A Statistical Premier League Preview: Part One (Retrospective)

I've nearly missed We Ain't Got No History's first birthday! Happy birthday, WAG. You've grown up so much, although I can't say that the main author has gotten any less boring in the past year. Anyway, what better way to celebrate the site's first birthday than with a season preview series where we get to show off some of the statistical work that we pretend to do when we're not covering various transfer rumours.

And, of course, before we should do that we should take a look at our previous year's predictions and see how we did. Let's take a poke through and see how well last year's season preview jived up with the actual totals...

Introduction

Last season I ran a series of studies to determine the relationship between previous performance and future goals scored and conceded, as well as taking into account transfer fees and league jumps. Then I worked out how to convert that all into points. The process is laid out in a series of five posts, and the result had Chelsea winning the league quite comfortably. Clearly, that was wrong, but it'll be instructive to see what the model did right, and where else it missed.

Manchester United

Actual: 80 pts, 78 scored, 37 conceded; 1st.

Predicted: 81 pts (-1), 79 scored (-1), 33 conceded (-4); 2nd (+1).

This is almost perfect: goal differential within three; points total within one. The model did an excellent job here.

Chelsea

Actual: 71 pts, 69 scored, 33 conceded; 2nd.

Predicted: 85 pts (-14), 91 scored (-22), 37 conceded (+4); 1st (-1).

Not so much with Chelsea. There was no bias - I just followed the math, but I suspect it was slightly delubed by the 103 goal season. That goals scored differential of 22 is the biggest in the whole set.

Manchester City

Actual: 71 pts, 60 scored, 33 conceded; 3rd.

Predicted: 76 pts (-5), 80 scored (-20), 42 conceded (+9); 3rd (-).

Manchester City were projected to do well because they sent a lot of money. The model missed because of the way Mancini deployed his players, and was fairly close on goal differential. And it got the position correct, which is good.

Arsenal

Actual: 68 pts, 72 scored, 43 conceded; 4th.

Predicted: 71 pts (-3), 73 scored (-1), 43 conceded (-); 4th (-).

Nailed it.

Tottenham Hotspur

Actual: 62 pts, 55 scored, 46 conceded; 5th.

Predicted: 63 pts (-1), 64 scored (-9), 46 conceded (-); 5th (-).

Came very close, although Spurs got quite lucky with the timing of their goals - the differential they ran doesn't often lead to 62 points.

Liverpool

Actual: 58 pts, 59 scored, 44 conceded; 6th.

Predicted: 61 pts (-3), 56 scored (+3), 42 conceded (-2); 6th(-).

Despite all the weird Liverpooly things, I got this more or less right as well.

Everton

Actual: 54 pts, 51 scored, 45 conceded; 7th.

Predicted: 55 pts (-1), 56 scored (-5), 51 conceded (+6); 8th (+1).

The model thought Everton were going to be exactly the same team only better at scoring and worse at defending. I'm not really sure why.

Fulham

Actual: 49 pts, 49 scored, 43 conceded; 8th.

Predicted: 49 pts (-), 42 scored (+7), 48 conceded (+5); 9th (+1).

Despite getting the points totals right, model underestimated Mark Hughes' Fulham.

Aston Villa

Actual: 48 pts, 48 scored, 59 conceded; 9th.

Predicted: 56 pts (-8), 53 scored (-5), 47 conceded (-12); 7th (-2).

A bad miss here. Every facet of Villa's play fell off, possibly due to the managerial turbulence that engulfed the squad. Of course, this sort of thing does happen.

Sunderland

Actual: 47 pts, 45 scored, 56 conceded; 10th.

Predicted: 44 pts (+3), 47 scored (-2), 61 conceded (+5); 14th (+4).

Despite the difference in the finish places, the projection did a decent job on Sunderland, I think. The defence was better than expected, and the forwards were slightly worse. All in all that's worth a decent little jump.

West Bromwich Albion

Actual: 47 pts, 56 scored, 71 conceded; 11th.

Predicted: 41 pts (+6), 55 scored (+1), 74 conceded (+3); 16th (+5).

West Brom were a little better than the model had them pegged for, but also got very lucky. Based on our goal differential calculation, they 'should' have been just out of the relegation zone.

Newcastle United

Actual: 46 pts, 56 scored, 57 conceded; 12th.

Predicted: 48 pts (-2), 57 scored (-1), 64 conceded (7); 10th (-2).

Not terrible, but a slight under-performance in all fields.

Stoke City

Actual: 46 pts, 46 scored, 48 conceded; 13th.

Predicted: 44 pts (+2), 37 scored (+9), 50 conceded (+2); 13th (-).

Stoke were probably the best team in the bottom half of the table. The projection isn't particularly good despite coming close on points and nailing them in 13th - they scored far more goals than the model was expecting, and got unlucky.

Bolton Wanderers

Actual: 46 pts, 52 scored, 56 conceded; 14th.

Predicted: 39 pts (+7), 42 scored (+10), 64 conceded (+8); 18th (+4).

I can't remember what Bolton did in 2009/10, but the model thought they were terrible, and continued to think they'd be terrible. Huge miss here.

Blackburn Rovers

Actual: 43 pts, 46 scored, 59 conceded; 15th.

Predicted: 45 pts (-2), 43 scored (+3), 54 conceded (-5); 11th (-4).

Slight under-performance means huge drop in position. I don't think the model missed by much here.

Wigan Athletic

Actual: 42 pts, 40 scored, 61 conceded; 16th.

Predicted: 31 pts (+11), 39 scored (+1), 73 conceded (+12); 19th (+3).

Roberto Martinez fixed the defence for the 2010/11 season. I don't know how he did it, but that was Wigan's major surprise.

Wolverhampton Wanderers

Actual: 40 pts, 46 scored, 66 conceded; 17th.

Predicted: 39 pts (+1), 34 scored (+12), 56 conceded (-10); 17th (-).

Much better at scoring than expected, far worse at defending. I wonder if Mick McCarthy was feeling ok.

Birmingham City

Actual: 39 pts, 37 scored, 58 conceded; 18th.

Predicted: 45 pts (-6), 39 scored (-2), 51 conceded (-7); 12th (-6).

Just nine goals worse than expected meant relegation. I don't think I was too far off here, but the defensive downgrades hurt.

Blackpool

Actual: 39 pts, 55 scored, 78 conceded; 19th.

Predicted; 29 pts (+10), 44 scored (+11), 81 conceded (+3); 20th (+1).

Much better than anticipated; still quite bad. Shame. Big miss in terms of goals scored.

West Ham United

Actual: 33 pts, 43 scored, 70 conceded; 20th

Predicted: 42 pts (-9), 49 scored (-6), 63 conceded (-7); 15th (-5).

Awful. Considering the players they had, my projection still seems more sensible than the results.

General Thoughts

Obviously, with any projection, there'll be issues - sports is inherently probabilistic and we're going to be off from time to time no matter what. That said, there can still be flaws in a system. Here, we seemed to run into major problems at the top of the goalscoring charts, with Chelsea and Manchester City hugely off in terms of goals scored. There's no really obvious way of fixing this, so I think we'll keep things as they are for this season and see how it works out.

Let's make a modern version next.