August 22, 2011

Hardy: 2011 Power Rankings model explained

Trevor Hardy
CFL.ca

The annual CFL bye weeks are upon us. More than anything, the bye week is valuable in that it allows clubs some time to rest and heal, as they gear up for the drive to the Playoffs.

Teams will also use this time to assess what has gone right and wrong during the first seven weeks of the season. It is in that spirit that we at CFL.ca have decided to do the same with our mathematical Power Rankings. How have they performed? What has gone right and what has gone wrong?

Remember that this year, the CFL Power Rankings are being determined using two methods. The first method has Matt Cauz, resident CFL.ca expert, providing his subjective rankings and commentary each week. The second method uses a pre-determined mathematical formula to estimate the scores each week based on nine key statistics.

The Mathematical Formula

This pre-determined mathematical formula was obtained by using a statistical tool known as regression analysis and is based on information gathered from all regular season games played during the 2004 to 2010 seasons, inclusive – 1,044 observations in total.

Regression analysis is an extremely popular and important statistical tool that is used in a wide range of professions and disciplines. Any valid regression analysis has the potential to tell us two important things: whether a certain variable has “predictive value” and how much impact that variable has on a result.

With this in mind, our regression analysis first suggested that – of all the statistics in football – the nine most “predictive”, or important, indicators of team success were:

1. Quarterback Efficiency Rating
2. Time of Possession, as measured by percentage
3. Rushing Yards
4. Turnovers on Downs
5. Fumbles
6. Sacks Taken
7. Punt Return Yards
8. Kick-Off Yards
9. Kick-Off Return Yards

The impact that these variables are estimated to have on the outcome of a game give us our “Mathematical Formula”, as follows:

Quarterback Efficiency Rating x 0.137, plus
Time of Possession (%) x 8.602, plus
Rushing Yards x 0.027, less
Turnovers on Downs x 1.299, less
Fumbles x 0.499, less
Sacks Taken x 0.629, plus
Punt Return Yards x 0.017, plus
Kick-Off Yards x 0.048, plus
Kick-Off Return Yards x 0.017.

In other words, what this means is that every Quarterback Efficiency Rating point is worth .137 points in any given football game, every Rushing Yard gained is worth .027 points in any given football game, and so on. 

Each week, we’ll determine a club’s predicted offensive score by inputting the statistics from the game into the formula (the defensive score is simply the opposing Club’s offensive score). In fact, we encourage you to try it yourself over the rest of the season so you can follow along!

The Results

The following table summarizes the results of the mathematical formula by game, and compares the results to the actual score:

Game # Predicted Actual
9 BC 32 MTL 40.0 BC 26 MTL 30
10 WPG 19.8 HAM 19.5 WPG 24 HAM 16
11 TOR 27.7 CGY 22.3 TOR 23 CGY 21
12 EDM 47.3 SSK 22.7 EDM 42 SSK 28
13 TOR 12.2 WPG 24.2 TOR 16 WPG 22
14 CGY 28.7 BC 27.7 CGY 34 BC 32
15 MTL 36.3 SSK 24.3 MTL 39 SSK 25
16 HAM 16.3 EDM 32.4 HAM 10 EDM 28
17 CGY 19.1 WPG 17.0 CGY 21 WPG 20
18 TOR 20.6 MTL 41.1 TOR 17 MTL 40
19 SSK 2.3 HAM 35.9 SSK 3 HAM 33
20 BC 16.4 EDM 36.0 BC 17 EDM 33
21 HAM 37.4 BC 25.4 HAM 39 BC 31
22 WPG 43.7 TOR 24.5 WPG 33 TOR 24
23 EDM 24.2 CGY 20.8 EDM 24 CGY 19
24 SSK 37.1 MTL 22.5 SSK 27 MTL 24
25 BC 24.5 WPG 28.0 BC 20 WPG 25
26 MTL 19.7 HAM 36.9 MTL 26 HAM 34
27 TOR 21.7 EDM 24.4 TOR 25 EDM 26
28 CGY 24.6 SSK 29.6 CGY 22 SSK 18
29 MTL 38.1 TOR 26.6 MTL 36 TOR 23
30 EDM 13.4 WPG 35.8 EDM 16 WPG 28
31 SSK 17.6 BC 23.5 SSK 11 BC 24
32 HAM 23.6 CGY 30.7 HAM 20 CGY 32
33 EDM 8.6 MTL 27.2 EDM 4 MTL 27
34 CGY 45.0 SSK 31.8 CGY 45 SSK 35
35 TOR 38.7 HAM 40.8 TOR 32 HAM 37
36 WPG 25.9 BC 10.4 WPG 30 BC 17

Summarizing these results, the mathematical formula has “predicted” the correct winner 27 out of 28 games – the only game where the incorrect winner was predicted was Game #28, where the nine statistics told us that Saskatchewan should have won the game against Calgary 30 to 25. Instead, they lost 22 to 18.

Overall, the mathematical formula has done an outstanding job of predicting scores – whether they are close games or lopsided games. For example, Game 14 between Calgary and BC was an extremely tight game. The formula reflected this, with a score differential of one point. On the other hand, Game 19 was a rather lopsided game in Hamilton’s favour, and the formula reflected this outcome, as well.

We hope that you enjoy reading the CFL Power Rankings each week – we encourage you to continue to track the results and remember to provide us your feedback, as well!