The Premier League season is begun, which to all fans means only one thing. Data. We now have data from the first ten matches of the season, and we can begin adding this data into our projection engines. What? Is that just me? Well, anyway.
I don't expect the numbers to tell us very much about team quality until November or so. I'm very slowly adding in the week's results as part of my team projections, and the effects will be small. I am using the Shots-in-the-Box-on-Target system I described over the summer to estimate expected goals for each team as the eventual basis of my power rankings and season projections. For now, these game results account for only ~3% of the overall team quality estimate, most of it is my preseason projection. By week 10 or so I'll be using mostly season results, adjusted for competition. For the projections, I'm using the Monte Carlo / bivariate Poisson method I described a few weeks ago.
Week One Results and Changes
There were a few small effects on team quality estimates. Manchester City destroyed Newcastle, even before an idiotic shove to the head got Steven Taylor sent off. It was a match City were expected to win, but their dominance still registered—5 BC, 8 SiBoT, 3 SoBoT against just 1 SoBoT for Alan Pardew's side. On the flip side, relatively easy wins for Tottenham Hotspur and Chelsea did not improve their ratings as good sides are expected to beat Hull City and Crystal Palace actually somewhat more soundly than the two big London clubs managed.
The more notable effects were seen not on the evaluation of team quality but on season projections. A club that lost what should have been an easy home match may not lose much in overall projected team quality—anyone can lose a match one week—but it affects their top four and title chances. As you can see, all of the other top five clubs benefited in expectation from Arsenal's loss.
At the same time, as I have pointed out before, 33% is a really high number. It's roughly, for instance, the same as Arsenal's top four odds as estimated immediately after last March's North London Derby. And we know how that turned out. 27% became 100%. So one-in-three is very, very far from a death sentence. At the same time, it's a bad idea to spot your rivals three points when no one has a particularly difficult fixture.So it matters, and it doesn't.
Projected Premier League Table and Odds
In the following table, I have a whole bunch of projection data. W/D/L are rounded projections of wins, draws, and losses. Pts is a rounded projection of points. Because of rounding, sometimes W/D/L doesn't add up to points. The points projection is a better estimate of final position because it only gets rounded once. GD is projected goal difference. Team+ is overall estimated team quality, compared to an average league side. So Manchester City is rated 54% better than the average team, while Aston Villa are 84% as good as an average team. The average side is roughly a Swansea or a Southampton. Then I have the chances of different season outcomes, and the change from the week before.
|West Ham United||12||10||16||45||-11||83||0.5%||+.5||11%||-4||0%||0||2%|
|West Bromwich Albion||11||10||17||43||-13||83||0%||-1||16%||+4||0%||0||1%|
- Aston Villa also saw a nice effect from their win, jumping to the head of the trailing pack thanks to a three-point outcome projected to only about a 20% chance before the match. They still don't project to be good.
- You can see from the Team+ numbers that my projections are really struggling to see any differences in overall quality between Villa, Fulham, Norwich, West Brom, Newcastle and West Ham. The stats actually like Norwich the best by an extremely slim margin, but their home draw to Everton was slightly below points expectation for the week, while Villa and Fulham each picked up away wins. I think Fulham's win was pretty clearly a fluke—they scored their only shot on target—but a two point lead translates to a small lead in the overall projected table over Norwich because the margins here are so slight.
- My stats actually liked Southampton more than Swansea before the season, but the subjective numbers moved the Welsh club above their south coast competitors. The first week wasn't enough to flip-flop their team quality ratings yet, but again a three-point advantage gives the Saints a small step up on the Swans in the projected table.
- One or both of Hull City and Crystal Palace are probably terrible.