Modern European Club Soccer: Partridges in Disparity?

Soccer Elo: The Rebuild, Part 2

Nearly 300 million Euros in one picture, and they got Gareth dirty. #raisedbywolves. Img: Sky Sports

Without the barriers of a salary cap or a luxury tax, club owners are free to make it rain. Rich teams that can afford the best players can usually afford to buy mega-stars at their peak value. If it works, it’s great for business, as it optimizes both play on the pitch and branding value (which sells shirts, ad space on shirts, viewership slots, etc.). More modest teams would hope to arbitrage the marketplace, developing academies of stars that they could sell for a huge payday (See: Gareth Bale, Adam Lallana, Ricky Lambert).

All of this happens, of course, at the cost of devaluing domestic league competitiveness. As borders open and the European soccer becomes the hotspot for the world’s most talented players, consumer demand wants to see the best play the best. If teams are willing to pay (sometimes literally) stupid amounts of money on the best players, then the world in its prevailing wisdom (and surely your American Euro-snob friend who spent a semester speaking English in Madrid) says stuff like this:

Bayern in the Bundesliga? Yawn, they’ll have that sewn up by September. Put them against Atletico Madrid in the Champions League and ja voll! Los geht’s Jungs!!!

Let the dweebs, weirdos, and DC United fans watch Motherwell-Hearts. Give me the Barca-Real Madrid Clasico or the Manchester Derby, it’s soooo much better.

If success in soccer is an allegory for success in life, then all you need is a rich, ethically-questionable sugar daddy to finance without question everything you want to do in this world. #SoccerBaybeez4Lyfe!!! #blessed #withgodgivenabilities #andunearnedoilrents (Pair that one with the next Neymar duck face selfie you see on Instagram.)

Whelp, that was depressing. But is it true?

ELO: the Arbiter of Parity

I introduced ELO as a chess rating, but it’s quite possibly the strongest measure of competitive parity to date. Why? Endogeneity! What does that mean? Let’s recall from the introduction:

• Every team that isn’t newly promoted starts with an initial rating of 1500 points.
• Every game will result in an exchange of equal points.
• In this version of ELO, there are no cross-border matches, which will have a disproportionate effect on the ranking systems.

Therefore, unless the league contracts in size, the average of every league in this study will always stay at 1500. I’m not a fan of excessive bold type-face, but this is important.

Crossroads at Tyneside: Leicester City [1546 and going up] vs Newcastle [1457 and going down], November 2015. Img: Yahoo

There are studies that attempt to explain competitive disparity in terms of inequality of league points differential, famously using the Gini coefficient among other values. These studies cannot take the team identity into account. If, say, a team were to finish 15th one year and 1st the next year (like Leicester City, 2014–2016) or do the opposite (Newcastle United 2011–2016), these studies would not be able to take that into account. Because it calculates every match at the team-level and adjusts accordingly, ELO can.

Now, let’s review the dataset’s parameters:

• Only first division domestic league games were considered. Football-data.co.uk was the main source for data from ten European Leagues, including the “Big Four” (England, Germany, Spain, Italy). For purposes of this study, Inter-European games will not be considered. For an Elo model that attempts to do this (well, but I still have problems with it), check out clubelo.com.
• 1996 was the first season used, as this was the first season in which all leagues instituted three points for a victory rules.
• Initial rating: All teams at the start of the 1996 season receive an initial 1500 rating
• The match importance parameter is a constant at K = 20, per clubelo.com’s formula.
• The margin of victory multiplier is also borrowed from clubelo.com and the When Stats and Soccer Unite blog. This is somewhat different from EloRatings.net’s formula, but is simpler and has very much the declining returns featured in EloRatings.net:

• Home field advantage starts at 100, but changes every day d for every country by a constant percentage of ELO points gained or lost by the home teams in each match n of the day d:

Home field advantage is not rounded off. This happens at the final step of the post-match ELO calculation.

• Adjusted End-of-Season Value: Per FiveThirtyEight conventions, at the end of each season, teams will retain 90% of their end of season ELO value, the remaining 10% will be calculated as a portion of the league mean ELO rating.
• Promoted teams will be awarded an initial rating of the mean of the adjusted scores for all relegated teams. If no teams are relegated (e.g., Scotland’s 2001 season), then promoted teams will be awarded an initial rating of the league mean ELO rating. In no case will a re-promoted team (i.e., Leicester City was relegated in 2002 and promoted in 2014) retain its end-of-season rating from a more previous year; all relegated teams forfeit their ELO ratings to their successors.

Full R Code on the Github.

Put it together, and?

It’s a little messy. First we have to tighten up the dataset to represent only end-of-year ELO values. Let’s stay in England for the moment and have a look at the Premier League’s dispersion of ELO values from 1996–2016:

The green line indicates the 1500 ELO average rating for the EPL each year. Each point represents one of the twenty teams in the league for that season. Looking at the data as a whole, there are two competing trends:

1. One grouping of teams is in the lead group, pulling away more and more from the peloton every year.
2. The peloton (or the grouping of teams tightly packed around the middle, seems to be drifting further and further below the 1500 average ELO line.

This seems to be the disparity everyone is wondering about. Let’s try to summarize it a little better using a box-and-whiskers plot:

Let’s define terms: The box represents the 25th to 75th percentile of teams in the league (6th place to 15th place), with the line in the box represents the 50th percentile (between 10th and 11th place).

The lines (whiskers) extend for 1.5 times that interquartile range if the data point is within 99% of that normal distribution.

The points outside that range are therefore true outliers (~100% significance of being outside normal distribution). Given the shape of the boxplot over time, there are three parts to comparing competitive parity:

1. Size of box: That is, the dispersion of the ELO scores of the middle ten teams in the Premier League. The bigger the box, the greater the disparity (a trend we see especially from 2010 onwards).

2. Placement of midline: That is, how much the median value (50th percentile) deviates from the mean. Let’s look closely at the 2012 season for an example:

City ended up winning on Goal Differential. #Boring.

In this case, four teams (West Brom, Swansea, Wigan, Sunderland) hover around the 1479 median, with four more teams (Norwich, Stoke, Aston Villa, and QPR) not far behind. This pushes the actual 50th percentile down way below the 1500 average, and creates an incidence of a top-heavy league where House Manchester .

3. Outliers: Outliers here are counted as part of the dataset and the average, which seems strange until you consider some of the actual outliers across Europe in 2016:

An Unfortunate Conclusion: What Exactly Is Parity?

Put in this perspective, England seems rather competitive compared to other leagues across Europe. Paris Saint-Germain and Celtic rule Ligue 1 and the Scottish Premier League as supreme dictators; PSV and Ajax top the Eredivisie in much the same way. Of course, one should not discount Leicester City’s improbable run in 2016, bringing English parity to heel after nearly a five year battle between Chelsea and House Manchester, a similar three-horse race currently being run in Spain between Barcelona, Real Madrid, and Atletico Madrid.

But the Leicester City counterexample doesn’t seem like it will last, as Chelsea, Man City, and the North London Horde (Arsenal, Tottenham Hotspur). are currently riding toward an epic battle for the Premier League Iron Throne over the next months.

N’Golo Kante Jumps from First to First: Probably 200 Elo Points in those Boots Alone. Img: The Mirror

Well, that kind of sounds like parity. So long as we discount the 15 other teams that have relatively zero shot at winning the Premier League. So does Belgium, which looks like it has been a relative free-for-all for years now (see the appendix below)… and that sound of surprise you’re hearing in my reading tone is the first time I’ve ever noticed Belgium in UEFA club competition. (Sorry, Belgium)

So what, exactly, is parity in soccer? What would define a league that would be at complete parity with completely unexpected results? And if we had complete parity, would we want it?

Or, to borrow from Tolstoy (will requisite apologies), are all league structures equally alike, but all league structures disparate in their own way?

I suppose that’s a question for another blog post. For now, this post has shown the awesome power of ELO in establishing an initial measure for parity within UEFA’s many league structures. Part 3 will look into simulating results based on the expected result variable and look closer at honing in on the balance between predictable and unpredictable results.

And, since I couldn’t find room for this, here’s a graph for parity across ten European leagues, with the Github link for all graphs here.