NHL Statistics Study
By D. Rehan, P. Salomone, K. Gahunia, and B. Simard
Worldwide popularity of sports has grown tremendously in the last few decades. One sport that is close to the heart of many
Canadians is hockey. Many hockey fans (sports fans in general) enjoy collecting and analyzing statistics of the game. Whether
this is for hockey pools, rotisserie leagues, computer simulations or just to learn more about the game, each year many magazines
publish special issues that try their best to forecast sports statistics.
As the use of data collection and statistics becomes more and more prevalent throughout the professional sports industry, the
uses for this data grow. The statistics can become important for players in negotiating contracts, for team coaches in deciding
game strategies, and for the league management in determining any significant trends, and how these trends can be manipulated.
This report examines the number of goals scored per season in the National Hockey League, partly because the data is of good
quality, and partly because this data has been collected over a much longer time period than most other statistics in the NHL.
This report focuses on forecasting the number of goals scored in a. It is an analysis of available statistical models that
could provide the most accurate forecast for total goals scored in the National Hockey League's (NHL) regular season in the near
future. The past decade has seen the NHL expand the number of teams in the league to reach the lucrative American market
(Refer to Appendix A-1 for expansions since 1946). Many experts argue that this expansion has watered down the talent in the
league. As a result, the NHL has been experimenting with rule changes to encourage an increase in goal scoring.
To conduct our analysis, the total numbers of goals scored per season from the 1946-47 to 1998-99 were used. The data collected
from this 52 year sample is included in Appendix A-2 of this report. It consists of the yearly statistics for the number of teams
in the league, the number of games played, total goals scored and the computed values of average goals per team and average goals
per game. With this data we expect to draw conclusions on the total goals per season and the average goals per game.
Several statistical models such as Linear Regression, Auto-regressive, Simple Exponential, Holt's Double Exponential and ARIMA
were used to find the most accurate forecasting model. The models will be compared using model validation statistics, error
statistics, and any other statistics that will help us determine the 'best' model.