FZ Soccer Store: EPL Mid-Season Table and Report

With a little over half of the EPL season complete and nearly half of the January transfer window closed, it's worth taking a look back at how the table has fallen out and identify teams that are over and under performing against various metrics.

Methodology

All data in this post was accurate as of December 28th, 2010 when a majority of the teams in the league had played nineteen or more matches (halfway towards each team's 38 total matches). To account for the different number of matches played by December 28th, each team's goal differential and point total were scaled up according to the number of games they had played to a projected value at season's end. The table was then resorted in descending order, first by points and then by goal differential. This produced a table that showed projected finish based upon performance through December 28th.

There are five metrics used to judge team performance within the table - four of them relate to table position, and the fifth relates to points accumulated. Each metric represents the results of a regression analysis that was performed on table position and that metric , or points and goal differential in the case of the fifth metric. Each metric is color coded red if the team is under performing versus that metric, green if it is over performing, and no color coding if it is performing as expected.

A quick note must be made regarding how over and under performing is determined. A post from last April introduced the concept of a prediction interval (PI) as part of regression analysis. Prediction intervals are communicated as percentages, and are centered on the regression line. Thus, a 50% PI represents the bounds of 50% of the expected finishing positions over time for a given predictor, and is expressed as the maximum and minimum table position (the bounds) in that distribution. Those bounds are the data points that are 25% above and 25% below the original regression line. Prediction intervals can be calculated along the entire range of predictors (MSq£, M£XI, etc.) to generate bounds around the regression line that communicates the result (table position, points).

In this analysis the 50% PI is used to determine under and over performance to each metric. Data that falls within the bounds of the 50% PI is considered expected noise around the regression line and equation - teams are expected to fall within these bounds given their performance versus the metric. Teams that fall outside of the 50% PI, high or low, are over or under performing. The direction of the relationship between the two variables determines whether falling above or below indicates over or under performance - positively correlated variables (points vs. GD) require falling above the 50% PI for over performance, while negatively correlated variables (table position vs. all variables) require falling below the 50% PI. The reverse conditions are true for under performance.

The metrics used within the table are:

MSq£: This metric is defined as multiple of the league average squad cost as defined by the Transfer Price Index . The concept of this metric and its relationship to table position has been covered extensively here .
M£XI: This metric is similar to MSq£, but it measures the multiple of league average starting XI cost. A comparison to MSq£ is made here, and a full discussion of its relationship to table position will be explored in future blog posts. All data for this metric has been compiled on a match-by-match basis by Graeme Riley, to whom I am deeply indebted.
Pts: Using the entire 18 season history of the EPL, a regression equation to predict table position based upon projected season points has been developed.
GD: Like points, EPL historical data is used to create a regression equation to describe the relationship between table position and goal differential.
Pts vs. GD: The fifth metric is used to judge the projected number of points a team will acquire versus their projected goal differential for the season based upon eighteen seasons of EPL data.

The performance to date of each team can now be judged in a statistical context against each of these five metrics.

The Mid-Season Table and Team Performance

The table below represents where each team would finish if they held their from they had demonstrated through December 28th (click to enlarge).

A number of observations can be made:

The table clearly demonstrates the superior performance of Manchester United versus the rest of the league, and the real, large gap between Manchester City and Tottenham Hotspur at the mid-point of the season.
While Manchester United has a health six point lead on Arsenal on December 28th, they are on track to have the fifth lowest point total of a champion in the 19 year history of the EPL. Indeed, this is confirmed statistically via their over performance in the third metric, table position versus points.
While Manchester City and Chelsea have nearly equal MSq£ numbers, Chelsea has a far higher M£XI number. In fact, Chelsea has the biggest jump from their predicted finish based upon squad cost to the predicted finish based upon starting XI. Their table position relative to their starting XI cost demonstrates just how much they are under performing.
While one might argue both Manchester teams are unfairly penalized by the regression model's over prediction of the importance of transfer costs,
Chelsea and Aston Villa are clearly under performing relative to their costs of their transfers. In the case of Chelsea, they're likely saddled with the huge costs of aging players that aren't delivering like they used to. Aston Villa must look to make some quick moves in the transfer market and get better performance out of their squad cost that should have them somewhere between 7th and 13th in the team.
Arsenal and Sunderland are the only teams with a significant backslide in predicted finish when moving from the total squad cost to the average starting XI cost.
Other than the three teams at the bottom of the table, Chelsea under performs the most by failing to meet expectations in four out of five categories.
While Liverpool were in ninth position at the mid-season point, they sank to twelfth in the table by Week 22 which clearly indicated underachievement versus the MSq£ and M£XI metrics. Combined with their underachievement against their historical norm, it was only a matter of time before Hodgson got sacked.
Bolton, Everton, Newcastle United, and Birmingham City are the only teams under performing in projected points compared to their projected goal differential. This suggests that either their GD will pick up in the second half if they're to maintain point form, or they will drop more points in the second half s they regress to a point total more fitting their GD.
Aston Villa's long-term drop from form this season is confirmed by the fact that they're massively under performing in the squad and starting XI metrics, are right where they should be based upon points, and are under performing based upon both goal differential metrics.