Tuesday, September 13, 2011

Unlocking the Key Pass

A couple of weeks back, we looked at the correlation between shots on goal and goals scored and (unsurprisingly) found a pretty strong link between the two. This ground breaking research probably won't make it into the next Freakonomics book but it might help illustrate players who are getting good looks at goal but not converting their chances. I went on to mess up my explanation of regression to the mean (h/t to mustardking for correcting me) but in essence, while players who have yet to capitalise on a number of shots on target, don't suddenly find themselves 'due' a goal, if a player continues to get in good positions and take good shots, over a period of time we would expect them to score at their historical rate going forward (normally around a goal every three shots on target for elite players).

Building on the shots analysis, we can now move onto to trying to predict assists. This is inherently trickier as the inputs we need to use will almost certainly involve some kind of judgement. 'Shots on goal' is a pretty unarguable number (I saw 'pretty' because there could be some issues with deflected shots, tame shots etc) and the number is provided by a number of sources. For assists though, we need to look at things like 'key passes', 'successful crosses' and 'through balls' which could be interpreted differently by two different people watching the same game. Nevertheless, we deal with what we have, and I will be using Opta data for all the below analysis, except where otherwise noted.

Trend Spotting
Opta provides a number of passing stats which can look at for signs of correlation with assists. Using data from the 2010/11 season, we get the below correlations:
  • Total passes made 45%
  • Successful dribbles 49%
  • Successful passes in opponents half 53%
  • Successful cross 69%
  • Total crosses 70%
  • Key passes made 73%
The first three are clearly, and unsurprisingly, not too useful and with the lack of distinction between successful and total crosses am I going to categorise these as a single item. That leaves us with crosses and key passes made, each of which are solid indicators of assist success. For now I am going to focus on key passes as this metric will include all player types, rather than leaning towards wide players or players from teams who play a certain formation or style.

Defining Average
Over the course of the season we are going to identify players whose key passes are not being converted at an 'average' rate and thus should see some improvement in the future. For example, let's imagine that after 15 gameweeks Chris Brunt has played 30 key passes but only has 1 assist. At that point he wouldn't be having a great fantasy season and would likely be hovering below the radar. However, the low rate of conversion is probably unsustainable, and while we we can't say that future chances will be converted at a greater rate to 'equal everything out', we can expect future chances to be converted at an average rate, and thus Brunt would be more valuable than he had been to date. But what exactly is the 'average rate'? 

Last year, Opta tells us that there were 4,941 'key passes' made by midfielders and forwards with 516 of these converted into goals (thus generating an assist). That gives a league wide conversion rate of 10.4%. This is a good start, but not all teams are equal. Slipping a ball through to Wayne Rooney and Javier Hernandez (who converted 121 shots in 24 goals) is very different to passing to Huge Rodallega and Franco Di Santo (who converted 122 shots into 10 goals) and some account needs to be taken for this. The conversion rate of 'key passes' to assists for each team last season were as below:
  1. Man Utd 16.4%
  2. Arsenal 13.1%
  3. Man City 12.3%
  4. Blackburn 12.0%
  5. Everton 11.4%
  6. West Brom 11.4%
  7. Wolves 11.2%
  8. Bolton 11.0%
  9. Newcastle 10.9%
  10. League Average 10.4%
  11. Fulham 10.2%
  12. Stoke 9.8%
  13. Aston Villa 9.2%
  14. Liverpool 8.6%
  15. Sunderland 8.4%
  16. Tottenham 8.1%
  17. Chelsea 7.9%
  18. Wigan 6.5%
While these numbers clearly vary considerably, we must remember that we are dealing with small sample sizes and not overreact. Only the top and bottom two teams lie more than a single standard deviation away from the mean so I think we should be fairly comfortable using the ~10% conversion rate as a league wide benchmark (taking care to note that, generally, good teams will have higher rates).

The Season So Far
It's obviously early to start making assumptions, but let's take a quick look at who has enjoyed the most key passes to date, and how this has translated to assists. Where this isn't the case, it's likely that these players are currently being undervalued and might be decent differentiators for the coming weeks:

Player
Key Passes
Assists
Conversion Rate
Lampard
15
2
13.3%
Murphy
14
2
14.3%
Silva
11
2
18.2%
M Petrov
11
2
18.2%
Nasri
11
4
36.4%
Nani
11
2
18.2%
Watson
10
0
0%
Faurlin
10
0
0%
Ramsey
9
0
0%
S Petrov
8
0
0%
J Morrison
8
0
0%
Larsson
8
0
0%
Duff
8
0
0%
Arshavin
8
0
0%
Agustien
8
0
0%
Taarabt
7
0
0%
Sessegnon
7
0
0%
Arteta
7
0
0%
A Young
6
4
66.7%
Mata
5
0
0%
Wright Phillips
4
0
0%
Milner
4
2
50.0%
Meireles
4
2
50.0%
Malouda
4
0
0%

This is not an exhaustive list of course but I have tried to highlight those players with interesting numbers to date:

  • It is interesting to note that the league wide conversion rate for the season to date has been 10.7%; very close to the comparable number last season (10.4%)
  • This doesn't categorically prove anything, but it is another piece of evidence suggesting that Silva may still be the choice pick over Nasri. Nasri has more assists to date but such a high conversion rate looks unlikely to continue all year. You can argue that maybe he plays better quality 'key passes' than Silva but playing the numbers alone, their value looks very close with regards to assists but Silva enjoys a goal scoring advantage.
  • Those players from Watson down to Arteta have been playing dangerous balls but have yet to have one of their gems converted by a teammate. Considering that we would have only expected a single assist, let's not get carried away and suggest they will enjoy huge seasons from here on,. That said, this might serve as a useful tie breaker for a player like Ben Watson who has also assumed penalty duties and logged a few shots on goal already this season.
  • Ashley Young has been great since signing for the Champions but that 67% conversion rate is unsustainable and unless he starts delivering more key passes, we would expect his assists to slow down a tad over the coming weeks as the conversion rate normalises.
  • I've been pretty down on Chelsea over these opening weeks but this data suggests Lampard will continue to generate assists, which coupled with his penalty duties, might just give him some value this year. I added Mata to this list just because his 5 key passes have been generated in just 96 minutes of playing time. If he keeps averaging close to 5 key passes a game, the assists will surely start to come his way very soon. Finally for Chelsea, Malouda led the league by some distance last season with 113 key passes (Henderson and Downing were second with 78) but he's managed just 4 in 217 minutes this year.
  • Shaun Wright Phillips looked very dangerous tonight for QPR and this is reflected in these numbers with 4 key passes in just 88 minutes. At 6.4m he is too expensive to be a bench player so we probably need to see some QPR goals before we buy in, but the potential for a return to prominence is there (playing alongside Barton and Taarabt, who was excellent in the first half, will also help).
I will continue to track these numbers as the season progresses and will highlight players with high 'key pass' numbers given the strong correlation with assists. Continue to check back this week or follow on Twitter @plfantasy as I plan to post this week on how to deal with the two Manchester teams along with the weekly look at the chalkboards. Thanks for reading and please post comments/suggestions/questions below or over on Twitter.

19 comments:

Mitchell Stirling said...

Maybe worth noting two of Young's assists were for nudging the ball a yard to Rooney vs Arsenal

S M said...

Is an assist always a key pass? I assume so, but not sure if it was spelled out specifically.

For example, assists are sometimes rewarded if someone scores from the rebound on your shot, or deflected passes etc. Do these count as a key pass?

Might mean assists are slightly overestimated at times.

Totally Frank said...

Is it worth pointing out that Silva has played 348 minutes while Nasri only played 210? Small sample size but according to that Nasri has a key pass every 19,1 minutes while Silva has one every 31,6. Think it's to early to draw conclusions and since Silva seems to be firing more shots and has overall been fantastic (especially the games he has started without Nasri) he might be the choice.
Another question: Is the succesful crosses included in the key passes category? Otherwise two of Nasris assists where crosses for Dzeko which would still leave him at 2 assists in 11 key passes, the same as Silva.

Thanks for a great site by the way ;)

Simon G said...

The City situation is difficult to say the least. I have Silva and Dzeko. Waht I would really like to know is whether Dzeko is going to gety any more playing time soon? Is Dzeko and Aguero up front not a workable solution, they have been the two in form strikers (and Tevez's poor penalty can't have done him any favours). I'm paniking and thinking about swapping Dzeko out for Aguero - but this will mean selling someone else to pay for it - so at least a 4 point hit. City have another ok fixture this week that could see them score goals - though probably not as many. Any advice?

Leo said...

Reckon Brunt was just being rested after international duty

http://www.wba.co.uk/page/MatchReport

Nibbsy said...

Great blog, and another really interesting article, thanks.

As a Baines owner, I'd be interested to know what his stats are like. I think you're basically buying him as a 6th midfielder who can also get cleansheets, so I'd be interested to see how he compares.

Also, are there stats available for how many key passes Nasri made for Arsenal last season? That would be interesting to see as he finished the season on 1 assist I think.

Chris Glover said...

Excellent point Mitchell, I forgot about those two. That makes his conversion rate more normal but still leaves us with questions about whether he is playing enough 'key passes' to deliver consistent assists over the season

SM - probably not but for simplicity's sake I am assuming they are. I am only trying to get approximate indicators, hence why I wouldn't now conclude that Nani > Young for the rest of the season purely based on this data.

Frank - That's a good point and I thought about tracking key passes per minute but for now I think it's too early. I too wondered wondered whether a successful cross can be both a cross and a key pass, but have not yet been able to find the answer. I am still just playing around with the data for now, so I should probably emphasize that people take these findings with a pinch of salt.

Simon - I plan to post a separate piece on City soon to try and deal with those questions (though honestly they will be educated guesses as I'm unsure too).

Leo - I saw the same thing with Hodgson suggesting Brunt and Morrison were both tired from the internationals. Normal service should resume this week.

Nibbsy - I do indeed need to look at full backs and will probably focus there on crosses (though will throw key passes in too) as this is the source of most of their value. Baines has 11 key passes and is 3/18 on crosses so far this year having racked up 11 assists on 63 key passes last year.

Nibbsy said...

Thanks for the info. That's interesting, Baines' stats sound excellent for a defender. I'll look forward to reading your piece on defenders.

James Richard Klien Jr said...

I think there is far too much ‘noise’ here for this to be a useful stat. Take the Young example, his assists are invariably from crosses. If the cross is good, it will more likely result in a goal, if not, it will end up in row Z and won’t be categorized a “key pass” thus bypassing this stat.

I love the thinking behind some of your more simplified analysis, such as identifying defenders with offensive points priced in. It takes a Black Swan/Options approach. Buying Baines is like buying Jagielka and you pay the price differential for an option on an Everton winger (you bag the offensive points if they come to fruition, if not, you lose the cash). How about using some more financial-esque analysis to explore potential “fat tails” (e.g. 80% of a players points coming from 20% of his games, or a disproportionate role of a rare event – e.g. Man Utds’ 8-2 drumming of Arsenal!) I imagine most players use a more Gaussian based system to team selection, pursuing the players with the highest total points or highest average points per game (even though points are seldom earned in such a linear fashion!). Think Muamba’s popularity after week one!

Chris Glover said...

James - I wouldn't suggest any stats on this site are 'noise free' as I'm working with limited data, time and, honestly, expertise.

On your specific point, I'm not sure I understand why this is an issue. A key pass is classified as an 'attempted assist' so the outcome of a pass has no bearing on its classification. I don't really see how a cross missed by Rooney is different to a through ball missed by him.

Crosses are shown separately in Opta stats so I openly concede that wingers may well be underrated in this list, but that's okay. If you analysed forwards based on goals last year, Drogba would have been average, but if you take his assists into account he becomes elite. This post is just a single piece of the puzzle, not a sure fire answer to picking midfielders.

As my initial paragraph explained, the correlation between crosses/assists and key passes/assists is pretty similar, though neither are strong enough to suggest this stat is perfect.

I like the idea about looking at players who've generated a high proportion of their value from a couple of games, and I started tracking this last year using basic standard deviation variance. I will try and take a look at this again this year. Thanks for the post.

James Richard Klien Jr said...

Chris thanks for the response. My thinking was that a good cross would have a higher conversion rate, whereas a poor cross will never result in a chance on goal i.e. a higher degree of variation as compared to passes in the final third. However as it is “attempted assist”, you’re right, this would have no baring.

I tried a similar method late in the season last year whereby I disregarded a players highest couple of game weeks (the outliers) when assessing a players potential so I wasn’t lulled into a Muamba trap! I also use a similar reasoning in balancing team selection - a somewhat 80/20 rule. The 80% being 10-12 players in my squad that are guaranteed starters and dependable point earners. The 20% being high risk gambles (Dzeko is as likely to score double digits as he is to score a singular appearance point). As a whole the team is balanced, I’m protected from a calamitous week but still being exposed to potential windfalls. That’s the logic anyway. Not always easy to apply.

Anyways love the website. Keep up the good work!

Mell 10-11 said...

So Chris, does captaining Roon essential this week to play catch up? Other obvious choices are on away assignment...Obviously most of us getting burn by not captaining him not the overall performance of the team

Steve said...

Chris, I have both Rooney & RVP who appear to be the most favoured (C) picks for GW5, but I'm looking for some kind of stats to help me decide between the 2.

I seem to remember RVP having a very good history vs Rovers?

David J. Pereira said...

Boas!

Devo dizer que gosto imenso deste blogue!

Podem adicionar os meus aos vossos links? Eu prometo que retribuo :p

http://davidjosepereira.blogspot.com/

Saudosos cumprimentos!

amtosh said...

Hey Chris,

I was just making my weekly transfer when I realised that you don't get to sell the players at their current market price. How much bullshit is that?

For example, to buy Rooney at the moment it'll cost me 12.4, but when I transfer him out I only get 12.2 (which is what I bought him for).

This pisses me off as it negates any advantage to buying a player cheap.

My team's apparent value at the moment is 101.6 m, but if I sell them I'll only get 100 m.

Chris Oien said...

Amtosh, that isn't quite right. When you sell a player whose price as risen, you get to pocket half the rise in value. If you buy at 10 and the player rises to 10.4, you sell for 10.2. Your Rooney example makes me think you actually bought him at 12.1 or just plain 12. Either would be consistent with now being able to sell him for 12.2

x00x said...

Thoughts on Benayoun? I am considering swapping Bent ---> Odemwingie which frees up some cash I can hold or take -4 to upgrade moses to Benayoun.

Although of course the moment I move Bent he'll score a hat tricK !

barraclough youths said...

Hi Guys

I payed 4.5M for all my defenders, but do not have the extra chash to upgrade 1 to 5M

S Taylor, Wes Brown, Ward, Woodgate, Zat Knight

Problem with Zat Knight, looks like he will not start every week

What are your thoughts on swapping him for Robinson or Steinsson???

Also got a problem with Woodgate, really want to keep him because Stoke are awsome, but its not sertain that he will stay fitt every week to score points, thinking maybe replace with a QPR or WBA defender, thoughts please????

Thanks ;)

Gummi said...

Just wanted to pop in and say thanks. Excellent analysis.

Regarding the "fat points" issue, this was exactly the situation with Berbatov last season, with most of his points coming from his high-scoring home games. Made for a frustrating experience for some (i.e. me) having transferred him out before his 5 goal haul.