When the Premier League restarted I decided to develop a model to predict relegation. The model was based on an assessment of the quality of the opposition and their likely motivation. Four rounds of games might be a bit early to re-assess things, but I think I have learnt enough to make some revisions. After all, learning about what works and what doesn’t has always been the reason for doing this – and there are only 10 games in total to work with.
My main points of reflection to date include:
- There is not much you can do about a 9000-1 shot such as Hawkeye not noticing that the ball was obviously over the line in the Watford-Sheffield Utd (an event Hawkeye later apologised to Sheffield Utd about). This got Watford a point that was not predicted.
- I won’t complain about 93 minute winners because that’s part of the game, and I’d just sound like a loosing punter, but as with the Hawkeye, there is a need to take into account unpredictable events.
- For both the reasons above I’ve decided to go for a spread of points in the prediction. You can look upon this as a cop out or as a more realistic assessment of an unpredictable future.
- Perhaps more importantly, I’ve learned that bad teams are bad teams for a reason and I originally over-estimated the value of motivation. The old adage that a motivated poor team will beat a less motivated better team does not seem to hold true, especially when the bad team are really bad. The calculations for future points are now more closely related to past performance.
The new predictions
So what does this mean for the new predictions? In the table below you can see the changes from the original Version 1 to the new Version 2 that was created after four rounds of games.
The predicted points for most teams have been reduced to reflect their own abilities and only the ranges for Brighton and West Ham are still within the original predictions. West Ham are the big winners, with the model predicting they have moved clear of relegation worries. This is mainly due to getting through their difficult restart better than expected and the poor performance for other teams around them.
The losers of this re-calculation are Watford, Bournemouth, Aston Villa and Norwich, although Watford should still be safe. Bournemouth were originally predicted to stay up by a whisker but anyone who has seen them lately will know that any model would struggle to justify that.
One other interesting note is how using a points range helps to show the level of uncertainty that inevitably emerges in sport. If Watford perform at the lower end of their range and Aston Villa at the higher end, then relegation could be decided on goal difference, or even the lucky bounce of a ball in injury time of the last game!
It is these uncertainties that make sport, and trying to predict it, so compelling. Rather bizarrely, adding greater uncertainty to the prediction has probably made it a more accurate model.
Sport.dat 