. The code then automates the analysis (including data scraping from NHL.com) for full transparency. As the season has developed they, and others, have drawn interesting comparisons between players and commented on high profile trades.
