The 2019/20 Hero ISL State of the Season Address
It's time to wrap up another disappointing international break, put heartbreaks in the past and look forward towards the rest of the ISL season. We are currently 4 games into the season, which is the perfect time to forecast how the season will move forward.
First up, it would do well to look at how the teams are performing this season and take stock of where they stand in terms of historical performances. An important concept to keep in mind here is "Regression to the Mean".
Due to the small sample size, some teams maybe be over or under-rated by the Expected Goals model. For example, in the below graph, most 2019/20 ISL teams are sticking to the periphery of the graph. This is because it's still early in the season and we do not have enough data to accurately rate these teams.
As the season progresses, we should see these teams fall back towards the pack, ie, perform more like past ISL teams.
Based on the above graph, it's important to remember the following points before proceeding,
Hyderabad, Chennai and Mumbai are not as bad as they seem.
Jamshedpur are very intriguing. They've been creating plenty of chances and look impressive, but are also conceding plenty of chances
Chennaiyin have scored no goals this season but are among the league leaders in xG. Eventually, they might regress to the mean and actually score a goal.....
Kerala 19/20 are considered one of the worst teams in the last 3 years of the ISL by the numbers. They neither create nor concede at an ISL standard rate. Surely, they won't continue being this boring?
Goa and Bengaluru might not be as dominant as they look.
Keeping the above in mind, we can take a look at the playoff qualification chances for each team. FC Goa and BFC are definitely being over-rated by about 15-20% here, in my opinion. The rest of the predictions, however, look good to me. It will be interesting to see how the ISL plays out, but I think it will be tough to argue with a top 4 of Goa, Bengaluru, ATK and Jamshedpur, with NorthEast and Mumbai the potential challengers.
With that pesky playoff business out of the way, we can forecast the final finish probabilities for each team. The model takes the xG created and conceded by each team, corrects it for historical performances in the league and against a particular opponent and simulates each match in the entire league 10,000 times. The league table at the end of each simulation is used to calculate the total probability of a particular team finishing in a particular spot.
Chennaiyin and Kerala's probability distributions immediately stand out from this graph to me. They both are almost replicas of each other, with Chennaiyin favored to finish slightly lower than the Blasters. This is incredibly interesting as Chennaiyin have been a very open team with plenty of chances created and conceded while Kerala have been the opposite - part of games where nothing interesting happens. It's going to be fascinating to chart how these teams perform as we progress through the league.
NorthEast and Mumbai are two other teams to keep an eye on. It looks like the jury's out on these teams as their (Mumbai in particular) performances have swung wildly from game to game. The model says that they have an almost equal chance of finishing anywhere between 3rd and 8th.
The rest of the predictions look reasonable with the key talking points being ATK's 34% chance of finishing 3rd and only 9% chance of finishing top 2 and Hyderbad being 96% to finish in the bottom 2.
Agree/disagree with the results? Find and tell me on Twitter - @SgtSaltnPeppa