Quick intro / recap

At the start of the season I mentioned that there would be a bit of a delay in getting the E Ratings live again. There were two reasons behind this:

1. The official data being collected for the National League, which had for years been at the same level as the EFL, was cut back significantly (presumably for cost reasons). To me this feels like a huge backward step that they’ll ultimately regret, but (that aside) it also broke the model, which needed that data to establish a starting rating for newly-promoted clubs from that division.
2. I wanted to make the model better, having read a slew of interesting posts over the summer about approaches that could potentially improve it, so I took it to pieces and started to rebuild it from the ground up. However towards the end of the summer my team at work won some additional contracts that soaked up most of my free time, so I kept having to park it.

I’ve still not gotten around to addressing (2), which is a huge job and needs a stretch of unbroken time (and lots of testing) devoted to it, but I was starting to get sick of not having the ratings around so I’ve slapped the old model back together and come up with a quick fix for (1) – basically using actual goals instead of expected goals to feed the ratings – which will have to do for now.

Latest ratings

Here is the latest update of the season to the E Ratings and how they predict the rest of the season will pan out. The rating system is explained here, but in a nutshell it’s based on the combined quality of chances that clubs create and allow, rather than their results.

The attack rating broadly equates to how many goals’ worth of chances a team would create against an average opponent (so higher is better), with the defence rating equivalent to the chances they’d be expected to allow (so lower is better). The overall rating is the difference between the two – effectively the average expected goal difference per match – so a positive number is good and a negative one is bad.

The graphic below lists each club in descending order of their overall E Rating and shows how this – along with their individual attack and defence ratings – has changed over the past 30 league matches. The red and green arrows indicate how the overall rankings have moved in the past month and the numbers in brackets show the ranks for each team’s attack and defence ratings.

It will shock nobody to learn that Wolves are the best-rated team in the division and continuing to climb, despite a shock defeat by Forest this weekend. Fulham have rediscovered their mojo lately with a sharp upturn in performances, underlined by a 6-0 demolition of struggling Burton.

The Brewers look to be the division’s weakest team, with Birmingham and Bolton also struggling, and it will be interested to see if Chris Coleman can arrest Sunderland‘s precipitous slide.

Predicting the rest of the season

Below I’ve used each club’s current ratings and those of their remaining opponents to predict how the rest of the season could play out. Each of the remaining fixtures has been simulated thousands of times, using the current E Ratings to generate probabilities for where each club will finish.

This graphic shows the cumulative probability of where each club could end up, in descending order of average points won. You can think of the ordering of the teams down the left hand side as a “best guess” of the final league table, with the coloured bars showing the relative likelihood of each club ending up in a certain section of the table:

Wolves are strong favourites to go up – again to the surprise of no-one – with Cardiff and Derby contesting the remaining automatic promotion spot most convincingly. There’s still plenty of life left in the play-off race, with Bristol City, Fulham and Brentford the outsiders looking most capable of carving out a top six spot.

The relegation battle is also wide open – while Burton and Birmingham – seem likelier to go down than survive, the final berth in the bottom three has at least two possible occupants in Sunderland and Bolton, but a few others could still get dragged in.