Tuesday, 29 May 2012

Predicting EPL Finishing Positions.

Sports statistics are able to tell you what occurred on the field of play at a particular point in a game.However,they vary in their usefulness because we aren't always able to say with any degree of certainty how the individual events contributed towards the final outcome of the contest or if they are repeatable skills that are likely to contribute to future success or failure.The last caveat is the most important for those who try to predict future events from past results.

Manchester City won the Premiership in 2011/12 partly because they equaled or outscored their opponents in  all but five of their 38 games and produced the kind of high class performances you would expect from such an expensively assembled squad.Their scoring and conceding record not only tell us how they won their first Championship for 44 years,but given that it was spread over an entire season,we would expect City to produce a very similar level of performance in 2012/13.The upper regions of the EPL changes very little over time and with their vast wealth Manchester City will be title contenders for the foreseeable future.So we shouldn't be surprised if the record of the Champions of one year is a very good predictor of their record in subsequent seasons.

The most basic measure of a team's performance over a season is their final points total and City's rivals United have recorded final points totals of 83,89,87,90,85,80 and 89 in their most recent seven campaigns and their previous raw,uncorrected for the talent spread of the league,totals are good indicators of what is likely to follow from season to season.A similar level of consistency exists lower down the pecking order where Stoke recorded 45 points in their first year of Premiership football and followed up by amassing 47,46 and 45 points in their three subsequent years.

If we plot the amount of points EPL teams get in one season against the total they achieve in the next we find that the previous two examples are far from unique and there is very strong visual correlation from year to year.R^2 is 0.57 for all Premiership teams since 1993.

For such a simple statistics,the correlation is extremely good,the only correction made to the totals was to allow for the occasional 42 game season earlier on in the sample run.However,improvement can be made.Total league points gained by teams over a season are a function of their win/draw record and it has been long recognised in sports such as baseball and latterly basketball and American Football that points/goals/runs differential and particularly points/goals/runs scored and conceded can reveal more about the true ability of a side than can it's raw win loss record.

The ultimate aim of all team sports is to put goals,runs or points on the board,but while more talented sides are likely to be better at scoring themselves and preventing their opponents from scoring,they may find that the goals arrive in inconveniently,unproductive sequences.Therefore teams may have identical goal differences or even identical goals scored and allowed totals,but may have slightly different win/draw records and thus different points totals.One team may win a sizeable number of very close games,whereas a similar team may see series a tight matches end in equal wins and losses.Although their respective win loss records my be different,their goals scored and allowed records may be identical and in these circumstances,it's common to assume that the "better" team is merely the luckier one.

So evaluating teams based on goals rather than league points accrued through their win loss draw records may highlight lucky teams and therefore be more predictive over subsequent years.Adapting this approach for football requires issues such as treating draws as half a win and accounting for the best and the worst teams tending to play in a slightly higher goal environment that the rest of the league.

Goals based league points projections are compared to a team's actual points total in the table for the 2011/12 EPL below.It's worth noting that Manchester City were meaner in defence and also outscored United,so they possibly should have secured the Championship slightly earlier than the 94th minute of the ultimate game.Based on goals,they were probably just over one win superior to United.

Team. Goals Scored. Goals Allowed. Actual
Goals Based Projected Points.
Manchester C. 93 29 89 88
Manchester U. 89 33 89 84
Arsenal. 74 49 70 67
Tottenham. 66 41 69 68
Newcastle. 56 51 65 55
Chelsea. 65 46 64 64
Everton. 50 40 56 59
Liverpool. 47 40 52 57
Fulham. 48 51 52 50
WBA. 45 52 47 47
Swansea. 44 51 47 47
Norwich. 52 66 47 43
Sunderland. 45 46 45 51
Stoke. 36 53 45 40
Wigan. 42 62 43 39
Aston Villa. 37 53 38 41
QPR. 43 66 37 37
Bolton. 46 77 36 33
Blackburn. 48 78 31 34
Wolves. 40 82 25 27

Both the strengths and the weakness of using seasonal goal totals to evaluate the amount of points a team "should"  have acquired over a season is highlighted in the case of Newcastle.Their goal difference of +5 is certainly atypical of the teams in the immediate neighbourhood,it's more indicative of a mid table finish rather than top six.Eight of their league wins were by a single goal margin and if a couple of those wins were turned into draws their points total would begin to fade badly because of the three points for a win system.Spurs,who finished just above Newcastle,scored 10 more goals than the Tynesiders and conceded 10 goals less.They had a much more conventional top six goals profile and they won just 5 games by a margin of a single goal.So Spurs would seem to be a more solid prospect to maintain their form next season compared to Newcastle.

However,Chelsea who finished just below Newcastle again with a conventional top six goals profile had one more single goal victory than did Newcastle.Clearly this approach is capturing other factors as well as simply highlighting streaks of narrow victories that may not be repeated in later years.

Wigan in 2009/10 finished with 36 points.Based on their goals record of 37 scored and 79 allowed they would have expected to have been comfortably relegated with 26 points,proportionally the biggest difference between projected and actual values in the last 20 years of EPL football.However,contained within their record was an 8-0 defeat at Chelsea following an early red card and a 9-1 reverse at Spurs with a full complement of players.The Latics had the good sense to concede 17 of their goals in just two games and for the rest of the season they played and accrued points befitting a competitive bottom six side rather than the relegation certs that their final goals record predicted.The following season Wigan again performed like a "good" bottom six side and they stayed up again with 42 points.In Wigan's case their actual points were more predictive than their record derived from goals scored and conceded.Newcastle's record this season didn't contain anything as extreme as Wigan's two catastrophic defeats,but they did ship five at both Fulham and Spurs and four at Wigan.

In short,rare and extremely large defeats can be shared out over the seasonal record making teams appear much worse than their raw win/draw record.Leading to the latter measure becoming the more indicative measure of future performance.

Having highlighted the issues that exist when using goals scored or allowed to get a projected league points total that attempts to account for an excess of fortunate,narrow victories,the final step is to see if overall these projections give a better prediction for subsequent years compared to using actual points from the previous year.

The plot pairs the goals based points projection with each team's actual points total from the next season for all years since 1993.Visually the fit is similar to the previous plot,but the r^2 is slightly higher at 0.61 indicating a better correlation.Although flaws exist,particularly when teams suffer large margin defeats,either because of red cards or simply by throwing in the towel,it appears that extra information can be captured by investigating a team's seasonal goals record.Newcastle are unlikely to repeat this year's points total of 65,but are equally unlikely to fall to levels close to the 55 points predicted by their goals based projection.Prior to any gain or loss of talent in the transfer window and most likely projection is 7th or 8th with a 2012/13 points total in the mid to high 50's.

No comments:

Post a Comment