To test the predictive validity of my "Fancy Stats" ranking model, which I explain here, I applied it to the 2013/14 season. The correlation between a team's rank in my model and their NHL points was r =.56, with an r-squared = 31.4%. This means 31.4% of the variance in a team's points is explained by the model. This is on the low side, but also note that the model does not account for the overtime period and shoot-outs. I don't count these situations in my model because they are irrelevant to playoff performance.
According to my model, the top 8 teams were:
1. Los Angeles Kings
2. Boston Bruins
3. San Jose Sharks
4. Chicago Blackhawks
5. New York Rangers
6. St. Louis Blues
7. Vancouver Canucks
8. Ottawa Senators
Among the top 5, 3 of the teams made it to the conference finals--LA, Chicago, & NYR. The Stanley Cup finalists were ranked 1st (LA) and 5th (NYR).
As for the other playoff teams in my model's top 8, Boston lost to Montreal (2nd round), San Jose lost to LA (1st round), and St. Louis lost to Chicago (1st round). Vancouver (25th in NHL standings) and Ottawa (21st) didn't make the playoffs.
According to my model, teams ranked from 9 to 16 were:
9. Dallas Stars
10. Tampa Bay Lightening
11. Columbus Blue Jackets
12. New York Islanders
13. Arizona (Phoenix) Coyotes
14. Detroit Red Wings
15. Philadelphia Flyers
16. Winnipeg Jets
Teams who made the playoffs, but missing from my model's top 16, were Anaheim (19), Pittsburgh (20), Minnesota (23), Montreal (25), and Colorado (26). How did these non top-16 teams fare in the playoffs? Anaheim lost to LA (2nd round), Pittsburgh lost to NYR (2nd round), Minnesota lost to Chicago (2nd round), Montreal lost to NYR (3rd round; conference final), and Colorado lost to Minnesota (1st round).
Seems like my model is good at predicting the playoff performance of teams in the top 8, but not so much below that. However, after the 1st round, the team ranked higher in my model won more often than lost. The one exception to this trend was the underdog, Montreal, who beat Tampa Bay (ranked 10th in the model) in the 1st round and Boston (ranked 2nd in the model) in the 2nd round.
I'm somewhat satisfied with the model. It's not overly strong in its prediction of NHL standings, but at least it's over 0.5. Also, recall that I didn't factor in overtime and shootouts, which includes anywhere from 10 to 20% of a team's games. My main interest, though, is whether the model can predict playoff performance. The model correctly predicted 11 of the 15 2014 playoff series, which is a 73% accuracy rate.
At a later date, I will apply the model to other seasons and see how things turn out.
Note. Data courtesy of stats.hockeyanalysis.com and NHL.com.
According to my model, the top 8 teams were:
1. Los Angeles Kings
2. Boston Bruins
3. San Jose Sharks
4. Chicago Blackhawks
5. New York Rangers
6. St. Louis Blues
7. Vancouver Canucks
8. Ottawa Senators
Among the top 5, 3 of the teams made it to the conference finals--LA, Chicago, & NYR. The Stanley Cup finalists were ranked 1st (LA) and 5th (NYR).
As for the other playoff teams in my model's top 8, Boston lost to Montreal (2nd round), San Jose lost to LA (1st round), and St. Louis lost to Chicago (1st round). Vancouver (25th in NHL standings) and Ottawa (21st) didn't make the playoffs.
According to my model, teams ranked from 9 to 16 were:
9. Dallas Stars
10. Tampa Bay Lightening
11. Columbus Blue Jackets
12. New York Islanders
13. Arizona (Phoenix) Coyotes
14. Detroit Red Wings
15. Philadelphia Flyers
16. Winnipeg Jets
Teams who made the playoffs, but missing from my model's top 16, were Anaheim (19), Pittsburgh (20), Minnesota (23), Montreal (25), and Colorado (26). How did these non top-16 teams fare in the playoffs? Anaheim lost to LA (2nd round), Pittsburgh lost to NYR (2nd round), Minnesota lost to Chicago (2nd round), Montreal lost to NYR (3rd round; conference final), and Colorado lost to Minnesota (1st round).
Seems like my model is good at predicting the playoff performance of teams in the top 8, but not so much below that. However, after the 1st round, the team ranked higher in my model won more often than lost. The one exception to this trend was the underdog, Montreal, who beat Tampa Bay (ranked 10th in the model) in the 1st round and Boston (ranked 2nd in the model) in the 2nd round.
I'm somewhat satisfied with the model. It's not overly strong in its prediction of NHL standings, but at least it's over 0.5. Also, recall that I didn't factor in overtime and shootouts, which includes anywhere from 10 to 20% of a team's games. My main interest, though, is whether the model can predict playoff performance. The model correctly predicted 11 of the 15 2014 playoff series, which is a 73% accuracy rate.
At a later date, I will apply the model to other seasons and see how things turn out.
Note. Data courtesy of stats.hockeyanalysis.com and NHL.com.
RSS Feed
