Darien student’s Elo ratings accurately predicting FCIAC basketball
Published 2:10 pm, Wednesday, February 10, 2016
It is safe to say this FCIAC boys basketball season is unlike any that has been seen in a long time.
Westhill dominated the league last year, with teams like Norwalk, Ludlowe and Stamford at the top of the standings all season.
The season before, it was Bridgeport Central leading the conference with a couple other top teams staying near the top from start to finish.
In those years, and for at least the last 30 years, it was pretty easy to predict which teams would be in the playoffs and which would be home come the end of February.
Not so much this season.
Trumbull and Danbury are both 8-1 in league play while Stamford is at 7-2, followed by four teams with three losses and three more with four losses. It is a jumble of contenders.
Having just passed the halfway point of the season, teams sitting at 2-6 are still in the mix for a playoff spot if things break right for them.
In fact, only New Canaan (1-8) and Central (0-8) seem to be out of contention, an astoundingly low number for the first week in February.
On any night there is no telling which teams will win and which will lose.
As a case in point on Monday night, Staples (3-6) beat Ludlowe (4-4) and St. Joseph (2-6) defeated McMahon (6-3).
So how can one accurately predict who will win night in and night out?
One FCIAC student may just have found a way.
Griffin Ott, a sophomore at Darien, has developed an Elo rating system he is applying to FCIAC contests with a good deal of accuracy, even in this topsy-turvy season.
For those not familiar, the Elo rating system was originally developed as a method for calculating the relative skill levels of players in competitor-versus-competitor games such as chess.
It is named after its creator, Arpad Elo.
In the last few years, Nate Silver founded a website called fivethirtyeight.com dedicated to predicting the outcomes of everything from football games to presidential races, adapting Elo’s original theory to fit team sports.
A fan of the work at 538, Ott decided to see if he could accurately predict the outcomes of FCIAC boys and girls basketball games, as well as calculating their chances of making the playoffs.
Ott is doing this for fun, he says, not for a class project, but just because he wanted a way to combine his passion for numbers with his love of sports.
“I just really like numbers, and sports is easy to do this stuff with because there is a lot of data to draw from,” Ott said. “So far, it has been pretty accurate, but it needs a larger sample size to become more accurate. Right now, it is still dependent on last year’s records, and teams in high school can change a lot but overall the results have been good.”
They have been good indeed.
Last week on Tuesday’s slate of games, Ott’s predictions were right in 13 of 16 games between the girls and the boys, including accurately predicting the Greenwich boys defeating then-unbeaten Trumbull.
Ott had Greenwich as having a 56-percent chance to win that game, and the Cardinals delivered.
On Monday, he picked every winner of the girls games correctly, but was off on four boys games. However, two of those losses were St. Joe’s beating McMahon and Staples upending Ludlowe, which he had going the other way.
The Elo system Ott uses measures wins and losses, margin of victory and strength of schedule among its factors.
Teams also get an extra boost in the predictions by playing at home, which is why Wilton was a 57-percent favorite to beat Danbury Monday night. Danbury won the actual game.
Each team receives an Elo rating which improves or declines based on wins and losses as the season goes on.
An average Elo rating is 1,500, and a team’s rating will adjust with each game. If the team wins their rating will move up, if they lose it goes down
When an upset occurs that swings teams’ Elo ratings even more.
Norwalk, for example, gained 111 Elo points by virtue of picking up a few upset wins the last few weeks, moving the Bears back to 1,500 Elo points for the season.
The Danbury boys are currently at 1,722 Elo points and the Warde girls are at 1,962 at the top of the FCIAC standings.
The biggest upset thus far?
The Darien boys had just a 10-percent chance of beating Ridgefield but pulled out a 54-48 win over the Tigers.
The girls side — where a bigger gap between the top tier teams and the bottom — has been easier to predict this season, but even with so many unknowns on the boys side, the numbers do not lie.
There is a series of math equations used to calculate the Elo scores which are too complicated to list here, but one aspect bares mentioning because it is the heart of the rating system.
The most important parameter to consider is what is known as “K.” K is the weight constant for the season played. The value of K in the NFL with just 16 games is 20 and in MLB with many more games K represents 10.
Ott sets K at 40 for his analysis. K indicates how much to update the ratings after each game. That may not mean much to anyone unable to compute these numbers, but it means everything to Ott’s analysis.
“I set it to 40, which was for a few reasons,” Ott said. “There is significantly more year-to-year turnover in high school sports than professional sports because people are graduating every year. The starting Elo ratings were based on last year and the year before, so I wanted each team’s rating to adapt more quickly to the current season.
“A good example of this is Central, who had a high starting rating of 1,725, but are clearly a lot worse this year. Their rating has already gone down to 1,299. Also, because it is a short season, I did not want it to be too slow in adjusting. When I was originally making the spreadsheet, I experimented with a few values and used my judgment on which seemed to produce the most logical amount of gain for a certain game.”
Ott adjusts his rankings every week, publishing a recap of the week before and a look at the week ahead on a Google site he calls FCIAC Sports Rankings.
So who does Ott project as the top three teams in boys and girls basketball going into play this week?
On the boys side he has Danbury with a 27-percent chance winning the championship, Wilton at 20.4 percent and Trumbull at 15 percent. In girls basketball, Ott has Warde at 34.9 percent, Trumbull 21.3 and Stamford 17.3.
The playoff odds are calculated using each team’s current Elo ratings and a simulation of the rest of the season run 5,000 times.
Of course, the teams still need to play out the season on the court, but Ott has given us an interesting way to analyze the season and predict outcomes off of it.