Are Hockey Season Predictions Meaningless?
It's that time of year when the prognosticators, armed with textbook knowledge of the game or sophisticated algorithms, attempt to predict the final NHL standings. While this exercise seems fruitless, there are rare occasions when the crystal ball gazers get it right; such success then stirs up enthusiasm for undertaking the exercise again and again.
Take Electronic Arts, for example. This video game company's simulations have a mixture of both stunning accuracy and spectacular failures. They have already predicted a Stanley Cup win by the Nashville Predators over the Montreal Canadiens for the 2016-17 season, with the Vancouver Canucks dead last in the NHL.
Electronic Arts isn't the only one predicting a bad year for the one-time Stanley Cup finalists. USA Today and The Hockey News had a similar pre-season assessment for the Canucks, with the latter foreseeing the Predators behind the Dallas Stars in their division, and the Canadiens behind both the Tampa Bay Lightning and Florida Panthers.
However, even The Hockey News questioned whether it was possible to accurately predict anything about such an unpredictable game. In an article posted just as the pre-season analysis commenced last year, THN said many forecasts are based on unpredictable factors such as player talent. The slippery criterion of "talent" comes from offensive shot rates, penalties and time on ice. Many commentators, including Rob Vollman, who made predicting hockey stats his own cottage industry, often admit they got it very wrong.
It's hard to argue when predictors get it right, especially when the details make it difficult to write off as a fluke. EA's "Madden NFL" simulation correctly predicted the Super Bowl winner nine out of the previous 12 years before Super Bowl 50. In 2016, EA predicted the Carolina Panthers beating the Denver Broncos by 24-20, while the Broncos actually won 24-10. In Super Bowl 49, however, EA not only got the score right but foresaw several key plays in the game.
At what point does a sound predictive model become a lucky guess? Does naming the winner 9 games out of 12 — getting it right 75 percent of the time, even using detailed predictive factors — really amount to anything better than the flip of a coin
In a hockey season with 82 games for each of 30 teams, 1,230 contests are played. It seems unfathomable that any predictions could account for the high number of variables that come into the win-loss tally of each franchise.