Quantifying Attack and Defensive Awareness

19 Jun

Building on my last post which used running correlation to quantify the movement relationship between dyads of players this post will introduce two other metrics to measure the impact of player and more specifically sub-system coordination.

To do this I will again be analysing the Brazil vs. Japan game but specifically Neymar’s ‘wonder’ strike for the first goal.

I am looking to analyse the player distances during the build-up to the goal for the Brazilian attacking (Neymar, Oscar and Fred) and the Japanese defending sub-systems (Yoshida, Konno and Nagatomo).  This will be done through measuring the radius of each set of players over the period of the attacking movement. The aim is to show how Japan’s defensive structure was destabilised by not only the exceptional off ball movement of the Brazilian trio but also the inept game intelligence of the Japanese trio.

How Do You Do That Then?

To analyse this I have again generated my own XY coordinates by noting down the positions of the 6 players after each 0.3s frame (roughly). See the last post for more detail on this process.  After this I then calculated the distance between each player in their sub-system and then in their attacker and defender pairings.

By calculating this it allows you to see the shape of the corresponding systems, how they interact with each other at key moments and become destabilised.

Example of Brazil and Japan Radius

Example of Brazil and Japan Radius

Analysing Defensive Stability

To start with we will look at the changes in overall radius between the two aforementioned sub-systems.  Using the running correlation technique you can compare the relationship between the changes in radius for Brazil and Japan.  The r -1 across the first 4 movements indicate that the two systems are changing radius in opposite directions with Brazil shrinking (-1.5m) and Japan expanding (+5.2m).  This is a very important observation as it potentially shows the difference between good and world class players.

Radius correl

How I hear you ask? Well this indicates that the Japanese defenders are only able to process one set of information (in this case the trigger of Marcello’s turning: Movement 3) and are not able to see that Neymar et al have  by movement 4 changed direction and started to aggressively shrink in radius.  Why is this important? well as @SoccerStatHunt realised:

Tweet

This is the beauty of the running correlation it’s a simple but hugely effective tool.  In this case it isn’t that the Japanese players are so much ball watching it’s that they assume the attackers are just going to move towards the goal (why wouldn’t they they’re attackers after all?). If we look closer at the Japanese inter player dyad distance above, we can see a large expansion between Yoshida and Nagatomo.  In the game footage we can see Yoshida has initially tracked Neymar forward while Nagatomo has not tucked inside Fred to keep the distance constant and the unit stable.

By movement 5 (Marcello preparing to cross) it is too late and Brazil have succeed in breaking their relationships with their paired defender.

This is exemplified by graph below showing the distance between Neymar and Yoshida.  The distance initially shrinks by 1.4m to a gap of 0m by movement 5 when the distance swings by a massive 3.4m by movement 9.  If you watch the goal you will see Yoshida starts to automatically retreat towards goal in anticipation of Neymar doing the same.  Instead Neymar has faked and then changed direction to create the initial peel of space.  By the time Yoshida has realised this (Movement 10) it is too late for him to close down the distance only managing to recover the gap to 2.6m upon Neymar’s shot.

Neymar Yoshida Dist

While all this has only happened in a split second, this lack of awareness in reacting in synch to the trigger sets in motion a series of movements that combine to destabilise the relationship in favour of Brazil.

Japan Crumble

As the graph below shows you can clearly see that the Japanese radius is 4m smaller than Brazil’s at the beginning.  This would be expected for the defending team who would look to narrow their shape to become compact and hard to play through.

Very quickly we see a crossing of the lines at movement 3 which coincides with Marcello turning with the ball.  The distance between the two systems, while having inversed, stays stable at 4m before Brazil’s front line blow the stability of Japans defence apart at movement 6 (Marcello crossing the ball).  From here we see  a race between the two systems with Japan only regaining compactness at movement 11 which was too late as this is the moment Neymar unleashed his wonderful shot.

change in radius

So while this shows the change in the two systems total radius over the whole movement when we look at the moving difference in each systems radius we see a different picture.

Moving difference

The above graph shows a really interesting relationship between Brazil’s and Japan’s defence.  We can see that for the first 2 movements the two systems are in a steady state.  The first large difference occurs at movement 3 which reinforces Marcello’s turn acting as a trigger point for the 3 Brazilian forwards whilst the Japanese defenders are stuck in typical defence mode.

We can see that the two systems then stop expanding and contracting at movement 5 indicating that the distance between the players has steadied before again at movement 6 (the cross) we see Neymar, Fred and Oscar drastically reduce radius by 8.64m which is led by Neymar peeling of his defender and Fred stepping in to receive the cross. From here we still Brazil’s radius continue to reduce but at a slower rate as the players now purposefully position themselves with the flight of the ball before the goal is scored at movement 6.

Conclusion

Hopefully this post has shown how by adding inter player distance alongside using the running correlation we start to add the context of the situation through numbers.  This clearly shows that inter personal space is a key metric in understanding how goals/shots are created, but also gives a good indication of the gap in game awareness between sets of players.

 

Analysing Brazil’s 3rd Goal: Understanding the Impact of Players Movement

17 Jun

A number of bloggers have really started to up the ante on the standard of data being collected and analysed recently.  In an attempt to support the now seemingly ever expanding quality and depth of analytical writing I have decided to attempt to apply some of my current research to goals scored in the Confederation Cup.

For this piece I have generated my own XY coordinates by basically being very sad and freeze framing every 0.3 and then capturing the positions of each player involved in the move using a simple X Y grid which runs 50 to -50 on the Y axis and 25 to -25 on the X axis.  As a result I’ve been able to produce a crude 2D representation of player’s movements.

For the purpose of this blog I have concentrated on Brazil’s 3rd goal against Japan scored by Jo.  The justification for this is that the goal allows the use of certain statistical tests, in this case running correlation, to determine the impact certain players have on others and at what point a defence moves from being stable to unstable (perturbation).

Running Correlation

I’m sure most people are now well familiar with correlations and will have seen these along with an r or r2 value quoted in a number of blogs which indicates the strength of relationship between different variables.  Just as a reminder ±1 indicates an almost perfect relationship (+ representing one that goes in the same direction and – showing one which goes in opposite directions) with 0 showing there is no relationship whatsoever.

A running correlation does the same, however, like a moving average takes a few samples of data and follows these over a period of time.  In this case I will be tracking the relationship between Jo (the striker) and Yoshida (the marking centre back) X coordinates (across the pitch) and Y coordinates (towards goal).

The reason for this is if we have the ability to capture type of movement that will lead to a quality goal scoring opportunity; we can then start to identify specific movement patterns to train with greater certainty which will hopefully lead to more success.  Also we can start to capture who has a higher level of movement intelligence in attackers and which defenders are able to recognise and prevent this type of movement.

Hopefully You’re Still With Me!

JoFirstly I plotted the movement of the 6 players involved; 3 attackers and 3 defenders.  As we can see straight away there is a general relationship between the 3 pairs of players which we can term a dyad:

  1. Oscar vs. Uchida
  2. Jo vs. Yoshida
  3. Lucas vs. Konno

As we know Oscar was the initial protagonist picking up the ball in his own half and running towards Uchida.  If we isolate this dyad to start we can see that Oscar is making a straight run while Uchida has to check to the left before the pair form a similar relationship moving towards the goal.

Oscar Uch

This is evidenced through the running correlation.  As the red line shows very quickly the relationship between the players movement in the X axis goes from one of stability (+1) to instability (0).  This represents the fact that both players are traveling in completely different directions on the X axis.  This is further backed up with the Y axis which moves for +1 to -0.7 which indicates that the two players are moving towards each other or to put it another Oscar’s affecting the movement of Uchida.  For the rest of the move we see what we call an anti-phase relationship between the movement in X and Y plains which means that the players are moving in and out of synch as Oscar tries to disturb the dyad to his favour while Uchida attempts to maintain stability.

As shown on the graph the Y correlation is near perfect 1 from movement 7 (toward goal) but is in the X plain you can see Oscar is trying to move left and right to win the battle.  Eventually he does by movement 10 were the relationship is totally destroyed (0 correlation).  This matches when he plays his sublime pass to Jo.

Jo vs. Yoshida

Ultimately a great pass is only possible if the supporting players provide great movement to a) create the space and b) actually receive the ball.  The Jo/Yoshida dyad is a fantastic example of a top class movement and unaware defending.

Jo Ypshida

This time we see a different relationship.  As in Fig 1 this dyad starts in the defenders D on the halfway line.  The correlation shows that there is a near perfect relationship between Jo and Yoshida as they both move toward the goal area (Y axis) all the way through until movement 9 where we see the correlation move from +1 to +0.76.  This represents the moment when Yoshida loses Jo and stops to see where he is, and where Jo then takes advantage and gets ahead to allow the pass from Oscar (remember Oscars relationship with Uchida plummets at 9 and stops at 10).

This isn’t the whole story though.  If we look at the X axis we can see that there is a very definite anti phase relationship with the correlation moving between +/- 1 up until the goal is scored at movement 14.  The most important part is at movement 6 where the correlation drops of -1.  This is where Jo moves away from Yoshida while the CB continues to move in the same direction unaware of Jo’s movement. (If you watch the goal you will actually clearly see this).  The next significant movement is at 10 (when the ball is passed) where get a correlation of -0.45.  This represents Jo making his move into the space the space created with Yoshida to close him down realising its too late.

What happens next is obvious with Jo scoring his first goal for Brazil.

A Pointless Exercise

I’m sure some of you will look at this and go well what’s the point of doing that when you can just see it.  Well the point of performance analysis and analytics is to provide objective data to either prove or disprove gut feeling.  By being able to demonstrate scientifically how movement influences the outcomes of attacks we will be able to collect this data for all successful attacking and defending outcomes.  Used properly we can then develop better suited training programme not just for senior but more importantly for junior players.

This is just a microcosm of some of the work I’m interested in but I feel very passionately about what it can help us achieve in the future.

It Takes a Team of Individuals to Win the League

31 May

There have been some great end of season posts recently about the EPL by the likes of @JamesGrayson and @FootballFactman.  Each has taken a detailed look at the performance of teams as a whole and demonstrated that there are several indicators of success.  However looking at the stats from Who Scored I was amazed to see just how little Manchester United featured at the top of any of the team statistics.  Admittedly they did finish top of what may be considered the two most important metrics; goals scored and passes completed (see table 1).  However, what’s interesting when calculating the Coefficient of Variation (CV%), basically the significant variation between scores, we can see that United score significant lower in a number of key areas including through balls (33.33%), total shots per game (13.45%) and dribbles (25.15%) than the top team, which you would normally associate with a better attacking team.  While they have scored significantly more goals than their nearest rival (6.83%) when breaking the goals down into those scored from open play (1.96%) and from set pieces (1.96%) they are not that far ahead.

Manchester United Team Performance

This led me to ask the question with United winning the league so convincingly without dominating the top of the ratings are individuals more important than the team?

There is a ME in TEAM

To test this hypothesis I again looked at the Who Scored stats on individual players as surely they must have players dominating at the top of these tables?

Looking at the overall player rating United suddenly started to show more dominance with the only team having 4 players in the top 20.  However both Liverpool and Tottenham both had 3 players who scored a higher average ranking.  Interestingly Arsenal and Chelsea only had 2 players and Manchester City had had a paltry 1.  So looking at this I started to conclude that maybe to be the best you need more individual players to perform at a high level over a season as opposed to a team of steady eddies.

Top Players

@TedKnutson recent great work on identifying players has targeted the importance of assists.  Looking at the table 11 out of the top 20 players play for top 4 teams however Arsenal (30 in total), Chelsea (29), and Manchester City (23 in total) have 3 players each with United having 2 (18 in total). Interestingly when looking at the key pass metric United didn’t have a single player in the top 20 while Arsenal had 2, Chelsea 2 and Manchester City 2.

Assist Top

Hmm so I wasn’t really sure if this proves the theory or not, so I then went on to compare the number of players each team had that contributed the following:

  1. An assist
  2. 3 or more assists
  3. A goal
  4. Scored 3 or more goals

Assists

As the results show United had significantly more players contribute an assist and more than 3 assists, in addition they also had significantly more players score a goal however there was no significant variance in the number of players who contributed 3 or more goals.

What is an interesting aside is the number of players who have both scored and create goals for all 4 teams with none being lower than 11 proving how much football is now a squad game.

Conclusion

This piece suggests that to win the league you need both a group of 3-4 top performing individuals throughout a season and a team that can reduce the pressure if one or more of these top players fails to perform.  This is quite clearly a blend the United have managed to create time and time again.  While they might not have had as many players in the top 3 sports as Chelsea or Arsenal across the different metrics they have had a greater depth of players contributing.  This is clearly demonstrated by the number of players who scored and assisted goals significantly out performing their nearest rivals.

This finding is quite intuitive with teams like Arsenal and Tottenham having to heavily rely on the same top performers i.e. Carzola and Bale week in week out in order to stay in touch.  This indicates that to be a top 4 team you need 3 top class players leading the way in several categories but to win you need a full squad (look at how many players contribute to United’s goals and assists) to be able to back them.

DEFF: Do Shots Result in Goals

3 Apr

Building on the data I’ve started collecting on DEFF I thought it prudent to ask the question do shots actually lead to goals?  This question on the surface of it may seem mad however if we examine the stats there is method to the madness.

The first thing to accept is that shots are a rare occurrence in football when taken into account for a 90 minute game, with goals being an even rarer occurance, hence the soaring costs of strikers.  Therefore do the number of shots a team concedes actually relate to the number of goals or is it luck/skill of the specific player taking it?

Method

I have collected the number of total goals conceded (GC), total shots conceded (TSC) and the SC in the box (SCB) for each team playing in the EPL between the 2008/09 season and 2011/12 season (props to @dankennett for the data).  This has provided a sample of 80 teams over 4 seasons and 1520 games, making the sample size reasonably robust in terms of carrying out statistical analysis.

From this data as with the previous posts I have used a simple correlation to determine the strength of relationship between the different variables.

Results

Looking at the relationship between TSC and GC there is a moderately strong correlation (r2 = 0.51) which indicates that 51% of GC are as a result of the total number of shots conceded by a team.  As you can see this is quite a strange finding which indicates that because a team concedes a large number of shots it doesn’t necessarily mean they will concede lots of goals.

TSC vs GC

Presumably however there will be a significantly stronger relationship between the number of shots conceded in the box and the number of GC as this is what we’re told by the experts? Well as you can see the r2 value has increased to 0.64 (rounded up) which shows that 64% of goals conceded are as a result of shots conceded from inside the box.  The relationship is stronger but again you would assume it shots in this key area should account for closer 80%.

TSCB vs GC

Conclusion

So what does this data tell.  Firstly that conceding or having lots of shots does not mean your team will score/concede lots of goals.  This leads to the several potential conclusions:

1. Luck plays a large part in goals being scored

2. The standard of shooting is extremely poor outside of goal scoring experts

3. Only a minority of players have the ability to convert chances

4. Defences are extremely efficient at preventing high quality goal scoring opportunities

5. Teams are extremely wasteful of opportunities or lack the patience to create a ‘genuine’ goal scoring opportunity.

The next step to add more clarity to this position will be to collect total shots on target data, who takes the shots and where are they taken from.  This will then allow us to understand just how much of a waste a time shots actually are.

DEFF: Relationship Between Goals Conceded and Points Dropped

1 Apr

I initially introduced the concept f measuring a teams true defensive efficiency (DEFF) in previous posts based upon several stats such as SoTC and SoTC/GC. This post looks at the relationship between Goals Concede (GC) and the number of points a team drops per season as opposed to wins.  Why look at this you may ask? Well as we know the team who drops the least amount of points wins the league and generally teams that have the worst goal difference get relegated (see Omar Chaudhuri’s excellent 5 added minutes blog for more on that).  However as seen in my last post the teams at the top and bottom of the goal difference table don’t necessarily follow the same pattern GCs. 

Aim

With this in mind I wanted to see if there was a relationship between the number of goals conceded vs the number of points lost during a season and league position. 

Method

Using data from www.statto.com for the 2010/11, 2011/12 and 2012/13 season I collected the number of goals each team conceded (GC) and calculated the number of points dropped (PD) by teams (points won – total points available).  I then divided PD by GC to calculate the average number of points dropped per goal conceded.

Results

The graph below shows the league position vs. average points dropped per goal conceded (PD/GC) (the variables).  The green line indicates the teams that finished in Champions League positions, the orange line shows the teams qualifying for the Europa League, the gray lines shows the half way point and the red line shows the teams in the relegation zone.  The legend on the graph show the different league seasons and the black line of bets fit indicates the basic pattern or distribution of the variables.

Image

As can be seen the is a very clear bell shaped curve with the teams finishing in the top 4 and bottom 3 having, on average, the lowest PD/GC with teams finishing in the mid table table cluster of 8-14th having the highest.

In order to finish in the top 4 it would generally appear that a team needs to have PD/GC -0.2 below the league average PD/GC.  Therefore as shown previously a team finishing higher up the league can afford to concede a relatively higher number of goals per game over the season due to their higher ability to score more goals.  Conversely teams who are in the relegation should maybe not head quite as much on the side of caution as was previously thought.  While keeping a tight defense is an important quality and one which is constantly pointed to by the pundits, the data indicates that a team should look to risk conceding more goals in exchange for increasing their chances of scoring more. 

Southampton are an excellent example of this. In the past 6 games they have the a record of W3 D1 and L2 with a goal difference of 11-9.  This is a PD/GC differential of 0.78 which is top 2 form.   So while Southampton have conceded 1.5 goals per game they have committed to being brave and attacking their way up the table.

Conclusion

The aim of the post was to see if there was a relationship between league position goals conceded and points lost.  The initial data shows that that the top teams have the lowest as would be expected however more interesting is the relationship between PD/GC and finishing in the relegation zone and mid table. 

If anyone would like to look at the data I’ve pulled together or look at collaborating on future work please feel free to get in touch.

Which Team has the Best Defensive Efficiency (DEFF)

28 Feb

Aim

In the last post I introduced the concept of giving a true measure of a team’s Defensive Efficiency (DEFF). For full details of the 6 metrics click here for the last post.

The aim of this post is to rank teams by their DEFF.  This will be done by looking at the number of goals a team concedes (GC), the number of goals a team concedes  from total number of shots they concede (GCTS%) and the number of goals a team concedes from the total amount of shots on target they concede (GCSoT%).

Data Collection

All data has been collected and verified from @StatsZone app and verified against the @Squawka_Sports website both of which are powered by Opta.

A shot on target is defined as a shot that has either been saved by a goalkeeper or resulted in a goal including penalties and free kicks.  A shot off target is defined as one that was blocked, hit the wood work or missed the target including penalties and free kicks.

Results

Goals Conceded

When looking at the GC the table generally follows the same pattern as that of the current standings of the EPL with the4 most efficient defences in terms of goals conceded being in the top 5 of the league with Stoke being an anomaly (5th vs. 10th in the league), providing some more objectivity to Stokes reputation of being hard to beat and not giving much away.  A similar pattern can be seen at the bottom of the table with 3 out of the bottom 4 teams (Wigan 51, Aston Villa 50 and Reading 48 GC) with the most GC being in the bottom 4 of the league.  The biggest surprises in the GC Table are QPR who lie 13th (41 GC) as opposed to 20th in the league, and Manchester United who lie 6th (31 GC) in the GC Table which indicates both how poor QPR’s attacking players are (19 goals scored) and how good United’s forwards are (62 goals scored).

Goals Conceded From Total Shots (GCTS%)

This is where the Table starts to become interesting and indicates a team’s true DEFF.  From there being a total goal difference of 19 goals from Manchester city (24GC) to Wigan (51GC) when converting to GCTS% there is only a 5% difference in efficiency from top to bottom clubs.  The first thing to note is the change in make up the top 4 teams; Sunderland, Swansea, WBA and Stoke, who only concede 8% of the total number of shots. What’s interesting is that these 3 teams also concede some of the highest number of shots yet concede very few.  Sunderland in particular show that while they have TSC of 443 they limit teams to low quality chances a good sign of DEFF.

The teams with the highest GCTS% again see’s Wigan as the worst club (13%), however Tottenham have dropped from 4th in GC to joint bottom with a GCTS% of 13% from 248 TSC which is actually the least number of shots conceded in the league.  This indicates that either Lloris isn’t particularly safe or the chances the team are conceding in highly susceptible to conceding shots in high % goal scoring areas such as the famed double 6yrd box. 

Table

Goals Conceded from Shots on Target (GCST%)

Once again there is quite a change at the top of the table with Chelsea and Swansea conceding the least (24%), Wigan are rock bottom again (38%). Swansea prove an interesting case conceding with one of the highest SoTC rates yet joint 2nd in GCST% indicating that they may be both lucky and that Vorn is an exceptional goalkeeper.   Suddenly there is bit more context to the DEFF of Manchester City.  Previously coming out top in GC and GCTS% they now lie 14th conceding 33% out of 77 SoT which for a team in 2nd place is an abysmal DEFF.  This indicates that while they may not concede many shots the quality of the chances are either very high or Joe Harte isn’t quite as good as we possibly think?

The other major eyebrow raises are Tottenham (34% out of 89 SoT), Liverpool (36% out of 95 SoT) and Everton (32% out of 99 SoT) all showing that while these three teams may be a bit more gun hoe in their attacking play they are extremely susceptible to conceding goals.   Interesting QPR come 9th (30% out of 138 SoT) reaffirming the previous statement on their inability to score as oppose to conceding goals being the reason for their league position.

Conclusion

Based on this initial analysis it would appear that GC may not tell the whole story of DEFF with GCTS% and SoTC% being a truer measure.  What is clear is that Wigan has by far the worst defence in the league and that QPR were right to invest so heavily in Remy to try and chase those elusive goals. Tottenham may have been a bit hasty in replacing Freidle as first choice keeper and that Liverpool are ‘work in progress’.

Moving forward I will look at the relationship between goals concede and TSC to determine just how much a causal effect conceding shots are on conceding goals.

Measuring Defensive Efficiency (DEFF)

20 Feb

Aim

The aim of this post is to introduce an objective assessment of individual teams Defensive Efficiency (DEFF) to determine if the team that concedes the most goals is actually the worst defense?  While a team may seem to have a relatively high DEFF does this actually hold true when comparing the number of goals conceded to the actual amount of shots they allow an opposing team to have.  For example a team may only concede 20 goals in a season which indicates a high DEFF but if those 20 goals came from 30 shots on target then you would have to question both the quality of the goal keeper and the quality of chances that the defenders are allowing the opposition to have.  It may be that a team doesn’t conceded due to them having such a high percentage of possession however they’re concentration when they don’t have the ball may actually be very poor consequently allowing attackers to find space to exploit or to gain overloads in key areas of the pitch.

Example DEFF for Arsenal

Example DEFF for Arsenal

To try and answer this I have looked at developing several metrics to give a truer measure of a defenses true DEFF.

  1. 1.       Goals Conceded (GC)

This does exactly what it says on the tin and is the total number of goals a team concedes over a season.  GC will be further broken down to GC at home (GCH) and GC away (GCA) to give an indication of whether a team’s DEFF is higher home or away.

  1. 2.       Goals Conceded Difference (GC-D)

This is an indicator of a whether a team’s DEFF is better at home or away and is calculated as follows:

GC-D = GCH-GCA

Where there is a +ive number left (i.e. +5) this indicates that a team has a lower DEFF at home (conceded more goals at home than away).  If there is  -ive number this indicates a team has a lower DEFF in away matches (conceded more goals away than at home). If the number is 0 this indicates that at a team has an equal DEFF home and away (conceded the same number of goals home and away).

  1. 3.       Total Shots Conceded (TSC)

This is the total number of shots conceded by a team over the course of a season.

  1. 4.       Shots on Target Conceded (SoTC)

This is the total number of shots conceded by a team that were on target over a season

  1. 5.       Shots on Target Conceded % (SoTC%)

This is the percentage of shots conceded by a team that were on target and is calculated as follows:

SoTC% = SoTC/TSC

Calculating the SoTC% will give an indication of how effective a defense is at reducing the number of goal scoring opportunities conceded which is a key trait of a strong defense.

  1. 6.       Goals Conceded From Total Shots % (GCTS%)

This is the percentage of goals conceded from the total number of shots conceded and is calculated as follows:

GCTS% = GC/TSC

  1. 7.       Goals Conceded from Total Shots on Target % (GCSoT%)

This is the percentage of goals conceded from the total number of shots on target conceded as is calculated as follows:

GCSoT% = GC/ToSc

The reason for calculating both GCTS% and GCSoT% is to give an indication of the quality of shots given away by a team i.e. the better the quality of a chance the more chance of it being scored but also gives an indication of quality of the goal keeper by assessing how many shots that they have to save are actually conceded.

Conclusion

The above 7 initial metrics will start to provide a more objective assessment of a defenses true DEFF.  This post has acted as introduction to how DEFF will be measured.  Rather than going straight into the results I will start to post blogs breaking each section of DEFF down to help build up the picture of DEFF in the league and if it can help predict league position and the number of goals a team will conceded in a season.  There are some initial restraints which are worth pointing out from the start.  Firstly the assessment doesn’t take into account the strength of the team or opposition which can change due to injury suspension or team rotation.  There is no weighting for the league standing of the opposition as well for example you expect a top 4 team to have more chances against a bottom 4 team.  Finally TSC and SoTC do not take into account home or away bias.

Follow

Get every new post delivered to your Inbox.

Join 38 other followers