Thursday, September 27, 2012

Gameweek 6 Preview


I've finally got some captain data to use this week, and while it certainly isn't a finished product, I'm happy enough to share it and see where it leads in the next few weeks. The short story is that the value in the below graph is a very approximate forecast of the points for the week, based on a number of factors that correlate fairly well to future production. For more information on the actual nuts and bolts of the calculation, see below the captain commentary under the table. As a caveat for all, these forecast numbers should be taken with a pinch, nay a drum, of salt and are really intended to guide our thoughts rather than make factual predictions. That said, I believe that over the course of a season you'll be better using them than the old 'gut' feeling. I always like getting feedback from readers but as this is very much a work in progress, suggestions, ideas and criticism would be even more welcome here.

Captain Commentary
Edit: Per BBC Sport, Dimitar Berbatov is out of tomorrow's game with a hip/thigh problem

Gameweek 6 looks like a fairly tough week to forecast with no real clear cut winner coming to the forefront. I generally shy away from going too unorthodox with captain picks but this is a week where someone like a Fletcher or even Baines might make sense.

Tevez and van Persie come with significant risk given the increasing fitness of their team-mates with Tevez being further hampered by a tough fixture. Spurs at home is probably easier than it sounds, with Villas-Boas's men conceding in every game this year while also missing Assou Ekotto, Kaboul and possibly Naughton for this one. If you own van Persie you need to be captaining him more often than not if you hope to repay that price tag and this is probably a game in which you do exactly that. Rooney played 76 minutes in midweek so I would suspect he might start this one of the bench, though the nature of his injury might indicate he doesn't need to be eased in. Either way, if I had to guess I'd say van Persie starts and is well placed for success.

As for Tevez, his status is less clear and I will personally probably look elsewhere this week. Aguero and Tevez are yet to start a game together this year - though they are certainly capable of doing so - and so even if they both play we need a few hints about the impact it will have on each's value before we start captaining Tevez again without hesitation. I still like Tevez and don't see a need to panic sell just yet but the armband looks risky this week.

Everton have the best fixture for the week, making Fellaini, Baines and Pienaar viable picks for the week. Indeed, given the fixture and the fact that so much of Everton's production has been through these three players, this is the direction I would go this week. Southampton have given up a league 'leading' 59 shots inside the box so while they were able to hold Villa to a single goal last week, doing so again this week looks like a stretch. For the year Everton have already notched a ridiculous 76 shots inside the box (102 total) with City trailing in second with just 54 (82). I hate jumping on bangwagons but given the upcoming fixtures this trio look too good to pass on and it may even be one of those extremely rare cases where I might consider a 4 point hit to bring them in.

Fletcher's top ranking is driven in large part by the fact that he has accounted for 100% of Sunderland's goals to date. Needless to say this won't continue and thus you need to exercise some caution with this ranking. That said, he is facing a somewhat shaky Wigan side who can be exposed and so he's a decent roll of the dice for the week. Would I take him over van Persie? No. But I can see a situation where you take his guaranteed start over someone like Tevez who might play for 9 minutes in garbage time.

Hazard does well here based on his excellent production in those first couple of weeks, but there's no way I would captain him at Arsenal, who have been much better at the back than many expected. I'm not crazy about taking Ba on the road either, though Reading haven't been particularly great at home and have yet to notch a clean sheet this season. There still exists a non-zero chance that Ba is dropped in favour of a standard 4-5-1 formation but having scored three in the last game and a half, that risk is greatly diminished for now.

I'm concious we're getting close to the deadline now, but please send your questions to @plfantasy or post them on Facebook and I'll try to get to as many as possible tomorrow evening (Eastern Time).

Captain Stats Methodology
As noted above, this is still a work in progress so the below will change over the coming weeks, but for, the forecasted value is based on a weighted average of three factors:
  • Total P90 - scoring points is the name of the game so I felt this had to be included somewhere. I went with points per 90 minutes rather than average to avoid skewing things early in the year for a player like Tevez who suffered a 1 pointer last week in 23 minutes.
  • Home / Away P90 - some teams and players struggle on the road and thus I felt we needed an extra factor to penalise players like Dempsey from last season who was simply dynamite at home but couldn't get much going away from Craven Cottage.
  • Cr% Forecast - this one is more complicated. Essentially we are looking at the percentage of goals a player accounts for multiplied by his team's expected goals for the week. I then add a final layer to account for the fact that some players will create more than they score, earning fewer points in the process. For example, if City are forecast to score twice, and Tevez accounts for (scores or assists) 40% of their goals, he would be expected to be credited with 0.8 goals this week. If for the year he has 10 goals and 10 assists his average points per credited goal would be 3.5 (10 x 4 per goal plus 10 x 3 per assist divided by 20), so multiplying this by his expected credited goals (0.8) gives him 2.8 points for the week. Add in his appearance points (2) and his weekly rating would be 4.8. This factor is then weighted with the others to give his final score.
Next steps
The next fix is an easy one, and I only omitted it because I didn't get a chance to write the formulae before this week. We all know that if a player scores a goal, particularly a winning one, he'll be credited with bonus points in the majority of cases so this needs to be added in to the above formula somehow. I propose simply working out the average points awarded to a goalscorer over the season, possibly drilling down by position and whether or not it was the game winner. This addition should be inserted next week.

The next big step is to include an element of regression in the model to try and eliminate the situation we currently have with Steve Fletcher where he's converting almost every shot he takes into a goal and thus accounting for 100% of Sunderland's goals to date. My proposal is to add an 'expected' goals and assist number into the formulae based primarily on league averages but adjusted when a player has shown a sustained ability to convert goals at a higher rate. If, for example, Rickie Lambert had 17 shots, 12 on target and 8 goals, we might suggest that the 0.66 conversion rate is unsustainable and thus only credit him with 4 goals to account for the fact that over the past few seasons 0.33 of all shots on target have been converted to goals, with few players showing a consistent ability to outperform that average. For this analysis I will use ESPN data as (a) it is freely available and I don't want to push my luck with the Opta data the guys at FFS pay for, and (b) I can get historic rates for players to ascertain if we can really categorise them as an outlier. Unfortunately, I will have to ignore data from other leagues for now so the likes of Lambert may be undervalued a touch as we will regress his numbers to the mean where as he may well be better than that, just unproven at the top level.

What else would people like to see included in this analysis? I don't want to add anything subjective as that will just make it a numerical presentation of my own biased narrative so things like 'form' won't go in there. Please post other suggestions in the comments, on Facebook or @plfantasy and we can hopefully take this to the next level in the coming weeks. Indeed, if we can really make some progress for captain picks, there's no reason this can't be extended to help us with lineup and even transfer decisions. 

19 comments:

John Doe, 2008 said...

So you have Fulham about a half goal favorite against City? Can't say I disagree as my analysis indicates Fulham has been playing superbly, especially offensively. Agree on the Arsenal advantage as well.

Any chance you can update the upcoming weeks forecasts as well?

jonpaul3 said...

Hi Chris, I've seen on a number of posts that you refer to completed passes, shots on goal, etc. for players.

I can find basic stats on places like ESPN but nothing like you mention. I'm guessing your info comes from someone like Opta?

If not, just wondered where you got all these stats from - assuming it doesn't cost too much that is!?

Cheers, Jon

Andrew said...

Who should I bench among Cazorla, Michu, Ben Arfa, Silva, and Podolski? I need to sit one. My thought is that Michu is the man to sit because of the tough assignment against Stoke on the road, but maybe it's Silva with his inconsistent minutes?

Gummi said...

What about adding total shots (TS) and chances created (CC) in the analysis to try to capture players that haven't delivered on their statistical promise yet?

Wouldn't that help in the early part of the season, especially with unproven players like Lambert?

If Fantasy points are truly correlated with TS and CC this wouldn't make much of a difference with the current top spots. However, it would help identifying players that have simply been unlucky.

dnivy said...

Thinking about brining in a Liverpool defender. Skrtel or Johnson?

Kevin Tan said...

I need start 2 players between them,
Tevez, Guthrie, Maloney, Ben Davies.

What's your advice?

Tev - might risk with 9mins jogging.
Guthrie - Set piece chances but weak attack in Reading
Maloney - Stat show only 9chances created & 4 goal attempts in 5 games but Penalty in pocket.
Davies - is swans def coming back?

IOpsych said...

This whole week, I've been indecisive about who I should captain. I have the following:

Foster
A. Cole Baines McAuley
Ben Arfa Michu Nolan Pienaar
Berbatov Podolski Ba

I've been deciding between Ba, Berbatov, Pienaar, and Baines; leaning towards Ba or Pienaar.

Thoughts?

Crunch said...

Would it be wise to leave tevez on the bench in favor of a 5th midfielder?

I have Cazorla, hazard, Fellaini, Nolan, michu
Berbatov, tevez, podolski.

Thoughts?

Ste said...

Hi Chris, Awesome work again. I am developing a similar system. Forgive me for linking to my site but I think you should take a look and we can combine ideas. http://www.shotsontarget.co.uk/2012/09/fpl-point-forecasting.html

In a nutshell I take expected goals and then take each player and work out his % of team shots taken, and % of team SoT taken. I then adjust this by his % of S in Box vs. All shots. (shots on target in the box are 30$ more likley to result in a goal than shots in target from anywhere). I will then factor in their position. Strikers have better conversion rates than midfielders for example.

I think we are on to something to be honest and would love to develop and test out ideas.

John Doe, 2008 said...

Chris -

A couple thoughts:

1. Projected numbers are definitely the way to go. The key, of course, is to use the "right" numbers for the projections (e.g., do you use total shots or shots on target).

2. How accurate are ESPNs stats?

3. While it won't cover all teams, the League keeps good shot numbers on their website. That should cover the majority of players from newly promoted teams.

4. What are you basing your Home/Away factor on? Team performance, player performance, or both? Also, how long do you go back to determine that factor? I see that factor as adding a lot of variance if not done correctly. I might just go for a standard adjustment for all teams.

5. P90 is a great metric, but it does not factor in rotation and sub risks. Someone like Silva is a constant sub off threat ad is just as likely to play 75 minutes as 95. Those lost minutes are potential points left off the board. While I might not try and factor that in the rankings directly, I might add it as a risk element to each ranking. That could also account for benching risks in your rankings. Something simple like red, yellow, green would do nicely I think.

John Doe, 2008 said...
This comment has been removed by the author.
John Doe, 2008 said...

@Ste -

Great stuff. I have been tinkering with some similar ideas as well but my work is not yet ready for public consumption. Looks like you are far ahead of the curve (statistics pun intended)!

John Doe, 2008 said...

@ Gummi -

I was hinting at that exact thing in my point #1. SOT is a handy metric, but might it be even better to dig deeper into the core components of performance. It could very well be that SOT is quite variable, while more rudimentary stats (total shots, penalty touches, final third touches, etc.) may be more consistently predictive.

The analysis would be heavily reliant upon OptaStats but I don't see anyway around it.

JayDavies said...
This comment has been removed by the author.
Eraser said...

Didn't Tevez and Aguero start together the first week of the season?

Ramzy said...

Hi Chris! Great stuff as always.
I am contemplating taking a 4 point hit to bring in Jelavic for Tevez. Mancini's comments about his fitness have convinced me he is not going to start against Fulham.

Ste said...

I've had a quite a hard look at some of the 'deeper' underlying data, such as Pen Box and F3rd Touches and there is not a direct correlation woth FPL pts. Shots on target is the primary driver, after goals ofc. Players like Cisse/Jelavic last season hardly touch the ball but obv. Scored lots of goals. Some teams play bv. Dirrct and don't need a lot pf touches. It is useful as an indicator ad to a player's position though, and where about they are taking their shots from. A good metric I've found useful is Pen Box Tch/ F3rd Touches. This indicates that when a player is involved in an attacking pay are they involved close to the goal or not, and this therefore can be used to bias their goal/assist potential. I'll do a post on this.

Matt Chan said...

Hi Chris,
I think a crucial ingredient you are missing with p90 is that players don't always play 90 minutes. Perhaps you could add a metric that is Minutes Played per Game Played. As a previous poster mentioned, this would impact the statistics for players subbed off early or subbed on for garbage time. (I would do this as opposed to straight minutes played, as that would be detrimental to players like Rooney that are fantastic, but happened to miss a couple weeks.)
Best,
Matt

Matt Chan said...

Another thought for later on in the season is perhaps weighting the ten most recent games more heavily, as they will be more indicative of how a player or team is currently playing.