You may have heard of FiveThirtyEight.com before.
You may have heard of Elo ratings before.
You may even have heard of FiveThirtyEight.com’s NFL Elo ratings.
Well this is neither FiveThirtyEight.com, nor the NFL, but we can certainly use the same method to attempt to gauge team strength in the CFL.
What are Elo ratings? The easiest answer is to go read FiveThirtyEight.com’s NFL Elo ratings Q/A.
The TLDR (too long didn’t read) version is that it is a credit system.
Teams have a rating (average is 1500) and gain points if they win a game and lose points if not. Each game begins as worth 20 rating points. Factors such as home field advantage, win margin and reduced returns for blowouts are used to adjust the value of a game for the winner and loser.
A high-rated team at home that wins by one over a low-rated team gains the fewest points, while a low-rated team on the road who wins in a blowout over a high rated team gains the most. The losing team always has the same amount of points taken away from their rating as the winning team gains to their rating.
Are Elo ratings perfect? Definitely not.
Elo Ratings are unaware of personnel, coaching, weather, match-up, scheme, venue and many other important influences on which teams win a game.
Are they useful? Maybe.
Elo ratings can be used to get expected win percentages for every game. This expected win percentage can be used to simulate how often certain standings and playoff situations will occur.
CFL franchise Elo ratings
Below is a franchise Elo rating chart for the 2019 CFL schedule up until the end of Week 15.
What is a franchise Elo rating? It is an Elo rating tracked from back before the CFL was even formed in 1958 for every team. Franchise Elo Ratings are regressed two thirds from the end of the previous year to attempt to represent the uncertainty of team change.
Franchise Elo rating has its own valid criticisms.
Sometimes a team undergoes so much change that the Elo rating at the beginning of the season can be over-inflated or under-representative.
The alternative to franchise Elo rating is season Elo ratings.
CFL season Elo ratings
For season Elo ratings every team begins at an average Elo rating of 1500 instead of a regressed value from the previous season. This means data — like the Stampeders’ franchise success under John Hufnagel or Dave Dickenson, for instance — is not accounted for.
Below is a 2019 season Elo rating chart for the CFL schedule up until Week 15.
You may have noticed that certain teams are at different values. Generally, by the end of a season, season Elo rating and franchise Elo rating are almost the same value.
Elo ratings can be useful. Elo ratings can let us simulate the odds that CFL teams end up with a certain position in their division. If we look at the odds headed into Week 16 in the CFL from franchise Elo rating we can judge that a couple of things are settled.
The crossover expectation is currently at 96 per cent.
The expected participants in the playoffs will not include B.C. in the West, nor Ottawa or Toronto in the East. The total between those three of any of them making the playoffs is a measly 5 per cent.
Hamilton is almost guaranteed to be first in the East (95 per cent) and host the East Final after a bye.
Montreal should be preparing to host Edmonton in the East Semifinal as they crossover from the West.
Hamilton is the favourite to make the Grey Cup (73 per cent). This is an accumulation of factors such as having the bye and then playing the weaker two of the other five playoff teams.
Things are close in the West Division. Saskatchewan is currently looking at third place in the West, which would mean two road games versus Winnipeg and Calgary to make the Grey Cup.
Grey Cup match-ups
Hamilton features heavily given how expected it is that they will make it out of the East over Montreal or Edmonton.
Hamilton is in the Grey Cup in 73 per cent of the match-ups with the most likely competitors being Calgary, Winnipeg or Saskatchewan based on their current standings expectations.
The first all-West Grey Cup match-up of Calgary and Edmonton, often talked about in Alberta, only has 3.57 per cent odds of occurring.
That’s not what the standings say/I don’t like your Elo ratings
Great. I don’t like them myself.
I already mentioned all the things Elo ratings can’t do. However, almost all simple analytical methods for team strength will have the same issues. ESPN’s proprietary NCAA FPI (Football Power Index) is one much more complicated method that uses many more features than Elo rating.
One fundamental thing to remember is the Elo ratings are very time-dependent. Credit is gained once. A loss at the start of the season to a team that was bad the year prior will look bad, however if it turns out that team is the best by the end of the season the rating points lost are unchanged.
An advantage of Elo ratings is that they are functional Week 1. Franchise Elo ratings have data regressed from the previous season and even season Elo ratings can be calculated immediately after.
Rating percentage index
One popular alternative to Elo rating is RPI or Rating Percentage Index. Rating Percentage Index is an attempt to take the win-loss record we all see in the standings and add some context to it.
That context is to create a win-loss percentage value which takes into account if the wins were against teams with bad records or good records. This value can be calculated newly after every game.
College sports (particularly in the NCAA) have used RPI variants for years to deal with unbalanced schedules and teams playing weak opponents. RPI is great for getting Strength of Schedules (SoS). Instead of just looking at the win-loss records of a team’s future opponents, we can now add the context of win-loss records of opponents not-including the team in question.
The chart below is comprised of RPI strength of schedule values and rankings for CFL teams. Both past schedule and future schedule are indicated. Winnipeg is on top of the CFL West but they have had the weakest schedule to date and the strongest going forward. Andrew Harris has returned but Matt Nichols’ possible return is unknown. One of the biggest challenges remaining is a home and home with Stampeders in Week 19 and 20.
So RPI sounds great, but now we are ignoring how many points were scored by teams. One of the most common arguments among fans is that one team beat another team by more points so they must be better. RPI, however, ignores win margin.
Markov Chain modelling
Another solution? Markov Chains.
You lost me. Yes, I probably did.
The TLINR (too long I’ll never read) version of Markov Chains is that allow us to connect every team by arrows weighted by how many points they scored. These points represent how much ‘strength’ flows from one team to another. This creates a network where ‘strength’ flows between teams. We can let this ‘strength’ network sit a while until the ‘strength’ settles with certain portions distributed throughout all the teams in the network. Then we can judge who has the most ‘strength.’
This creates a method that is much like fans arguing about ‘X’ team scoring seven more points versus ‘Z’ team than ‘Y’ team scored against ‘Z’ team.
One danger with Markov Chains is that an extreme unlikely result — like a team without their starter for one game who gets blown out — will unbalance the whole process. Unlike fans who may be able to rationalize away considering this game, Markov Chains treat it as equally valid as any other. Additionally, like RPI this method is quite inaccurate or even unable to be calculated early in the season. Both methods need to gain data before they can be of use.
One way to do rankings is not do them.
Okay, if you have to do them, I prefer to look at a variety of angles and let rankings be more educational than argumentative absolutes.
Below is a ranking table of CFL teams including all three methods described. Both franchise and season Elo ratings, as well as Markov Chains and RPI. The table is ordered by the season values being changed into relative rankings and averaged.
The result is tiers of ranking.
2. Calgary; Winnipeg
3. Saskatchewan; Montreal
5. B.C.; Toronto
But, but, these rankings are lame
Think these rankings aren’t good? Make your own.
This is the best thing about rankings.
They mean nothing and anyone can make them.
You don’t even have to be consistent — they can even be about pork products.