Championship At Best!

Lately, I've been trying to come up with a defensive equivalent to the goals created statistic. Ideally, I would like to be able to add the two, and come up with an overall value of each player. I'm not quite there yet, but I've found something that makes sense for defensive contribution. When looking at Marginal Goals, we can determine how many goals (compared to a fictional "marginal" team) we prevented. With the percentage of shots saved by the goalkeepers, we can break it down further to see how many goals were prevented by the keeper, and how many were prevented by outfield players.

In the 06/07 season, I rated our outfield defence 15.768 goals better than the marginal team (think Sunderland of 05/06). How do we divide this up among individuals? The most fair method I've come up with this equation:


Here, we have MGP for marginal goals prevented; DA as defensive actions - including all clearances, blocks, interceptions, tackles, and offside provocations; and M as minuses - the number of goals conceded with a player on the pitch. The same variables with a lower-case t are the team totals for each. This gives us the following table for 06/07:

So what this means is that, by having Aaron Hughes start at CB instead of a generic player from the 05/06 Sunderland team, we prevented almost 3 goals. Brede Hangeland and Dejan Stefanovic still managed to prevent almost 2 goals apiece in their limited time. I think that all of this tells us that there isn't a huge difference between our top 4 CBs, but the difference does amount to about half a win, or 1-2 draws. As we have found out, that can be the difference between staying up and being relegated.

Another thing I found interesting is how this relates to our current central midfield dillema. I predicted that Danny Murphy and Jimmy Bullard will finish this season with almost exactly the same total of goals created, with roughly the same number of minutes. In defence, however, they differ quite a bit. Murphy, while not an idea defensive midfielder, was still ahead of Bullard. Leon Andreasen, projected over a full 3000-minute season, would have been more than a full goal better than Bullard. Would that goal make up for the loss in attacking ability? Or would Andreasen's presence allow Murphy and the rest of the team to attack more effectively? It's all difficult to say, but we're getting a little bit closer to understanding just what it means to replace a player in the lineup. The fun begins when we can look at marginal goals created. That's (hopefully) coming soon, and it should give attack and defence statistics a common ground, a way to say "Player A is better than Player B".

The past couple of days have been spent gathering data on all of our players, plugging it all into the computer, and telling it to figure out how everybody is going to do this season. I've talked about goals created a few times, which works on the assumption that each goal can be credited half to the goal scorer, and half to playmaking. Not all goals are set up by a second player, so this actually ends up being a little more than half of a players goals + assists, and differs from team to team. Another nice thing about this stat is that it allows us tp pull one player from one team, drop him in a new team, and determine how much better a player is in attack than another. For example, a weighted average of Bobby Zamora's last three seasons would project him at 1448 minutes this season with just over 4 goals created. The player he replaced, Brian McBride, created 2 goals in 1206 minutes. If you put each player in a vacuum (meaning that playing time is determined by only the player himself, and not the team's depth chart, squad rotation, etc.), this would indicate that Zamora is a two goal improvement over last season's version of McBride.

If I project all of our new signings the same way, and subtract the players we have lost, we're left with a net gain of about 7 goals. The raw numbers were actually higher than that, but I adjusted for minutes played - the Ins were projected to play quite a bit more than the Outs. Compared to last season, that would put us at 45 goals. In addition, several players underperformed in 07/08 and could bounce back somewhat. Danny Murphy, despite his late season heroics, was better in each of his previous two seasons. Jimmy Bullard missed at least half the year. David Healy, Paul Konchesky, Steven Davis, Seol Ki-Hyeon, and Collins John were also below their normal levels. Simon Davies and Diomansy Kamara, unfortuntely, are candidates to regress slightly. All in all, though, we should expect a slight improvement overall from our existing squad. My projection is 3 extra goals from our remaining 07/08 players, putting us up to 48 total.

Just for fun, I broke down every player's predicted GC into goals and assists. To do this, I used a league average goals/assist ratio, and determined that a 48 goal scoring team should pick up around 33 assists total. Then, I took each player's individual G/A ratio and broke down their GCs into goals and assists. Here's what I came up with:

Some might be surprised to see our top two striker's goal totals so low. My numbers suggest that Andy Johnson is going to play quite well for us, but his projected playing time is on the conservative side. A few things went into determining this - first, Johnson has had some injuries that limited his minutes at Everton. Second, recent Fulham teams have rotated their strikers quite a bit. Last season, none of them (aside from Dempsey, who also played in the midfield) saw more than 2000 minutes. I went into a bit more detail about Zamora yesterday, but the biggest issue with him is that he simply wasn't all that good last year. What the calculator doesn't see, however, is that injuries played a big part of that. If Zamora is fit to start 30 games this season, he could be our best player. Between those two and Erik Nevland, that does not leave a whole lot of playing time for Healy, Eddie Johnson, and Collins John. Diomansy Kamara is scheduled to resume training in December, which would likely put him back on the pitch by January.

No big surprises in the midfield. Bullard, Murphy and Davies are our best attacking options, and Zoltan Gera is not far behind. It may prove difficult to fit all four into the lineup at the same time, but we will try. Davies and Gera will see the most minutes, as they can play on the wings, with Bullard and Murphy being replaced at times by the more defensive players Andreasen, Andranik and Smertin.

Obviously, this table doesn't tell us a whole lot about defenders. The fullbacks can be expected to pick up a few assists on crosses, but these players generally aren't expected to attack much. Their real contribution is in how many goals they can prevent. Our friend Rich expects a 6 goal improvement over last season, and I think he's right. LB and one of the CBs is unchanged. We should see some improvement with a full year of Brede Hangeland in the other CB slot. Fredrik Stoor and John Paintsil are probably an improvement over first-half RB Chris Baird, but might not play quite as well as January loan signing Paul Stalteri. We don't know a whole lot about the new RBs, but it is safe to say that they should be at least equal to last season's players as a whole.

That's it for now! My current mission in life is to squeeze all of our defensive metrics into a quantifiable number of number of goals prevented, much like we have done already with goals created. Until then, we have our educated guesses. 48 goals scored and 54 conceded should leave us in a comfortable mid-table position. With a bit of luck on our side, this could push us up into Europe, or with luck working against us, back into another relegation fight. It all begins tomorrow at Hull, and I know I'm not the only one looking forward to it.

I'm working on a much longer 2008/09 projections post, but I wanted to share something interesting. Specifically, how good Bobby Zamora was prior to last season. You can see Fulham's Goals Created leaders for last season on the right-side of this page. This gives a good picture of the attacking contribution, essentially assigning credit for each of the team's goals to individual players based on goals scored and goals assisted. Thus, adding up the total GC for each player on the team will always equal the team's goal total.

In projecting Bobby Zamora's 08/09 season, I found that he led West Ham in goals created over the 05/06 and 06/07 seasons. He was better than Yossi Benayoun and Nigel Reo-Coker in the same period, better than Carlos Tevez in 06/07, and also more productive than Dean Ashton and Carlton Cole's 2007/08, the season in which these two players surpassed Zamora on West Ham's depth chart. GC is a counting stat, so perhaps even more impressive is that Zamora did this in somewhat limited playing time.

Zamora's 7.138 and 7.833 GC in those two seasons would be good enough to place him right behind Simon Davies at #2. I had to include the fitness concerns in my final projection (which I hope to be posting today or tomorrow), so they may not look impressive, but my gut feeling is that he's going to outperform and surprise us.

Remember when I said that we didn't need Andy Johnson now that we had the equally-capable Bobby Zamora? Turns out that my point was pointless, as we now have both. The fees for both players were high, but I'm pretty sure this is a good thing.

So, who starts? We still have a good sized stockpile of attacking midfielders who probably should play, but our newly acquired high-priced talent up front suggests that we may stick with some form of a 4-4-2. Which brings us to the subject of this post - who starts, who is on the bench, who doesn't play at all?

We've already established that most strikers look a lot better off the bench than they do in the starting lineup. But do any players look a lot better? Do we have any players who are close to the same as a starter OR as a sub?

I took all of our strikers (minus Eddie Johnson - detailed MLS stats are suprisingly difficult to come by), and divided their scoring rate as a sub by their scoring rate as a starter, over the past three seasons:

This suggests that Andy Johnson and David Healy are almost three times as likely to score when brought on as a sub, while Clint Dempsey is almost equally prolific either way. Does this mean we should start Dempsey, with AJ on the bench? Not at all. What this doesn't take into account is how well each player does in each situation compared to each other. AJ may still be better than everyone else as a starter, despite being so good as a sub...if he scores once every 3 games as a starter, and every else scores once every 4 games, AJ should start. Plus, we didn't just pay £10m for a bench player. I think most would agree that AJ and Bobby Zamora are the two best strikers in the squad. As such, those two deserve to start most games. What if one of those two goes down with an injury, though? That is where the numbers above may come in handy. If we had to choose between, say, Healy and Nevland to make a spot start, Nevland might be the best choice to start with Healy coming on late as a sub.

I did have another thought in all of this, though. Plenty of people have suggested that Erik Nevland is best utilized as a "supersub" - a player who can be counted on to come in late and give us a goal-scoring threat. Nevland may or may not be the best choice for this role, but the idea is a good one. Instead of rotating players in and out of the starting lineup, how about keeping one player out of the starting rotation, but on the bench and always coming on at 60 minutes? We saw several instances last season where strikers coming on late have turned the game around, and several other instances where late defensive replacements have failed to hold a lead. This is just an observation - I have no data to back this up, but I do get the feeling that the best defence late in a game might just be a fresh attack. Especially with 7 subs on the bench, every team will have tons of attacking options available.

On an unrelated note, I'm still working on getting the missing portions of the site ready for the new season. The tricky part is moving my old, custom player database schema to Drupal's format. It's a pain to set up, but it should allow us to do some neat things once it is all working.

As you can see, this page looks a little different today. There's actually a lot more going on behid the scenes - this site runs on the excellent CMS Drupal, and I finally upgraded to the newest major release. There are lots of things that don't work right, and I know my layout looks weird in IE, but I figured that I wouldn't ever get around to fixing these things if I didn't bring the new site live.

The stats pages are broken (and, as you can see, the links aren't even there), but never fear - all of that stuff is in The Fulham Review, so you could just get order one of those...

While putting together some numbers for my previous post, I was a little curious to see what sorts of numbers were considered good, what was average, and what was not so good. I listed two different percentages, which I meant to explain a bit more, so I might as well do it here: SH % is the ratio of goals to shots taken. SG % is the ratio of goals to shots on target. A third ratio, we'll call SA % (for shot accuracy), relates the two - this is the percentage of total shots that ended up on target.

My suspicion was that SA % varies from player to player, while SG % is relatively constant. Whether or not a shot is on target is largely up to the shooter. Whether or not that shot goes in the net depends on much more - goalkeeper, the posts, defenders, etc. To find out, I added up the total number of goals, shots, and shots on target for every player in the Premier League over the past two seasons:

Season	G	SH	SG	SH %	SG %	SA %
2006/07	894	9133	4362	0.098	0.205	0.478
2007/08	924	8837	4547	0.105	0.203	0.515

The SG % didn't change much, but the SA % did. Pretty much what I expected. The league average save percentage for goalkeepers last season was just over 68%, which accounts for most of the non-goal scoring shots on target - the others being things like the posts/crossbar (especially when you hit it 3 times), blocks by other defenders, and referees who hate David Healy.

Now, if we look at individual players, the SG % actually does vary quite a bit. This suggests that some players shots are more difficult to save than others, which is definitely something I would expect, but the variance was more than I thought it would be. Below are all players who scored 5 or more goals in each of the past two seasons. In the interest of larger sample size, each player's numbers include seasons prior to 06/07, if available:

Name		SH %	SG %	SA %
McCarthy	0.250	0.422	0.593
Doyle		0.195	0.398	0.489
Viduka	0.243	0.396	0.615
Keane		0.231	0.385	0.600
Yakubu	0.202	0.368	0.550
Drogba	0.167	0.352	0.475
Adebayor	0.203	0.343	0.593
A. Johnson	0.219	0.342	0.641
Barry		0.149	0.326	0.455
Bent		0.188	0.321	0.586
Cahill	0.169	0.290	0.582
Agbonlahor	0.174	0.288	0.603
Kalou		0.163	0.280	0.581
Saha		0.145	0.265	0.547
Martins	0.127	0.256	0.497
Berbatov	0.151	0.255	0.592
Mwaruwari	0.153	0.247	0.618
Van Persie	0.133	0.246	0.540
Tevez		0.124	0.231	0.535
Defoe		0.143	0.231	0.618
Crouch	0.127	0.224	0.567
Anelka	0.136	0.223	0.611
Lampard	0.110	0.211	0.523
Ronaldo	0.101	0.203	0.499
Rooney	0.109	0.192	0.567
Gerrard	0.093	0.183	0.509

Most of these players are above the league average, but some are consistently far above that standard. Benni McCarthy, Robbie Keane, and Mark Viduka have been above 30% for just about all of their careers. Emmanuel Adebayor, Didier Drogba, Ayegbani Yakubu, and Kevin Doyle have been very close. Out of curiosity, I also looked at some leading strikers in foreign leagues. Luca Toni, Karim Benzema, Djibril Cisse, Daniel Güiza, David Villa, Alessandro Del Piero and David Trezeguet, as well as many others, are not only consistently above 30%, but have finished the season over 40% in most cases. In fact, even many of England's top scorers (Adebayor, Fernando Torres, Roque Santa Cruz, Robbie Keane) fell into this 30% or higher club. The league's top scorer, Cristiano Ronaldo, seems to suffer from a bit of Kobe Bryant syndrome - take a zillion shots regardless of quality, and you're bound to score a bunch. Interestingly, Man Utd's other leading scorers, Carlos Tevez and Wayne Rooney, seem to follow this strategy.

What is the point of all of this? I believe it can help in telling us which players are the best finishers, the best pure goal scorers. Not necessarily the best forwards - obviously, Dimitar Berbatov and Carlos Tevez are great players who excel at creating lots of their own chances - but this is a big part of the equation. We can see which players thrive when provided with quality service from the midfield. More or less, which players would do well for a team like Arsenal. With Fulham's growing stockpile of quality attacking midfield players, it would appear that this is the direction Roy is taking us. Signing players from this group would make a lot of sense.

Best of all, there are players who may be available, and in our price range! In England, Kevin Doyle is the most realistic target. He had poor season in 07/08 after a very strong 06/07, but his past history for both Reading and the Republic of Ireland suggest that last season may have been a fluke. He's probably the biggest risk of the bunch, but also the cheapest. Benni McCarthy is getting older, but he's had two very good seasons since moving to England, and likely has a few more left in him. A few others who aren't on the above list include Man City's Rolando Bianchi (.229 SH/.500 SG split between City and Lazio) and Marlon Harewood (.238/.417 in 07/08; .181/.369 career), both of whom I believe to be obtainable.

One more note - a lot of this depends on how a player is used. It seems clear that Man Utd's coaches encourage players to shoot on sight. As such, a player like Louis Saha, who I would have expected to score better on this scale, varies quite a bit from year to year:

Season	SH %	SG %
02/03		0.135	0.227
04/05		0.059	0.111
05/06		0.167	0.412
06/07		0.123	0.195
07/08		0.278	0.556

Some players vary quite a bit when changing clubs. Take Darren Bent:

Season	SH %	SG %
05/06		0.247	0.450
06/07		0.212	0.412
07/08		0.167	0.286

Tottenham seems to do things in a similar fashion as Man Utd, so it make sense that Bent's numbers last season would look more like Berbatov and Defoe's numbers (and what does that say about Robbie Keane? One could make a strong case for him as one of the more underrated players in the league). Djibril Cisse is another who looked like a completely new player after moving from Liverpool to Marseille. Again, there are other key qualities that are important for strikers, but this tells us something about who the best pure finishers are. And more often than not, those players score a ton of goals.

Major League baseball player Chris Capuano is a middle of the road pitcher, a one-time All-Star, and a frequent visitor to the disabled list. He has one thing that he does really well, and that is picking off runners at first base. With a runner on, he's often able to pull outs out of nowhere, which have saved him from sticky situations on many occasions. Several people, including his teammates, have said his pickoff move is illegal, yet he rarely gets called on it by the umpires.

Andy Johnson also has a pickoff move. He wins penalties, and lots of them. Unfortunately for Johnson, in the past could of years referees have begun to agree that his falls in the box are not legitimate, and his penalty totals have dwindled accordingly.

With news stories popping up that we may put in a bid for Johnson, there has been plenty of discussion on the message boards. In the meantime, we went out and signed Bobby Zamora from West Ham. The move was highly unpopular at first, but was met with reluctant acceptance once the deal was made official. Zamora isn't a bad player by any means, but fans were hoping for a bigger name. Perhaps an international for a high-profile footballing nation. Someone like Andy Johnson. Going by the rumors, it would likely cost over £10m in transfer fees to land the Everton striker, while Zamora's cost was about half of that. Is Johnson twice the player that Zamora is?

Since 2003/04, both players have spent three seasons in the Premier League, and two in the Championship. Good, this makes them easy to compare:

At first glance, Johnson does seem quite a bit better. More goals, better shot percentages, better strike rates. But look at all of those penalties! Fact is, they haven't been a huge factor since the 2004/05 season. Since then, his numbers have been fairly average. Let's compare the two again, this time ignoring penalties:

Suddenly, the two look very comparable. Taking into consideration Zamora's other advantages - size, ball-winning ability, other non-goalscoring attributes, and of course the magic that are INTANGIBLES - I'd even go as far as to take Zamora for the same amount of money.

Rich made a very good point yesterday that strikers will generally fare better off the bench than from the starting lineup. Indeed, this holds true for us (including Clint Dempsey's appearances through January, since he was mostly played in the midfield after that):

Last season, about 59% of all goals scored in the league came in the second half, and over 26% came in the final 15 minutes. Should teams start games with their best goalscorers on the bench, and let a McBride or Helguson try and wear down the defence for 60 minutes? That's probably not the best tactic for most games (as a majority of goals are still scored by starters), but in situations where you may be overmatched and playing for a draw, it might be a way to steal a few points.

I think I see the beginnings of a stamina rating.

Even though I've set a policy for myself to ignore any transer rumors that do not involve a manager naming a player or a player naming a team, I felt the need to do a little digging after a usually-reliable poster on TIFF named Marlon Harewood as a target. As one other poster pointed out, Harewood scored 5 goals this season, despite only starting one game for Aston Villa. I found that interesting - sounds like a very good backup/bench player, so I checked the numbers:

I'd welcome that kind of contribution to our squad. He's really only had one bad year (2006/07), and West Ham were awful that season. Obviously, I'd prefer to see him coming off the bench, but we could do worse. We all expect Roy to dig up some unknown continental player (and why not? - they are generally cheaper and better values), but Harewood seems to have dropped off of everybody's radar after rough 06/07 season, so he may fall under that same "value" umbrella. Depending on the asking price, of course - there is still a risk (is 06/07 his true form?), and thus £4m would probably not be the best value. Cut that price in half, and it's a deal.

A lot of people prefer goal differential to points as a method of evaluating the overall quality of a team. I do too, but there are a couple other ways of looking at it. I've gotten a little carried away in some new statistical analysis of the season recently, but I'd like to share some of the basis behind it.

The first few columns show the final league table as we all know it. MG represents "Marginal Goals" - more on this a bit later. Next is Real Win %. This is each team's W-D-L record represented as a percentage. For this, a draw counts as half of a win. Pythagorean Win % is the results of a simple formula used to predict how many games a team should have won:

Pythagoras didn't really have anything to do with this, but the name comes from the formula's resemblence to his theorem.

Bill James (who else?) applied this to baseball originally, and later took it a step further while developing his Win Shares statistic, which quantifies individual player contributiion to his team's record. This is where MG comes into play. Marginal Goals is the sum of a team's goal differential and the league average goals scored. In 2007/08, this was just over 50 goals. If we divide MG by 2 times the league average goals, we get another predicted winning percentage. This one generally isn't quite as accurate as the Pythagorean method, but it generally does a good job for teams who win between 30% and 70% of their games. More importantly, it gives us something to break down further and evaulate individual performance.

A "marginal" team can be described as a team of players at the lowest level of quality in the league. Think of players who would do well in the Championship, but would be lucky to make it off the bench for an average Premiership team. Sunderland of 05/06, who finished that season only 4 goals better than a marginal team, is the best example in recent memory. For most teams, 2-3 goals scored or prevented will earn another win. This season, Fulham averaged 2.007 MG per win, so every two goals we scored/prevented above that line in the sand we call "marginal" would have earned us an additional win. Manchester United, who had plenty of wins already, would need 3.6 more goals to win one more game. Derby County's negative total shows that they really didn't belong at this level, something most of us knew. Fulham, Newcastle, and all other teams who gave up points to them at any point in the season must have really been off on those days.

Right now, none of this (at least on its own) means much. The Pythagorean formula is nice to tell us where we should have been on the table, and it works for predicting future performance. Marginal Goals tell us how much better we are than a very bad, non-existent team. The real good stuff is when we use this to determine individual player value.