Assessing Team Performance

Yesterday, Chelsea lost a reasonably close game away to Manchester City. Obviously, this was a disappointing result and we'd all have been fairly happy with a draw. However, two weeks ago, Chelsea hosted Blackpool at Stamford Bridge. Would anybody have been pleased with a draw then? Clearly not.

What makes a draw a decent result versus a calamity? The location of the match and the quality of the two teams. Wolves, for example, would be quite happy to draw all of their games against Champions League calibre teams. Chelsea would not. We have the framework to run win probability calculations between any given teams, which means that we can also make points earned predictions for each side in every match, as long as we have an estimate of their true talent attacking/defending ability.

Fortunately, we have such an estimate as well - it's included in our weekly power rankings. These represent our best guesses as to a team's actual ability at any given time. Combine this with points probability and actual results, and then you have a metric that tells you whether a team have over or under performed in any given game. When West Bromwich Albion visited Chelsea for the opening game of the season, how many points did they actually drop? Even the most optimistic of fans couldn't have hoped to walk away with three points, so what should the expectation have been?

Chelsea vs. West Brom at Stamford Bridge (again, with my current numbers) gives the visitors a 5.9% chance of winning and a 6.4% chance of securing a draw. This means that the Baggies might have expected around a quarter of a point - any result would be a big over-performance, and losing isn't embarrassing in the slightest. What happens when we apply a similar philosophy to Chelsea in that match? Although they got three points, they were expected to get 2.69 anyway, so they only really 'earned' 0.31. These numbers are clearly biased against the favourites in that they can't record big upsets, only suffer them, but remember that we're just looking at performances relative to expectations. It's much easier to over-perform if everyone expects your side to lose.

Let's apply this to an entire week of fixtures:

Figure 1: Team over/under-performance during Premier League week six. Green corresponds to a positive result, red to a negative one, and yellow is neutral.

I think this is pretty neat, because it tells you just how much of a blow/boon any given result really is. For example, Chelsea cost themselves almost 1.5 points with their loss to City, but that's not too much more expensive in terms of expected points than Manchester United drawing 2-2 against Bolton at the Reebok Stadium (and our losing at City is a better result than a hypothetical draw against Bolton at Old Trafford for United). If we keep track of these figures, we can also remove the impact of a team's schedule from their numbers; if Chelsea have a total positive figure for the season, we can say that the Blues have done better than expected no matter how hard their schedule has been. As it turns out, Chelsea are in the positives, barely. I've compiled the figures for the whole league below: