This research examines two aspects of hockey teams in the National Hockey League (NHL). The first aspect of the research was to determine significant seasonal factors and corresponding weights using discriminant analysis in predicting which teams would make the playoffs. Data was collected over seven seasons initially considering 60 variables. Total goals against, total goals scored, and takeaway totals for a season, were enough to correctly predict whether a team made the playoffs 87% of the time. The second aspect of this research uses regression analysis to create models estimating the probability of a hockey team winning the game, and also estimating the difference in goals scored between the two teams playing in the game. In developing these models, a random sample of games was taken from the 2009-10 and 2010-11 season and the in-game values on 60 variables were recorded. A logistic model was then developed estimating the probability that a team would win the game if the values of the following in-game variables found to be significant were known: save percentage margin, shot margin, block margin, short-handed faceoff percentage, short-handed shot margin, and even-handed faceoff percentage. A second model was formed for estimating the goal difference of a game that was also based on the random sample of 52 games taken from the 2009-10 and 2010-11 seasons. In the goal difference model, save percentage margin and shot margin were significant and accounted for over 93% of the variation in score difference. The probability model and the goal difference model were validated using the actual in-game values from a different data set that was not used in the development of the models. The probability model and the goal difference model were correct 98% and 100% of the time for these 52 games when the actual in-game values of the variables were used. The probability model and the goal difference model were then used in predicting the results of another random sample of 60 hockey games from the 2011-12 season when the actual in-game values of the variables were not used, but instead averages of the values of these variables were used based on the previous three hockey games both teams had played. In this case, the models correctly predicted the results of the hockey games 65% and 66.7% of the time. These percentages were found to be significantly larger than the percentages of the time one correctly predicted the winner of a game by always selecting the home team, or always selecting the team with the better record, or always selecting best team as determined by a handicapping website. It was found also both over the entire season and over the short term that defense has a stronger impact on winning the game than offense. Perhaps teams may want to rethink their most important players.
© Copyright 2014 International Journal of Sports Science. Scientific & Academic Publishing. All rights reserved.