Topic: Statistics and Analytics - insight into our performance

[Homo rubrum]

Jesus, now this thread has got me thinking, "Stats!  Finally, a way I could conceivably get into professional sport!" 

What a distorted childhood dream that'd be.   

[barbudo]

If I were running a club, I'd want a decent stable of them. Some of them with coaching and practical football experience and some of them solely data nerds with limited previous biases. I know the Houston Rockets in the NBA have a militia of data crunchers/visualizers, etc. etc. with varying backgrounds.

Anyone see the Opta contest? I want to come up with something, even if the cash prizes can only go to students.

http://optasports.com/showcase-pages/110.aspx

What about something simple  - defender on one, both or neither post during corners?  Combined with subsequent offsides as defenders leave the box?

[The Woolster]

Another post related to the skill/no skill finishing debate posted today

http://www.statsbomb.com/2014/02/comparing-actual-and-expected-goals/

If you're on Twitter and have found the skill/no skill debate on the last few pages interesting, then check out @cchapas who wrote it, the stattos (including the guy who wrote the 'no skill' piece) are currently discussing it now (although its a bit hard to follow with Twitter's conversation format)

[The Woolster]

Cheers The Woolster, very interesting posts. Glad the Optaforum went well!

I do wonder whether the fact the title of that particular piece was 'How to scout a striker?' maybe meant PoP & Elston thought they were further apart than they were/are?

Woolster is fine, "The" only needs to be used in more formal situations 

[The Woolster]

This thread has evolved into an absolute belter.  I wish I had the time to jump into some of these conversations.  As it is, I'm just lurking and enjoying the debate 

Hey Prof, here's an alternative to the Alternative Premier League Table

http://www.stathunting.com/2014/02/15/the-improved-epl-table/

[ElstonGunn]

Neither. I'm saying that you seem unaware of how scouts do their job, and what it is they look for. Of course data is helpful, and eliminates bias - but the things that scouts actually look at are the same things the ExpG is showing. My interpretation of your initial post is that football is doing it all wrong, and that they are looking for the wrong things. I stand to be corrected on that, though, if I've misunderstood?

Not at all! I don't see what I said that would make you think I'm saying football is doing it all wrong--in fact, I said, I thought scouting was absolutely necessary and stats are a supplement that are primarily useful to identify targets in the first place (who you then scout) or double check yourself against making huge mistakes--but I certainly didn't mean to convey that at all. Apologies for a lack of clarity again. I understood you to be saying scouts already know these conclusions intuitively, so stats aren't adding any value.

[Prof]

Hey Prof, here's an alternative to the Alternative Premier League Table

http://www.stathunting.com/2014/02/15/the-improved-epl-table/

Thanks 

[trembles97]

I'm taking Sports Econometrics this semester and the stuff we are talking about is very interesting. A lot of the talk is about baseball, probably because it is the most statistically based sport as well as having extensive records going back 100 years. But, my professor said we'll be taking a look at football later in the semester, and how he did work with our coach at the university about how corners are negatively correlated with win percentage, and how set pieces in general are overrated.

Anyways, I'm wondering if anyone has a preferred site they use for football stats? I know there is WhoScored and Squawka, but I'm looking for something more that has the football as well as financial data. After all, what better way to determine value for money than to regress performance stats on wages 

[Prof]

Just to add my own line of thinking to the shooting/goal scoring debate...

PoP has alluded to the classification of skill as being integral to unpicking the problem.  Finishing is clearly an open, externally paced skill.  There are aspects of the conditions of each shooting chance that will always be different.

In addition, the result of the shot is also an externally affected factor.  A bad shot can result in a goal, a good shot can be saved etc.

In order to simplify this debate, consider it like this.  If the skill of shooting was a (largely) closed, internally paced skill, like a golf shot, or a place kick in rugby, the typical dispertion range would be around 8-12% for an elite player.  That means for a 100 yard golf shot, you'd be within 8-12 yards of the target.  A 40 yard place kick would have an error range of 3.2-4.8 yards.  We can therefore assume that a striker would also have this range of error in their shooting in a closed, self-paced situation.  So a 12 yard penalty kick would have a range of 3-4 feet.

The open and externally paced nature of shooting in football means that the amount of error in a shot could increase, but this would be influenced by the effect of the new variables.  A situation in open play where the striker has time and space would have a closer resemblance in terms of error to the penalty.

We can assume when talking about elite strikers that their shooting skill levels in 'safe' environments are comparable to one another.  This is the case in all the sports I have seen data from.  Where differences will occur in competition will be largely affected by the variable nature of the situations they find themselves in, and the skill of the player at managing those situations (mentally and technically).

The only true way of measuring the ability/skill of the striker at managing these situations would be to measure the accuracy of shots as a % error.  This would require the researcher to know where the striker was aiming.  Measuring the outcome alone is quite misleading as the outcome is affected significantly by external factors.

[Roblaar]

The only true way of measuring the ability/skill of the striker at managing these situations would be to measure the accuracy of shots as a % error.  This would require the researcher to know where the striker was aiming.  Measuring the outcome alone is quite misleading as the outcome is affected significantly by external factors.

It's possibly more complicated than that as shot velocity is also important, and different players will probably have different shot accuracy/velocity profiles.

[Prof]

It's possibly more complicated than that as shot velocity is also important, and different players will probably have different shot accuracy/velocity profiles.

The speed of shot might vary between players, that's true, but it would also vary within the player too.  Therefore I'd assume that there wouldn't be a huge difference in their average shot speed.  It would certainly be an interesting bit of research for someone to do though.

The accuracy of shots is the thing I'd like to see measured as a percentage error value.  Extrapolating evidence from other sports, the amount of variance in the elite players will likely be insignificant for closed skill shots, but open play situations will probably show where the differences occur.  I'd also suggest that this is where 'form' has the biggest impact.

If it could be measured (which I concede might be too difficult in matches), you could then explore the best practice methods to improve this accuracy value.

[The Woolster]

I'm taking Sports Econometrics this semester and the stuff we are talking about is very interesting. A lot of the talk is about baseball, probably because it is the most statistically based sport as well as having extensive records going back 100 years. But, my professor said we'll be taking a look at football later in the semester, and how he did work with our coach at the university about how corners are negatively correlated with win percentage, and how set pieces in general are overrated.

Hmm, not sure about that corner finding, so would like to hear about it, and if true I think would say its correlation rather than causation.

Corners are definitely low scoring opportunites, but my guess is that the number of corners is correlated with the number of shots, and the number of shots correlates with win percentages.

The only explanation I can think of just now is that socre effects is playing a role, as the losing team tends to take the greater % of shots, and perhaps the type of shots they tend to take when losing, and the way the winning team defends when winning, might see an increase in the number of corners for the losing team.

Anyways, I'm wondering if anyone has a preferred site they use for football stats? I know there is WhoScored and Squawka, but I'm looking for something more that has the football as well as financial data. After all, what better way to determine value for money than to regress performance stats on wages 

For me its EPLIndex, but you now have to write for them to get teh stats due to Opta now providing the stats for the Premier League. Otherwise you can pay a couple of quid a month for fantasyfootballscout which has more detail than those other 2.

Transfermrkt has some basic stats and may have contract details, so you could try there

[u27ellisj]

I think a lot people have noticed that there are more and more good statistical writers in football these days. Andrew Beasley (AKA BassTunedtoRed) and Dan Kennett are the most prominent Reds around, and do consistently good work. Then you have others, like MCofA, from cartilagefreecaptain, whose projected Premier League tables we've started to see quite a bit around RAWK; Ted Knutson, Ben Pugsley and Colin Trainor, who have made the new site Statsbomb.com a must read; James Grayson (with a blog named after himself), the best source of pure stats analysis on the web, and even bitters like Paul Riley (of differentgame) and Mike Goodman (of Grantland) who are doing great stuff. Our very own Prof also posted something very interesting about final third entries the other day. (And I'm sure I've missed a bunch of others.)

Instead of cluttering up threads like the Alternate Premier League Table (which is intentionally much simpler than a statistician would make it), I thought it would be good to have one place where we can keep track of developments in the stats world for those of us interested.

To start off, I did my own relatively simple statistical look at how we've played with and without the Lucas-Gerrard duo this season. It's been a little bit of time now since the partnership has been broken up, so I thought I'd take a look at what the stats indicate. I've long maintained that our issue in midfield was never about Gerrard himself (or Lucas), but the lack of mobility we had when both play. Obviously, that may have some effect on my analysis, but I think the facts below speak for themselves.

I focused particularly on 4 stats: total shots for and allowed, shots on target for and allowed, Total Shots Ratio (TSR), and Shots on Target Ratio (SoTR). The first two are straightforward, and are strong indicators of how well we're playing.

TSR and SoTR may be a little less familiar, but they're actually pretty straightforward too. TSR is just a measure of what percentage of the total shots taken are taken by a given team rather than its opponents. So, if Liverpool plays Everton and both sides have 10 shots each, Liverpool's TSR is .50, because it took 50% of the 20 total shots in the game.

There's some debate about how valuable the stat is--Villas Boas' teams always underperform their TSR's, for instance, for obvious reasons--but in general, having a good TSR is strong indicator of being a good team. (The r^2 between TSR and points gained in a season is .67, and it is arguably the single best stat for predicting future performance, which indicates it reveals to some extent "true ability.").

SoTR is the same thing as TSR, but using only shots on target instead of all shots.

When Lucas and Gerrard started together
Total shots against (on target) / shots for (on target): 179 (68) / 195 (76)
Shots against per game: 13.8
Shots for per game: 15
SoT against per game: 5.2
SoT for per game: 5.8
TSR: 52.1 (would be 10th in the league, virtually tied with MUFC)
SoTR: 52.7 (would be 11th, virtually tied with Swansea)

Lucas and Gerrard (only games where Suarez played)
Total: 109 (38) / 133 (50)
Shots against per game: 13.6
Shots for per game: 16.6
SoT against per game: 4.75
SoT for per game: 6.25
TSR: 54.9 (would be 9th in the league)
SoTR: 56.8 (would be 7th, just ahead of Spurs)

When only one of Gerrard or Lucas play:
Total: 82 (20) / 145 (53)
Shots against per game: 11.7
Shots for per game: 20.7
SoT against per game: 2.8
SoT for per game: 7.6
TSR: 63.8 (league best for the season is Man City with 63.7)
SoTR: 72.6 (!!) (league best for the season is Man City with 66.3)

In short, when Lucas and Gerrard have played together, our underlying stats are those of a team on the outside of a Europa League chase*, even if you only include the games when Suarez was available.

When we've only started with one of them, we've been playing like the best team in the league.

It's obviously important not to read too much into something like this. The sample is small, and there are other factors. In particular, our fitness may be improving as a result of Rodgers' periodization methods, and we've exclusively played 4-3-3 and 4-2-3-1 in the games since the Lucas-Gerrard partnership was broken up, whereas we played a variety of formations, including 3-5-2 and 4-4-2 before, when Sturridge was fit.

Nonetheless, I think we've all observed the improvement in our energy and mobility in midfield recently. I think it's a mistake, however, to think the change started with the Spurs game. For me, that added mobility was also evident against West Brom and West Ham, when Allen came in for Lucas and Gerrard played. My tentative conclusion, then, is that Henderson and Allen should be regular starters in midfield, while it should be a straight choice between Gerrard and Lucas for now.

*There are caveats worth noting here about the quality of the shots we've generally allowed and taken. Our opponents generally take shots from pretty poor locations, and we take shots from pretty good locations. There's a lot of debate in the stats world, however, about how repeatable "shot quality" is.

I would have never expected that, wow facts can scare you. Interesting though, great way to view our performances and line-ups 

[redmark]

Corners are definitely low scoring opportunites, but my guess is that the number of corners is correlated with the number of shots, and the number of shots correlates with win percentages.

Corners perhaps correlate better to attempted crosses (blocked at source by the fullback, headed behind) than they do with shots. As everyone except David Moyes knows, crosses do not provide goalscoring opportunities as effectively as through balls. (There might also be some correlation between corners and blocked long shots, another poor goalscoring choice).

[owens_2k]

Just to add my own line of thinking to the shooting/goal scoring debate...

PoP has alluded to the classification of skill as being integral to unpicking the problem.  Finishing is clearly an open, externally paced skill.  There are aspects of the conditions of each shooting chance that will always be different.

In addition, the result of the shot is also an externally affected factor.  A bad shot can result in a goal, a good shot can be saved etc.

In order to simplify this debate, consider it like this.  If the skill of shooting was a (largely) closed, internally paced skill, like a golf shot, or a place kick in rugby, the typical dispertion range would be around 8-12% for an elite player.  That means for a 100 yard golf shot, you'd be within 8-12 yards of the target.  A 40 yard place kick would have an error range of 3.2-4.8 yards.  We can therefore assume that a striker would also have this range of error in their shooting in a closed, self-paced situation.  So a 12 yard penalty kick would have a range of 3-4 feet.

The open and externally paced nature of shooting in football means that the amount of error in a shot could increase, but this would be influenced by the effect of the new variables.  A situation in open play where the striker has time and space would have a closer resemblance in terms of error to the penalty.

We can assume when talking about elite strikers that their shooting skill levels in 'safe' environments are comparable to one another.  This is the case in all the sports I have seen data from.  Where differences will occur in competition will be largely affected by the variable nature of the situations they find themselves in, and the skill of the player at managing those situations (mentally and technically).

The only true way of measuring the ability/skill of the striker at managing these situations would be to measure the accuracy of shots as a % error.  This would require the researcher to know where the striker was aiming.  Measuring the outcome alone is quite misleading as the outcome is affected significantly by external factors.

I think a lot of finishing can be instinctive. Some players do better when given little time, yet struggle when given too much on, say a one-on-one. These are factors which can't be measured.

[The Woolster]

Corners perhaps correlate better to attempted crosses (blocked at source by the fullback, headed behind) than they do with shots. As everyone except David Moyes knows, crosses do not provide goalscoring opportunities as effectively as through balls. (There might also be some correlation between corners and blocked long shots, another poor goalscoring choice).

When teams go behind, they attempt more headers, which is why I said that game states could be a cause, and really, it should perhaps be double checked using data for only when teams are level to see if correlation still exists

[The Woolster]

Just to add my own line of thinking to the shooting/goal scoring debate...

PoP has alluded to the classification of skill as being integral to unpicking the problem.  Finishing is clearly an open, externally paced skill.  There are aspects of the conditions of each shooting chance that will always be different.

In addition, the result of the shot is also an externally affected factor.  A bad shot can result in a goal, a good shot can be saved etc.

In order to simplify this debate, consider it like this.  If the skill of shooting was a (largely) closed, internally paced skill, like a golf shot, or a place kick in rugby, the typical dispertion range would be around 8-12% for an elite player.  That means for a 100 yard golf shot, you'd be within 8-12 yards of the target.  A 40 yard place kick would have an error range of 3.2-4.8 yards.  We can therefore assume that a striker would also have this range of error in their shooting in a closed, self-paced situation.  So a 12 yard penalty kick would have a range of 3-4 feet.

The open and externally paced nature of shooting in football means that the amount of error in a shot could increase, but this would be influenced by the effect of the new variables.  A situation in open play where the striker has time and space would have a closer resemblance in terms of error to the penalty.

We can assume when talking about elite strikers that their shooting skill levels in 'safe' environments are comparable to one another.  This is the case in all the sports I have seen data from.  Where differences will occur in competition will be largely affected by the variable nature of the situations they find themselves in, and the skill of the player at managing those situations (mentally and technically).

The only true way of measuring the ability/skill of the striker at managing these situations would be to measure the accuracy of shots as a % error.  This would require the researcher to know where the striker was aiming.  Measuring the outcome alone is quite misleading as the outcome is affected significantly by external factors.

I've said some similar stuff in what could possibly be described as a Twitter sermon going somewhat against the blog that started this discussion 

I can't get on Twitter on work pc though to copy my thoughts, so you'll have to check my twitter if anyone's interested @The_Woolster

#lookingformorefollowers 

[redmark]

When teams go behind, they attempt more headers, which is why I said that game states could be a cause, and really, it should perhaps be double checked using data for only when teams are level to see if correlation still exists

But on the point I quoted you on, regardless of game state, my guess is that the number of corners may correlate better with teams which use width and cross the ball a lot, and (probably to a lesser extent), teams which shoot from 20-30 yards a lot. Teams which carve out more chances in the box from through balls will tend to get a differently-slanted set of results (goal, miss, offside, some corners of course).

If we also take evidence that both long shots and crosses are less efficient at creating good chances than intricate approach play and through balls, then it would be reasonable for teams with a higher emphasis on inefficient approach - and therefore lower success rates, i.e. wins - get a disproportionately high number of corners.

[The Woolster]

But on the point I quoted you on, regardless of game state, my guess is that the number of corners may correlate better with teams which use width and cross the ball a lot, and (probably to a lesser extent), teams which shoot from 20-30 yards a lot. Teams which carve out more chances in the box from through balls will tend to get a differently-slanted set of results (goal, miss, offside, some corners of course).

If we also take evidence that both long shots and crosses are less efficient at creating good chances than intricate approach play and through balls, then it would be reasonable for teams with a higher emphasis on inefficient approach - and therefore lower success rates, i.e. wins - get a disproportionately high number of corners.

We were the team with the most corners last season, and this season I think it is Man City. An inefficient strategy may well lead to a higher proportion of attacks ending in corners, but I do think it is the teams that attack the most (probably best measured by total shots) who probably get the most corners in total

[ElstonGunn]

The only true way of measuring the ability/skill of the striker at managing these situations would be to measure the accuracy of shots as a % error.  This would require the researcher to know where the striker was aiming.  Measuring the outcome alone is quite misleading as the outcome is affected significantly by external factors.

Granting everything else you said, the question for me isn't really about how to measure finish ability--I'm happy to grant that measuring outcomes can't do that with precision. The pertinent question for an analyst is, how many goals can I reasonably expect this player to score (at my club, in my system, etc.)? Here's where we get into fairly conceptual issue, but the data on the lack of correlation from year to year in G/ExpG indicates (though, of course, doesn't prove) to me that even if you somehow knew that your striker was an elite finisher, you still wouldn't be able to reasonably expect that striker to score more goals in the next season than an otherwise identical striker with merely average finishing ability. That's an important finding.

[PhaseOfPlay]

Granting everything else you said, the question for me isn't really about how to measure finish ability--I'm happy to grant that measuring outcomes can't do that with precision. The pertinent question for an analyst is, how many goals can I reasonably expect this player to score (at my club, in my system, etc.)? Here's where we get into fairly conceptual issue, but the data on the lack of correlation from year to year in G/ExpG indicates (though, of course, doesn't prove) to me that even if you somehow knew that your striker was an elite finisher, you still wouldn't be able to reasonably expect that striker to score more goals in the next season than an otherwise identical striker with merely average finishing ability. That's an important finding.

So let's take Suarez. If we can work out where Suarez takes his shots from, and find a comparable striker in terms of position of shots, but who scores less goals, is this model suggesting that next season we wouldn't be able to have any reasonable expectation that Suarez would score more goals?

[ElstonGunn]

So let's take Suarez. If we can work out where Suarez takes his shots from, and find a comparable striker in terms of position of shots, but who scores less goals, is this model suggesting that next season we wouldn't be able to have any reasonable expectation that Suarez would score more goals?

Suarez is maybe a bad example, but yes. I think you'd have trouble finding that comparable striker though. Suarez is one of the couple best players in the world because he has an extraordinary ability to take huge volumes of good shots and create them for others.

He's the poster boy for how large the swings in finishing outcomes are though. In 2011-12 he was a historically bad finisher. This year he's been a historically good finisher.

Here's one model's representation of how likely an average finisher would have been to score at least as many goals as Suarez in 2011/12:



(i.e., almost 100%)

And the same for earlier in 2013/14:



(Almost 0%)

Note again, that the claim I made becomes less true as you increase the volume of shots. So if you absolutely knew you had an elite finisher, you could reasonably predict he'd score a small number of extra goals over the average finisher over something like a 3 year period.

[PhaseOfPlay]

The thing that worms its way into my thinking, though, is "What has changed between 2011-12 and 2013-14 for Suarez?"

And this is where I have trouble with the concept, because it seems to leave an obvious crucial issue out?

[ElstonGunn]

The thing that worms its way into my thinking, though, is "What has changed between 2011-12 and 2013-14 for Suarez?"

And this is where I have trouble with the concept, because it seems to leave an obvious crucial issue out?

It doesn't completely leave it out. I think there obviously some things that changed, some of which are almost definitely missed by the ExpG calculation (like Suarez not having to both create and finish as many chances, which goes to your earlier point about number of touches before the shot). However, I asked Colin at the time what Suarez' ExpG2 was (which is based on both shot position and then where it hit the target) and there was still a very large (though not as large) gap between the expectation and the actual goals scored. And, regardless, the change is just so stark that I find it impossible not to the think there's a very large chunk of simple random variation behind it. He literally was the worst finisher in England under Kenny (by outcomes), and is by far the best this season.

[PhaseOfPlay]

It doesn't completely leave it out. I think there obviously some things that changed, some of which are almost definitely missed by the ExpG calculation (like Suarez not having to both create and finish as many chances, which goes to your earlier point about number of touches before the shot). However, I asked Colin at the time what Suarez' ExpG2 was (which is based on both shot position and then where it hit the target) and there was still a very large (though not as large) gap between the expectation and the actual goals scored. And, regardless, the change is just so stark that I find it impossible not to the think there's a very large chunk of simple random variation behind it. He literally was the worst finisher in England under Kenny (by outcomes), and is by far the best this season.

I see what you're saying. But my thinking is that we had a major tactical shift between Kenny and Rodgers, and I think that played its way into increasing his scoring totals, because as I was saying earlier in the discussion - the final ball into the forward is as important as the strike itself. To repeat an old coaching adage - the pass has to allow the receiver to make their next move with as little fuss as possible. People think that a pass is a pass, but so much more goes into it that the wrong pass into a striker in the right position can mess with their statistical chances more than a shot from an undesirable position (possibly). So in other words, although Aspas' numbers were from "bad" positions, it's possible that Celta Vigo's attack was designed around getting him the ball in the way he prefers it, whereas at Liverpool now, he's getting the final ball in a different way, and it affects his ability to finish. Similarly, under Kenny we were WAY more direct than we are now, with more crosses, long balls, and more direct play - and this possibly doesn't fit into how Suarez needs to play the game. It's no surprise to me that with the new counter-attacking emphasis at the moment, that his scoring total has gone down while Sturridge's has remained much the same. On the other hand, when Sturridge was out, and we played more to Suarez's strengths, he was knocking them in for fun. So my query about the model (and I DO like the model) is that there is a lot more going on in goalscoring than positioning alone, and I wonder if the model takes this into account - and if not, how could it incorporate these elements if they are not (yet) statistically identifiable in discrete situations such as goalscoring?

[iamrobk]

I see what you're saying. But my thinking is that we had a major tactical shift between Kenny and Rodgers, and I think that played its way into increasing his scoring totals, because as I was saying earlier in the discussion - the final ball into the forward is as important as the strike itself. To repeat an old coaching adage - the pass has to allow the receiver to make their next move with as little fuss as possible. People think that a pass is a pass, but so much more goes into it that the wrong pass into a striker in the right position can mess with their statistical chances more than a shot from an undesirable position (possibly). So in other words, although Aspas' numbers were from "bad" positions, it's possible that Celta Vigo's attack was designed around getting him the ball in the way he prefers it, whereas at Liverpool now, he's getting the final ball in a different way, and it affects his ability to finish. Similarly, under Kenny we were WAY more direct than we are now, with more crosses, long balls, and more direct play - and this possibly doesn't fit into how Suarez needs to play the game. It's no surprise to me that with the new counter-attacking emphasis at the moment, that his scoring total has gone down while Sturridge's has remained much the same. On the other hand, when Sturridge was out, and we played more to Suarez's strengths, he was knocking them in for fun. So my query about the model (and I DO like the model) is that there is a lot more going on in goalscoring than positioning alone, and I wonder if the model takes this into account - and if not, how could it incorporate these elements if they are not (yet) statistically identifiable in discrete situations such as goalscoring?

If nothing else, could you not say that this is statistical evidence that Rodgers system works better than Kenny's?

[PhaseOfPlay]

If nothing else, could you not say that this is statistical evidence that Rodgers system works better than Kenny's?

Possibly. It definitely works better for Suarez, for sure.

[ElstonGunn]

I see what you're saying. But my thinking is that we had a major tactical shift between Kenny and Rodgers, and I think that played its way into increasing his scoring totals, because as I was saying earlier in the discussion - the final ball into the forward is as important as the strike itself. To repeat an old coaching adage - the pass has to allow the receiver to make their next move with as little fuss as possible. People think that a pass is a pass, but so much more goes into it that the wrong pass into a striker in the right position can mess with their statistical chances more than a shot from an undesirable position (possibly). So in other words, although Aspas' numbers were from "bad" positions, it's possible that Celta Vigo's attack was designed around getting him the ball in the way he prefers it, whereas at Liverpool now, he's getting the final ball in a different way, and it affects his ability to finish. Similarly, under Kenny we were WAY more direct than we are now, with more crosses, long balls, and more direct play - and this possibly doesn't fit into how Suarez needs to play the game. It's no surprise to me that with the new counter-attacking emphasis at the moment, that his scoring total has gone down while Sturridge's has remained much the same. On the other hand, when Sturridge was out, and we played more to Suarez's strengths, he was knocking them in for fun. So my query about the model (and I DO like the model) is that there is a lot more going on in goalscoring than positioning alone, and I wonder if the model takes this into account - and if not, how could it incorporate these elements if they are not (yet) statistically identifiable in discrete situations such as goalscoring?

Our point of disagreement isn't that those things matter (I certainly agree that they do), but how much they matter. Team ExpG correlates very very well to goals scored, IIRC, which indicates that it captures most of what goes into a goal. Like I said, it does incorporate a few things other than positioning (i.e., whether the pass was a throughball, a cross, whether it was taken with the head or feet), but it does miss a lot of other things. That the correlation with actual goals scored continues to be high, however, indicates that those other things aren't likely to produce a difference like you see with Suarez from 2011-12 to 2013-14.

[ElstonGunn]

Possibly. It definitely works better for Suarez, for sure.

Another thing to add here is that Suarez is better now not only in terms of finishing but in ExpG too. There's no doubt from me either that Rodgers' has gotten more out of Suarez than Kenny did.

[BabuYagu]

The only explanation I can think of just now is that socre effects is playing a role, as the losing team tends to take the greater % of shots, and perhaps the type of shots they tend to take when losing, and the way the winning team defends when winning, might see an increase in the number of corners for the losing team.

Also teams who rely on set pieces (eg - Pulis' Stoke) would actually play for corners & throw in as wingers would wait until they were closed down then kick the ball off shins a lot. For most teams a corner is a result of a saved/blocked shot but for some it is seen as the outcome of a successful attack.

[The Woolster]

Also teams who rely on set pieces (eg - Pulis' Stoke) would actually play for corners & throw in as wingers would wait until they were closed down then kick the ball off shins a lot. For most teams a corner is a result of a saved/blocked shot but for some it is seen as the outcome of a successful attack.

This is true, but if 2 or 3 teams are using those tactics, whilst it would affect the correlation, I am not sure if it would be enough to be the over riding factor

[The Woolster]

The 2nd presentation write up from the Opta Forum is now available

http://www.optasportspro.com/about/optapro-blog/posts/2014/blog-the-optapro-analytics-forum–-long-range-shooting-as-an-option.aspx

[JackWard33]

Another thing to add here is that Suarez is better now not only in terms of finishing but in ExpG too. There's no doubt from me either that Rodgers' has gotten more out of Suarez than Kenny did.

There's some evidence that he isn't better at finishing and in fact that finishing as a whole isn't a particularly important factor in how many goals a striker will score (ie there isn't that much difference between players in finishing ability - or nowhere near as much as we think)
In fact its number of shots and location that is the biggest correlation with number of goals a striker will score
See this article

http://differentgame.wordpress.com/2014/02/20/luis-roo-and-bob-too/#comment-1203


It does seem that there is such a thing as 'finishing skill' but that it is nowhere near as pronounced between players and that it may not be quantifiable in a useful way
(see this for more detail http://cartilagefreecaptain.sbnation.com/2013/12/3/5169884/shot-matrix-v-identifying-player-shooting-skill)

The basic problem is that a season of shooting (finishing) simply isn't enough to get a reliable conversion rate or measure of finishing skill ... in fact an entire players career might still be subject to variance depending on how much they shoot

People constantly view results as a true picture of reality and don't like to attribute it to statistical variance (ie luck)

Earlier it was asked whats the main difference between Suarez 11-12 and Suarez 13-14
The two key differences are i) he's taking more shots (over 6 a game compared to 4 a game) and ii) he's run WAY above expectation this season and ran below it in 11-12 (ie he was unlucky 2 years ago and has been lucky this year)
These are much more likely explanations than him suddenly becoming twice a good a 'finisher' as he was 2/3 years ago

[ElstonGunn]

There's some evidence that he isn't better at finishing and in fact that finishing as a whole isn't a particularly important factor in how many goals a striker will score (ie there isn't that much difference between players in finishing ability - or nowhere near as much as we think)
In fact its number of shots and location that is the biggest correlation with number of goals a striker will score
See this article

http://differentgame.wordpress.com/2014/02/20/luis-roo-and-bob-too/#comment-1203


It does seem that there is such a thing as 'finishing skill' but that it is nowhere near as pronounced between players and that it may not be quantifiable in a useful way
(see this for more detail http://cartilagefreecaptain.sbnation.com/2013/12/3/5169884/shot-matrix-v-identifying-player-shooting-skill)

The basic problem is that a season of shooting (finishing) simply isn't enough to get a reliable conversion rate or measure of finishing skill ... in fact an entire players career might still be subject to variance depending on how much they shoot

People constantly view results as a true picture of reality and don't like to attribute it to statistical variance (ie luck)

Earlier it was asked whats the main difference between Suarez 11-12 and Suarez 13-14
The two key differences are i) he's taking more shots (over 6 a game compared to 4 a game) and ii) he's run WAY above expectation this season and ran below it in 11-12 (ie he was unlucky 2 years ago and has been lucky this year)
These are much more likely explanations than him suddenly becoming twice a good a 'finisher' as he was 2/3 years ago

Yes, I fully agree. You may want to read back the last several pages though to see some very forceful counterarguments!

[PhaseOfPlay]

Yes, I fully agree. You may want to read back the last several pages though to see some very forceful counterarguments!

[JackWard33]

Yes, I fully agree. You may want to read back the last several pages though to see some very forceful counterarguments!

Hah! Yeah I've done this after posting

People are always going to be resistant that stats that don't fit in with a narrative
It's an even bigger problem when all analytics can provide as a conclusion is ..... 'well it's probably down to variance'
Most people hate HATE luck being the answer - you can't construct a narrative like ('suarez has become one of the deadliest finishers in the world') around statistical variance; 'Suarez is running hot this year' isn't exciting

[ElstonGunn]

The top 4 race is over:

Michael Caley:



Colin Trainor:

[robgomm]

Perhaps backs up my feeling we'll finish third though I'm still hopeful of that outside title shot

[drmick]

The top 4 race is over:

Michael Caley:



Colin Trainor:

Would like to know he Michael Caley works those % out. It looks like a smile ppg projected forward. Am pretty sure the relegation stats are well off!

[ElstonGunn]

Would like to know he Michael Caley works those % out. It looks like a smile ppg projected forward. Am pretty sure the relegation stats are well off!

Easy enough to find the methodology at cartilagefreecaptain. He's the only one who publishes all the details of his model.