Once again this year I thought I’d have a go at predicting the outcome of the premiership so it’s time to look back and see how accurate they were. My predictions can be found here, and, courtesy of 5addedminutes, I also have the predictions of the EuroClubIndex.
To see how much our knowledge has improved our ability to predict the standings I’m going to compare these predictions to two other ‘dumb’ scenarios:
a) each team scores exactly as many points as they did the previous year
b) each team scores an equal amount of points
So how does each set of predictions compare to how the teams really did?
There’s a few noticeable exceptions. Everyone underestimated the Manchester clubs but that’s kind of understandable – all of these systems are based solely on prior performance and so transfers aren’t taken into account. In the case of United I had them lower because they simply didn’t have a lot of possession in ’10-11, I was genuinely expecting a drop-off. Newcastle were a surprise but I think a lot of the improvement is smoke and mirrors, they’re much closer to Everton in terms of performance than Chelsea. Wolves badly under performed, which they couldn’t afford as they were always likely to be part of the relegation dog-fight.
So how do they projections compare statistically? To determine this I took the magnitude of the difference between a teams predicted points and their actual points and calculated the mean and standard deviation across the league for each system. A perfect system would have a mean of 0 and a standard deviation of 0 (i.e., lower numbers are better)
For the record my numbers last year were mean 5.9, stdev 4.7 so this year seems to have been more unpredictable.
Based on this what proportion of teams would each system have got correct to within 5 or 10 points.
Basically being so far off with regards to Manchester City killed everyone here.
So in conclusion if we had to pick one set of predictions that best described the ’11-12 we’d have been best off if we’d just guessed that they would do exactly the same as last year. I’d already decided that, alongside the current method, I was going to try another way of doing predictions for next year so lets see if there’s an improvement in ’12-13.