Long term readers will know that 11tegen11 started out, back in 2010, as a pure tactics oriented blog, but since then, slowly the analytics part has crept in. Pure tactical analysis has become a rare commodity here, after analytics took over. The main reason behind this, and I can safely say that now, is that tactical analytics is all hindsight bias.

Early this year, Richard Whitall put it very nicely in one of his State of Analytics columns…

This is not to say that this kind of subjective interpretation of tactical trends, strengths and weaknesses is without value, but I do think it is subject to abuse. For example, it’s too often the case that some writers will imply a strong causal link between a certain, game-specific formation and a final outcome or set of outcomes.

These two sentences kind of bring together why I gave up writing tactical match reports. I missed the evidence to make my statements and basically, explaining tactics in the context of highly luck-driven outcomes felt like the abuse that I just quoted.

So, let’s move on and hope that analytics and tactics will soon merge. I firmly believe that trend has recently started, and it won’t stop soon. Analysts are nothing without tactical content, just as tactics are empty without empirical evidence to back it up.

This post was intended to tackle the issue of finishing quality, so let’s continue and do a little thought experiment. Try and answer this simple question…

What does it take to score a goal?

Simple, right? Creating a shooting opportunity and finding the back of the net. Correct.

In the previous posts, we’ve focused on the first step: creating shooting opportunities. Teams differ in two respects here: better teams create more shooting opportunities and they create better shooting opportunities. Both the number of shots created and the amount of ‘Expected Goals per Shot Created’ nicely correlate with the final league table.

But what’s up with step two? You may create less shooting opportunities, or shots with a lower ‘Expected Goals Scored’ number attached to it, as long as you make up for it by finishing more chances, you’ll be fine. Today’s post will identify Finishing Quality and come up with a simple parameter to judge teams or players by.

Remember that we’ve recently established a number for Expected Goals for each team. The number of Expected Goals is quite simple. Categorize each shot by strike zone and game state and look up the league average conversion rate for that shot. Add the total for all shots, and here we are, a total number of Expected Goals Scored.

We’re now just one step away from establishing Finishing Quality and that requires a comparison with the actual number of goals scored. Score more than the average Eredivisie team does from your shots (same number, same strike zone, same game state) and you’re an above average team when it comes to Finishing Quality. The next graph ranks all teams by Finishing Quality, defined as the number of Actual Goals Scored divided by the number of Expected Goals Scored.

Ajax and PSV, unsurprisingly, come up as the best finishing teams in the league. Vitesse and Roda complete a quite distanced top-4. Heracles is somewhere behind them in 5th place, but certainly higher than their league ranking would suggest. At the back end, Groningen’s problem is uncovered without shame. Perhaps also surprising, Feyenoord, with a FQ of just 0.84.

We’ll take this one step further and provide split numbers per team per zone. In order to keep the number of graphs within limits, I’ve created this Tableau graph (scroll down) where you can flip through all of the teams and see their performances for yourself. I’ll go over all teams in brief, as I believe there are some very interesting numbers.

ADO
Overall, ADO has slightly overachieved, with an FQ of 1.05. This has led to two more goals than expected, with the best number from Zone 2. Small differences between Expected and Actual Goals though, so hardly any different from the average Eredivisie team.

Ajax
With an FQ of 1.25 Ajax scores a quarter more than the average team would do from their shots. We can see this mainly comes down to long and medium distance shooting. Ajax’ FQ from Zone 3 (20/13 or 1.67) and Zone 4 (13/7 or 1.86) is downright impressive and indicates that they’re doing things quite right from distance.

AZ
Overall, with an FQ of 1.02 AZ are an average team when it comes to finishing. It’s just from distance (Zone 4) that they seems to overachieve a bit.

Feyenoord
Now here’s an interesting one. Third ranked Feyenoord proves to be one of the worst finishing teams of the Eredivisie, who’d have thought. The previous post had identified them as creating the best chances in the Eredivisie, but their FQ comes in at just 0.84. We can see that this problem arises in all four zones, but scoring only one goal from Zone 4, where five were expected is quite poor. They took a total of 159 shots from Zone 4, which deserves a more detailed examination in a later piece.

Groningen
The worst team when it comes to Finishing. Regular followers of the Eredivisie will probably know that already, but Groningen’s strikers really can find the back of the net. Their long distance performance is poor, but it’s Zone 2 that really catches the eye. Nine goals behind an expected tally of 32, that’s quite the difference between a firm top-6 spot and mid-table.

Heerenveen
Despite having a top striker in Finnbogason, Heerenveen scores below par with an overall FQ of 0.87. Their deficiencies are mainly all over the pitch, as they underscore in each zone apart from the true tap-ins of Zone 1.

Heracles
Here’s another remarkable chart. Heracles ranks 5th overall with an FQ of 1.12, and it’s mainly because of their performance in Zone 3. That’s the area just outside the box, or slightly off-central within the box. Another lead for further investigation is born.

N.E.C.
I can’t keep saying they’re all interesting, right? Overall, N.E.C. is poor at an FQ or 0.84, but their long distance efforts overachieve, while their performance from Zone 2 is dreadful. Saved for later.

NAC
Overall around average at an FQ of 1.05. The spread looks a bit like N.E.C., with an overachievement from distance and underachievement closer by. This time, though, the differences are small, and may be not even significant.

PEC Zwolle
With an overall FQ of 0.93, PEC Zwolle don’t surprise, but they graph point out their quality was mainly in Zone 2, where this newly promoted team outperformed the average Eredivisie level.

PSV
The top scorers of the Eredivisie with 99 non penalty goals have the second highest FQ, at 1.24, just behind Ajax. But in contrast to their title rivals, PSV overachieved from every zone.

RKC
Overall, RKC came in okay, at 0.97. The graph shows it’s their Zone 2 performance that did the trick, while the medium distance shots were the problem area.

Roda
The sign of a team with an excellent striker. Massive overachievement in Zone 2, while the rest is at par. Will be studied on player level, if only to satisfy Sanharib Malki’s fans.

Twente
With an FQ of 0.92, Twente underperformed. Mainly from distance it seems, although a team that held title ambitions prior to the season start should have a better close range strike force than this sub-par Eredivisie level.

Utrecht
Overachiever in the table with their 5th spot, Utrecht did not do so on the basis of Finishing Quality. At just 0.93 overall, they were on par from Zone 2, but disappointed from further out.

Vitesse
Much like the pattern at Roda, Vitesse overachieved (1.22) and mainly did so from Zone 2 and 3. Wilfried Bony, anyone? Stay tuned.

VVV
At 0.84 an unremarkable overall FQ, and the spread across the pitch is quite even.

Willem II
Their FQ is level with VVV at 0.84. The problem has not been to score from distance, but the close range has let them down.

It’s been two years since I wrote these words in an article named ‘A chance is a chance is a chance?’. Unfortunately, breaking down chances into expected goals, rather than simply counting shots has not made it to the Eredivisie, or any other league, yet. But times are about to change…

Strike Zones and Game States

Using our recent explorations on strike zones and game states, we can stratify shots according to location and match situation and come up with expected goals per shot. This is much more valuable than simply adding shot numbers, as it removes the basic – and incorrect – assumption that all shots are of equal value.

Shot location may be the most influential factor when it comes to shot quality, as we’ve learned from the days when StatDNA still posted quality analysis pieces on their blog, but location isn’t the only factor involved. The most difficult – and therefore often unmentioned – factor is defensive positioning, or defensive pressure. Measuring this in detail would require GPS tracking of all players on the pitch, which I’m sure is done behind closed doors at present, but it generates huge amounts of data, which complicates the analysis a lot. And more importantly, data at such a level of detail is not widely available yet.

We’ll have to do with what we’ve got, and game states serve as a nice proxy for defensive pressure, as we’ve seen that teams trailing by a single goal give up significantly better chances than teams defending a single goal lead, a gap that measures up to around 25%.

 
Expected Goals

The challenge for this post is now to convert all our recent explorations of strike zones and game states into something handy and simple. We need a single number to indicate the quality of shots that teams create and concede, or at player level, a number that indicates the quality of shots that a player took. Simply said, we should know how many goals the average Eredivisie player would have scored from the attempts that a team, or a player, has had. We shall term this ‘Expected Goals Scored’.

Actually, it is a very simple concept. Let’s take Ajax and examine their shots in detail. In total, Ajax created 544 shots, of which 2 penalties are excluded. Here’s a table of Ajax’ 542 remaining shots created per zone and game state.

	GS -2	GS -1	GS 0	GS +1	GS +2
Zone 1	0	0	2	0	1
Zone 2	3	22	82	41	44
Zone 3	3	19	64	29	46
Zone 4	2	19	97	29	39

Our previous explorations have shown how many goals are scored per shot for each combination of strike zone and game state. We can now easily compute the expected amount of goals for Ajax’ 542 shots by multiplying both tables.

For example, from Strike Zone 2 at GS 0, Ajax took 82 shots. The league average conversion rate for shots from Strike Zone 2 at GS 0 is 0.190. Therefore, the total Expected Goals Scored for Ajax from Strike Zone 2 at GS 0 is 82 * 0.190 = 15.59.

We can repeat this exercise for all combinations of Strike Zones and Game States and add all the subtotals. This will show that Ajax had 65.35 Expected Goals Scored with their 542 shots. In other words, the average Eredivisie team would have scored 65.35 goals from Ajax’ shots, if we correct for Strike Zone and Game State. Only one small step to go, divide the Expected Goals Scored by the number of shots, and now we know the quality of shots that Ajax created: 65.35 / 542 = 0.121 Expected Goals Scored per shot.

Quality of Shots Created

We can repeat the trick for each team to come up with the following graph. The bars represent the quality of the shots that teams created.

There is a considerable spread in quality of shots created. PSV and Feyenoord may expect 0.129 and 0.128 goals per shot, while Willem II creates chances that result in only 0.101 goals per shot. In other words, the type of shots that PSV and Feyenoord create are generally worth 27% more than shots by Willem II. PSV and Feyenoord are followed by the teams that also complete the top-6 in the final league standing, and Roda. In general, the quality of shots created nicely correlates with the league table, with Roda being the big exception. Roda finished 16^th in the table, but comes up 4^th in terms of the quality of shots created.  

Quality of Shots Conceded

We can do the same thing for shots conceded, and measure the quality of shots conceded. This time, of course, lower bars indicate less shots per goal conceded, as an indicator of quality defending.

Again, there is a considerable spread when we compare the best team, Groningen, with the worst team, Heerenveen. Groningen earns their top spot in this chart by doing an excellent job in forcing their opponents to shoot from low quality positions (Zone 4), as we’ve seen previously. In contrast to the Offensive Shot Quality, there is no clear correlation between Defensive Shot Quality and the final league positions. It seems that quality of shots created is a better way than quality of shots conceded to tell good and bad teams apart.

In the end

It’s always a good thing if analysis and observation start to overlap, and with more detailed information to work with, we’re slowly getting there. The mini-series of posts this past week has now lead to a simple parameter called Expected Goals, which we can either express per shot, over a match, or over a series of matches. It has an offensive and a defensive side and the former can be applied to teams and players, while the latter is limited to team level, since shots conceded can’t be linked to single defenders.

Next up will be a series where we will compare the outcome in terms of goals scored and conceded with the Expected Goals scored and conceded. The Expected Goals parameter estimates the chance of the shots, while the difference with the actual outcome is an indicator of Finishing Quality, or its defensive equivalent.

Strike zones

Last week’s post has identified four different strike zones on a football pitch. These zones have been chosen so that Zone 1 identifies near-certain goals: shots with conversion rates of over 80%. Shots from Zone 1 are rare (< 1%), but nevertheless have a huge impact. Zones 2 to 4 all contain over 25% of shots, but with sharply diminishing success rates. Shots from Zone 2 are considered high quality shots, and roughly 1 of 4 shots results in a goal. From Zone 3 this number is 1 of 13 and from Zone 4 just 1 in 30. For convenience, we could say that shots from Zone 1 are worth three shots from Zone 2, which are worth three shots from Zone 3, which in turn are worth three shots from Zone 4.

Using these zones allows us to study the basic fallacy of our beloved shot metric, Total Shots Ratio (TSR). Remember, TSR is a simple, but very successful metric, that uses the fraction of shots that teams take in their matches. TSR has shown to correlate very well with the amount of points that teams obtain and it has proven its usefulness in identifying teams that experience lucky or unlucky streaks.

All shots are equal, or not?

The main concern with TSR, however, is that all shots are treated equal. As simple as the TSR model is, shots from 40 yards out are counted the same as one yard tap-ins. Over the long run, most of the differences between shots will even out, but only if we assume that shot quality is evenly balanced out between teams. But is that really the case? Don’t certain teams create better chances than others? If so, we could better understand why certain teams did not live up to the expectations by the TSR model, while others have seemingly over performed.

The build-up of this post will be to look at all Eredivisie teams, and identify where they took most of their shots. Let’s get started with the raw numbers. I’ve excluded penalties for this analysis.

	Zone 1	Zone 2	Zone 3	Zone 4
ADO	2	118	101	159
Ajax	3	192	161	186
AZ	4	161	126	160
Feyenoord	6	206	152	159
Groningen	2	157	93	174
Heerenveen	3	154	148	144
Heracles	3	141	138	176
NAC	3	118	82	131
N.E.C.	1	171	118	164
PEC Zwolle	1	133	99	165
PSV	8	225	172	213
RKC	4	100	122	135
Roda	4	137	89	110
Twente	4	181	134	152
Utrecht	3	178	118	153
Vitesse	2	172	109	146
VVV	2	120	112	147
Willem II	2	122	87	152

Thanks to my friend Benjamin Pugsley, this table format has now been improved a lot, and you can even sort the columns yourself! But from an analytics view, this data is not really accessible, is it? Let’s go through a few numbers in detail and see what we can work out from there… PSV have dominated the shots department and have created the highest number of shots in each of the three zones. Other outliers are a remarkable high number of shots from Zone 2 by Feyenoord – that’s a good thing – and a remarkable low number of shots from that zone by RKC – that’s a bad thing. Roda tend to shy away from taking long shots, with only 110 attempts from Zone 4, but they tend to shy away from creating shots from any zone, so, alas, not necessarily a good thing for them.  

Relativity

Clearly, this table needs relativity. Teams like Roda may do well by avoiding futile shots (Zone 4) at goal, but they hardly create any quality attempts (Zone 2) either.

This is basically the same information, but presented in a slightly different way. PSV stood out in the previous table, because they had the highest amount of shots in each zone, but the spread across the zones is not remarkable. RKC stand out as a team with big trouble creating quality shots (Zone 2), while Feyenoord, Vitesse, Utrecht and Twente succeeded in creating a high proportion of shots from Zone 2.

Willem II, PEC Zwolle, ADO and Groningen take over 40% of shots from Zone 4, where teams on average need around 30 shots for a goal, while Feyenoord, Heerenveen, Roda and Twente limit themselves to less than 33% of shots from Zone 4.

Defense

We can repeat this trick for shots conceded, to try and identify teams that give up high quality chances and teams that give up mainly low quality chances. Here’s a similar table to the one above, but this time for the defensive side of things.

Painful news for Heerenveen, who give up a record high 45% of shots from Zone 2, followed at some distance by ADO and VVV at 41% and RKC at 40%. Better numbers in that regard for AZ and Groningen, who lead a bunch of teams with just over 30% of shots from Zone 2.
Remeber, you can sort the table yourself by clicking on the header.

Twente and Groningen perform well in terms of forcing their opponents into low quality chances (Zone 4), while ADO, Heracles, Heerenveen and VVV could do a lot better in that regard.

In the end

Lots of numbers, but don’t let that scare you away. We’ll wrap the concept of shot quality differences up in a later post, where we’ll fuse a lot of this concepts to make things simpler.

Using shot locations and league wide conversion from our well defined zones, we can see which teams did well in terms of creating shots from quality positions, while preventing their opponents from doing so. At the very least, it seems there are relevant differences between teams in terms of shot quality, which in itself is bad news for models based on TSR. And for the nearby future, this type of detailed analysis can help us to slowly build towards a better synthesis of analysis and tactics…

Four zones

Yesterday, we’ve seen that we can reasonably split a football pitch into four zones, conveniently named Zone 1 to 4, with each higher numbered zone cutting the conversion chances about a third. For clarity, here’s the diagram of the zones again.

Shots from Zone 1 are rare (< 1%), but they do deserve their own category because of their extremely high conversion rate (>80%). I will leave Zone 1 out of the next table, since the low numbers only make things messy without adding value.

Conversion

The next table focuses on conversion rates at different Game States in Zones 2 to 4. Note that Game States represents score differential, but Game States -2 and +2 contain all teams chasing or leading two or more goals.

As we already know, overall conversion drops sharply as we move up the Zones. However, the most interesting finding is that conversion within a zone is strongly linked to Game State. This means that using shot location is important, but independently, one should also factor in Game State. So, high quality shots are even more high quality if a team is leading a match.

Obviously, shot quality is not directly influenced by the score board, but we consider it a proxy for defensive positioning. As teams chase a goal, they give up defensive effort to gain more offense. Subsequently, as we now find out, their opponents can generate higher quality shots, independent of location.

Leading and chasing

So leading teams fire in better shots, but do Game States also influence the amount of shots that teams take from different zones? Do leading teams fire in more close range shots?

To answer that question, we need to study shots, zones and Game States in a slightly different way. The next table shows the relative amount of shots that teams take from Zones 2 to 4 given their particular Game State. Remember, Zone 1 is left out, so numbers in the columns don’t exactly add up to 100%.

Overall, teams take most shots either from Zone 2, which is mainly the central penalty box area, where conversion lies around 22%, or from Zone 4, which is the near hopeless area of 3.4% conversion. But Game States do influence where teams take their shots from.

As expected, teams that defend a lead, either by one goal or by two goals or more, decrease the frequency of low quality shots (Zone 4) to below 30%. Also, leading teams have the highest proportion of high quality shots (Zone 2, > 38%).

Interestingly, the highest proportion of low quality shots (Zone 4) is not fired in by teams chasing a lead, but at an even Game State. This links in well with the earlier observation that conversion rates dip significantly at this Game State. It is true, however, that teams chasing a single goal (GS -1) also fire in a high amount of low quality shots (Zone 4, > 37%).

Conclusion

If we link these two tables together, we can learn that leading teams take more shots from good positions (Zone 2) and less from hopeless positions (Zone 4), but their conversion rate from each zone is also significantly higher. So, it works two ways when teams chase leads: they sacrifice defense by giving up more Zone 2 shots that also stand a higher chance of finding the net.

To kick off the summer period, this piece will look at shots and identify four different strike zones on the pitch, where vastly different types of shots are being produced. This may not be all that shocking, as most of it connects with common sense of watching football games, but at the very least, it will provide a nice base for future explorations.

Basically, we could take two different approaches to identify strike zones. Either top-down, by adhering to known areas on the pitch, like the penalty box, the 6-yard box, etc. This obviously has the advantage of easy communication. But the disadvantage would be that the discriminatory power would be less, as shot quality may differ quite a bit within these classical zones.

The alternative, and the method I will use, is a bottom-up approach by using a fine grid to identify where on the pitch the biggest drop of in shot quality occurs, and then work out the best zones from there.

Here’s an Excel-produced half football pitch, split into small grids, with the conversion being printed in each grid and color marks indicating zones of high and low conversion. Conversion, in this case, is goals per shot and data covers the past two Eredivisie seasons. It’s not intended to be fully readable, but the basic point is that it allows the identification of four different strike zones.

The continuous line marks the borders of the pitch, and for reference, the penalty box. The dotted lines mark zones, where you can see a clear demarcation in shot conversion. We will use these boxes to identify four different strike zones, each with unique characteristics, as we’ll find out later.

Unsurprisingly, right in front of the goal, nearly all shots are converted into goals. I will therefore refer to this zone as Zone 1. The next most threatening part, Zone 2, covers mostly the central penalty box area, but stretches just beyond the edge of the penalty box. Then comes Zone 3, covering the wider penalty box area as well as longer distance central shots. The rest of the pitch will be Zone 4.

This table shows the mean conversion rates per zone, indicating that each zone has a distinctly different conversion rate, going a factor three down as we progress down each zone. Shots from Zone 1 are rare, but due to their extremely high conversion they deserve their own group. All other zones contain a reasonably high number of shots, indicating a good distribution of shots among these zones.

These strike zones will prove a useful reference point to assess difference between teams in terms of strategy, difference between players in terms of quality and will help to explore our beautiful game more in depth.

The fan would probably not understand the very fact that you’re asking this, particularly at this time of the year with the Eredivisie that just finished. Obviously, it would be Ajax, duh. They just won the league for the third time in a row, didn’t they?

The manager would maybe find a bit more nuance and say that this season Ajax was just a bit stronger than PSV, particularly near the season end, and Ajax’ 3-2 win in Eindhoven put them in the driver’s seat towards the title.

The pundit would push Ajax forward as the strongest team, and immediately start feeding you correlations disguised as explanations as to why Ajax is stronger than the other teams. Ajax is building a ‘successful revolution’, the team plays a ‘recognizable style of football’ and they allow home-grown talent a decent shot in their first team.

The analytics blogger would probably take most time in answering this question, and to be fair, if you would ask me, I would not be quite so sure whether to pick Ajax or PSV.

Limitations

We’ll now look into the important question of determining the best team in the league, based on a few assumptions.

Firstly, playing 34 football matches is a poor way to determine which team has the most quality, as was demonstrated quite elegantly by @penaltyblog.

Secondly, luck has more influence in football than it’s generally credited for. If skill would be the prime factor involved, wouldn’t betting companies be capable of calling the correct winners of football matches in more than the 50% to 55% of cases that they do now?

Experiment

Imagine this thought experiment…

It’s August 10, 2012 and the 2012/13 Eredivisie is about to start in Tilburg, where newly promoted Willem II is about to host NAC Breda. Just prior to the kick-off, we press an imaginary ‘save’ button and quickly fast-forward to May 12, 2013, the final Match Day.

Here we reload our ‘save game’ from August 10, 2012 and we run the same season again. And again, and again and again… A million times.

Using bookmaker odds as an estimate for team strength, we can do just that. After all, bookmaker odds can’t be too far off, otherwise we would be allowed an easy occasion to make some cheap money exploiting them. And if we would all do so, the odds would get corrected quickly. Admittedly, the odds are never perfect and there is quite some evidence that strong home teams will be overestimated, but only slightly so.

The winners

Using some recently acquired skills in R we can run a simulation of one million seasons in just under four minutes, so let’s go straight to the winners…

The 2012/13 Eredivisie, as we can see, was more of a two horse race than it was held for. On the other hand, this simulation also turns out that only two teams – now relegated VVV and Willem II – did not manage to win a single simulation out of a million runs. All other teams, had at least a one in a million shot at the Eredivisie title, if we use bookie odds as an indicator for their chance of winning matches.

Process versus results

This should be an important piece of information for people in charge of professional football clubs. The old adage had always been that ‘the winning coach is always right’, and going by the recent laudation of Frank de Boer as up and coming manager, that adage is presently as strong as ever. Results dominate our image.

In fact, it’s the underlying process that should get our attention. Give me a manager that substantially increases my team’s chances of winning the league, rather than a manager that wins the league without an underlying improvement of performance. Professional football organizations that appoint managers on the basis of results only, run a serious risk of dumping a manager who was developing the underlying process well and trading him for a manager that has experienced just a bit more luck.

The separation of process and results will be an important goal at 11tegen11 during the upcoming season, and using simulations will be a crucial tool to do so.

Some of you may have noticed a decrease in the frequency of new articles on 11tegen11. I can assure you that this is a temporary thing and activity on 11tegen11 will pick up over the summer. It is in part due to the start of Dutch Volkskrant blog ‘De Zestien’, where you are more than welcome to read articles by @SimonGleave, @Tijsrokers, @MichieldeHoog and myself. I hear Google translate does a decent, and on occasion surprisingly entertaining job in making the Dutch articles accessible to foreign readers.

With Game State we indicate the score differential of the match in-play. Each match opens with both teams at GS 0, and a scoring team moves to GS +1, with the conceding team moving to GS -1. This Game State obviously has a big influence on how teams approach the game at hand. However, in traditional – if I may say so in this young business – football analytics groups all match events together, regardless of Game State.

The best concepts in football analytics make rational sense as well as intuitive sense. And such is the case with Game States. A team holding a narrow lead is a different team than a team that defends that lead. Obviously, better teams hold more leads than they defend, but even within teams, the shifts that occur when Game States change are fairly homogeneous. We’ve learned before that moving from GS 0 to GS +1 brings an average team a 10% decrease in Total Shot Rate, while the opponent increases 10% simply because of the shift in Game State.

On this day before the big game, PSV – Ajax, we look at the two best teams of the Eredivisie with a focus on their performance levels at the most crucial Game State: GS 0. The main reason for doing so, and I can safely say this out loud now, is that I have my doubt about the accuracy of the Total Shot Rate model used to predicted the final Eredivisie standing. It has significantly overestimated PSV and underestimated Ajax.

The model uses the Relative Shot Rate (RSR) to estimate the total points at the end of the season. RSR is a variation on the Total Shot Rate (TSR). Early in the season, the RSR has advantages over TSR, because teams have encountered a different strength of opposition, but by now those advantages have gone and RSR is nearly equal to TSR. At the moment, PSV’s TSR stands at 0.671 with Ajax at 0.632. Now what does this figure tell?

PSV has a higher ratio of chances created and conceded. Does this single figure make PSV the better team? No, because you may generate all the chances you want, you’ll need conversion as a skill to turn shots into goals.

PSV’s shooting percentage stands at 17.0%, which compares favorably to Ajax’ 15.4%. Does a higher TSR in combination with a higher shooting percentage make PSV the better team? No, because you can score all you want, you’ll need to prevent the opponent from scoring from their shots too, and this is where saving percentage comes into play.

PSV’s saving percentage is 87.4%, compared to Ajax’ 89.5%. But wait, that’s about the same difference as we found at shooting percentage, only this time PSV comes out on top. That’s true, and so both teams have a comparable PDO, which is the sum of shooting percentage and saves percentage. PSV’s PDO is 1044, and Ajax’ PDO is 1048.

So, if we wrap these numbers up we can safely say that PSV generates a higher ratio of shots. Taking shooting and saving into account, both teams are roughly equally efficient. Now why doesn’t PSV live up to the expectations of our TSR model?

The answer is to be found in game states. We can repeat the exact same exercise of looking at shot rate, shooting percentage and saves percentage for each game state. I won’t go over every single number, but instead focus on the most crucial Game State: GS 0. The average Eredivisie team plays out nearly 50% of shots at this Game State, but since Ajax and PSV are the two top teams, they can be expected to play out less shots at GS 0. Of all shots in Ajax’ matches, 41.9% take place at GS 0. For PSV this number is 35.4%.

Here’s the table for PSV and Ajax in terms of TSR, shooting and saving efficiency, and PDO at    GS 0. Note that PDO in this case provides a nice summary of efficiency, wrapping up both offensive (shooting) and defensive (saving) skills.

The TSR tells us that at the most crucial Game State (GS 0), Ajax is by a distance the better team in terms of shot creation. PSV partially makes up for the lower TSR with their shooting percentage of 15.1%, which is higher than Ajax’ 12.8%. However, PSV loses this advantage in saves percentage, because their 86.7% is much lower than Ajax’ 92.4%. The combined efficiency is higher at Ajax, indicated by a PDO at GS 0 of 1052, compared to 1018 for PSV.

So, analyzing all shots in every match in one group, PSV seems the better team.

But at GS 0, the most crucial stage of the match, Ajax creates a better shot ratio, and is more effective. They gain more leads, which is a good thing in itself, but it also allows them to play more time at favorable game states, leading to an even better performance.

This post is a translation of yesterday’s article for ‘De Zestien’, the football blog of Dutch national newspaper ‘De Volkskrant’. Admittedly, it turned into a rewrite, more than a translation.

Providing an answer is not the hard part here. Providing a correct answer is.

Let try this, often heard answer. It’s simple. Better teams have the better players. Better players hit more difficult shots, leading to more goals per shot.

Game states

The problem with this answer is not that it isn’t correct. Because it is. Better teams have better players, and these players turn more shots into goals than weaker players.

The problem with this answer is that it stops most people from looking beyond it and consider other factors that come into play here. And you’d probably guessed from the title of this article already, that it’s game states and conversion that I would like to link today. It turns out that game states may well explain more of the variation in shot conversion between better and weaker teams than player quality will ever do.

Two weeks ago I wrote about Total Shot Rate (TSR) and Game States. Let’s recall the graph that was central in that piece.

Let repeat this exercise for shot conversion. So, here’s the same graph, linking shot conversion and game state. Please note that this graph contains all shots from all Eredivisie teams in the present season until match day 27. This time I’ve concentrated on GS -2 to GS +2, to prevent the low numbers at more extreme Game States from disturbing the picture. The shot numbers at different Game States are 371 shots at GS -2, 1160 shots at GS -1, 2885 shots at GS 0, 922 shots at GS +1 and 363 shots at GS +2.

It turns out that, like TSR, shot conversion is also related to Game State. TSR had a complicated shape, with an inverse correlation at close GS, but shot conversion is a lot easier to digest.

In general, the more favorable the Game State, the better shot conversion is. The only exception is at GS 0, where shot conversion is lower than at GS -1 and GS +1. Overall shot conversion for the league is 11.7%. Shot conversion at favorable Game States is significantly better, with GS +1 at 14.1% and GS +2  at 17.1%. The most interesting observation is that shot conversion at GS 0 (10.4%) is lower than at GS -1 (11.6%).

Things become more interesting when we combine the conclusions from TSR and shot conversion at different game states.

GS 0

At GS 0, both teams are by definition balanced in terms of TSR, as both team are at the same game state and each team’s shot created is a shot conceded by the other team. In terms of conversion, this is not a fruitful game state. This may well be due to the fact that teams are inclined to be more cautious at this score, since they have a point to lose, particularly near the end of games. A further explanation may be that this Game State, by definition, occurs more in the opening stages of matches, and teams may well be more conservative at the start of a match than they are at the end.

GS +1

At GS +1, there is an interesting trade off. The TSR declines over 10% to 0.443, while the shot conversion rises to 14.1%. The tricky situation with TSR is that it works two ways. The leading team creates 44.3% of shots, but it concedes 55.7% at this game state. So, overall, the TSR of a chasing team is 26% higher (0.557/0.443) than the TSR of a team defending a single goal lead. Despite the fact that the conversion at GS -1 is better than at GS 0, teams at GS +1 convert 22% (0.141/0.116) better than teams at GS -1.

If we combine the shift in TSR and shot conversion for teams at GS -1 and GS +1, we find that a team at GS +1 pays 26% of TSR to gain 22% in shot conversion.

So, generally speaking, teams have a slightly worse chance of scoring when leading by a single goal than when chasing a single goal.

GS +2

At GS +2, there is a whole new world. Teams at GS +2 have restored their TSR to 0.495, while their conversion rises further to 17.1%. Teams at GS -2 fall back in terms of TSR (to 0.505), while their conversion drops to 8.6%. Overall, both teams create a roughly equal amount of chances, but the team leading by two goals converts nearly twice as much.

In the end

Let’s turn to the opening question once more. Why are better teams able to convert a higher percentage of their shots into goals? They take a much higher proportion of their shots at favorable game states.

Who are the conversion kings of the Eredivisie? Right, PSV at 16.9%. They took 20.6% of their shots at GS +2 or higher, compared to 8.3% on average for the other Eredivisie teams…

Or, quoted from Nate Silver’s inspring book ‘The Signal and the Noise’, “a failure to think carefully about causality will lead us up blind alleys”.

Football analytics

Too many people think that football analytics revolves around fancy individual player analysis, multivariate scouting models or complex GPS tracking data of on field events. While it may be true that these ‘holy grails’ get most attention, it’s the simplest of questions that yet remain unanswered. With this tendency to run before we can walk, we run the risk of falling and hurting ourselves over and over again.

Think about this very simple question: why do better teams win football matches?

Obviously because they score more goals than weaker teams.

This bears down to two factors involved in goal scoring: creating shots and converting shots. And, of course, to the defensive equivalent: preventing shots and saving shots. This post will focus on shot creation, and a quick follow-up post will take on shot conversion later.

Total Shot Rate

In order to assess shot creation and prevention in a single number, we’ve become familiar with the concept of shot rates: Total Shot Rate (TSR) and Shots on Target Rate (SOTR). I’ve made no secret of my preference for TSR over SOTR, simply because it has three times more shots to work with and variation in offensive and defensive shooting accuracy between teams is mainly noise. For this analysis, however, I will also include SOTR to serve the audience of people who still believe certain teams to be substantially better at hitting the target than others.

Let’s look at shot creation and prevention by way of TSR for different game states (GS). With GS, I mean score differentials while the shot is taken. Each match starts with both teams at GS 0, until one team scores a goal and moves to GS +1, with the opposing team moving to GS -1.

This is a very interesting graph, containing a wealth of information in a single line. Broadly, the line moves from the lower left hand corner to the upper right, indicating that either leading teams create more chances, or the reverse causation, teams that create more chances end up leading games.

But there’s much more to this graph than just that. Let’s start with GS 0. This point of the graph will always be 0.500. Football is a closed model, meaning that one team’s created shot is another team’s conceded shot, and at GS 0, both teams are at that same state. So each created shot is automatically a conceded shot in that same category. Likewise, shots created by teams at GS +1 are conceded by teams at GS -1 and vice versa. In short, the graph will always be a point symmetrical around GS 0 and 0.500.

Now, while it’s true that the line roughly indicates that there is a positive correlation between TSR and GS, the catch is that over 80% of shots take place at GS -1, GS 0 or GS +1, the so called close game states (CGS).

And in these CGS, there is a negative correlation between TSR and GS. In simple words: teams that go a goal up create over 10% less chances, and allow over 10% more chances at the same time. The shift in TSR is over 25% in favour of the team trailing the goal.

Shots on Target Rate

We can’t really judge this trend without taking the accuracy of shots into account. Therefore, I’ve also included the same graph for SOTR and GS.

To cut a long story short, both graphs are virtually identical. So, the hypothesis that teams at GS -1 take overly hopeful pot-shots does not gain ground from these data. At least, the shot accuracy between GS -1 and GS +1 is virtually identical. This does not mean that the quality of the shots is comparable too, but we’ll go into that when we look at conversion rates.

First, I want to stress the implications of the fact that TSR is negatively correlated with GS for over 80% of the match. This means that teams that spend a lot of time trailing a single goal, will have an inflated TSR, while teams that spend a lot of time defending a single goal lead will be underestimated in terms of TSR.

Predicting

In early September, only four matches into the season, we’ve gone bold by publishing a predicted final standing of the Eredivisie based on TSR. There are some interesting over- and underestimations in this prediction to learn from.

Willem II, now bottom last and near-certain relegation candidates, had been predicted 14^th with 37 points. Of their shots created and conceded, 25% took place at GS -1, compared to a league average of 17%, and only 7.3% at GS +1. This has significantly overestimated their qualities.

Another overestimation is RKC. Top-half in terms of TSR, they are now 14^th in the actual league table and are battling to avoid the relegation playoffs. Needless to say they won’t make their predicted 7^th spot. Of RKC’s shots, 23.5% took place at GS -1, and a somewhat substandard 15.5% at GS +1.

There are also some interesting underestimations in the model. Feyenoord do better than their TSR would suggest, as they spent over 40% less time at GS -1 than the average team and nearly 30% more at GS +1. Vitesse are another example, with nearly 20% less time at GS -1 and over 25% more time at GS +1.

PSV and Ajax are less affected, because they also spent quite some time at GS +2 and higher, where leading teams create a dominant TSR.

In the end

In short, TSR depends on Game State. Over 80% of all shots are contested at Close Game States (GS -1, GS 0 or GS +1), where TSR and GS are negatively correlated. Teams that spend a lot of time trailing single goals are overestimated in terms of TSR, while teams that spend a lot of time defending single goal leads are underestimated.

For those unaware of the terms RSR and PDO, let’s start with the latter for a short summary. A more extensive description can be found in the post on separating luck and skill, which introduced the concept of PDO on this site. The term PDO is adapted from ice hockey analysis, where it was introduced by Brian King (whose internet alias happened to be ‘PDO’) and picked up by James Grayson, who first applied it to football and has taken it further from there.

PDO is the sum of a team’s saves percentage and shot percentage, where saves percentage is the fraction of conceded shots that don’t result in a goal, and shot percentage is the fraction of shots created that results in a goal scored. For convenience, PDO is multiplied by 1000 to get rid of the decimals.

The RSR is an extension to the TSR, which stands for Total Shot Rate. A team’s TSR is computed as the fraction of shots created from the total number of shots in all matches played by the team. The RSR is a slight adaptation, which compares a team’s number of shots created and conceded with the league average against the same opposition. More details on the method behind TSR and RSR are found here.

Without further ado, here’s the EPL league table, updated with Match Day 22 results, including RSR and PDO. Remember, a high RSR signifies a relatively high ratio of shots created, and is a strong characteristic of sustainable good performance. A high PDO signifies a high ratio of shots converted and/or saved, which has proven to be a lot less sustainable over the longer term.

In general, high PDO teams are found in the top half of the table, and low PDO teams in the bottom half. The team with the highest PDO (1062) is Manchester United, mostly due to their immense conversion rate of 17.4%, which is over 50% better than the league average of 11.1%. The other exceptionally high PDO is Chelsea, but also Stoke, West Ham and Swansea punch above their weight, with PDO’s at a level that could only be sustainable by top teams. From a recent long term PDO analysis, we’ve learned that PDO’s outside of the 980-1020 zone seem unsustainable beyond the scope of a single season, while inherent differences in team quality may account for variations within this zone. So, we may expect Manchester United, Chelsea and to a lesser extent Stoke, West Ham and Swansea to drop off a bit in the remaining part of the season.

Remarkably low PDO teams are clustered near the bottom, where all of Newcastle, Aston Villa, Southampton, Wigan and QPR look set for an improvement on their points-per-game haul so far. Also, Liverpool and Tottenham rank low in terms of PDO, which means an improvement in terms of points-per-game is just around the corner.

RSR

In terms of RSR, there is quite a clear top-3, with Manchester City, Liverpool and Tottenham the only teams above the 0.600 mark. This means that, with hypothetical equal conversion rates, these three teams would be in a close fight for the title. And, since shot rates are a lot more sustainable on the long term than conversion rates and saves rates, these three teams reflect the best underlying performance level. Behind them, Everton and Arsenal are a close 4^th and 5^th, with Chelsea at 6^th place. Perhaps remarkably, league leaders Manchester United come in just 7^th in terms of RSR. This indicates that Sir Alex Ferguson’s team is highly reliant on substantially higher conversion and/or saves rates, which seems a precarious base for future success. However, so far, their exceptional PDO has earned them a gracious seven point lead over rivals Manchester City, which may well be enough to win the league.

Based on these parameters, the top-3 will most likely be United, City and Tottenham, with a close battle for fourth between Chelsea, Everton, Arsenal and Liverpool.

A the bottom of the table, both RSR and PDO spell doom for recently promoted Reading. They are the bottom team in terms of RSR, and by a distance, but their PDO of 1017 indicates that their shots and/or saves percentage has been above the average EPL level, which is more than can realistically be expected of this side. A PDO at the low side of the 980-1020 zone seems more realistic and a disconnection with the pack battling for survival seems imminent for Reading.