Sunday, March 3, 2013

Shot on Target ratio

One of the areas we need to work on over the summer is how shots on target are converted into goals, specifically the split between types of shot. We've observed that, on average, shots on target become goals at a league wide steady rate (30%-33% depending on position), but on an individual level things are more complicated. We consistently see players like Adel Taarabt show up as outliers in that they've taken a lot of shots, hit the target but yet not received the final reward.

One of the obvious starting points is to look at the ratio of shots taken inside and outside the box, and this is something that is already baked into the current model. However, playing around with some data recently, I put together a graphic which hopefully lays out some of the questions that remain unanswered, and I hope to spark some debate to push this forward in the offseason.

SiB Ratio - percentage of shots taken inside the box
G/SoT - percentage of shots on target converted into goals (excluding penalties)

 On first glance, the signs aren't too promising here. We don't observe a particularly strong relationship between the two ratios and we see successful players scattered all over the chart. If we focus on those players who've notched more goals this year (represented by a larger circle) though, we can see something develop, with only a handful of players managing to really defy the logical trend of having a lower G/SoT rate is they taken less shots inside the box:

One thing that could potentially explain some of this variance is the fact that this quartet of players all take a large number of set pieces, all ranking in the top 10 among midfielders:

1. Bale 24 attempts, 0 non penalty goals
4. Gerrard 18 attempts, 0 non penalty goals
7. Cazorla 16 attempts, 0 non penalty goals
10. Lampard 14 attempts, 1 non penalty goal

I must say it wasn't particularly intuitive to me to think that taking set pieces would be a bad thing for a player's stats, but it does seem that in terms of their impact on ratios, scoring a long range free kick does seem harder than scoring a long range effort in open play. I therefore think that for future iterations of the model we need to further distinguish between not just SiB and SoB but also shots taken from set pieces. The issue is that from the Opta data I currently use, we don't seem to have a distinction between shots from set pieces in terms of long range efforts and those where a player gets on the end of a cross from a set piece (which presumably have a decent chance of being converted). As I say, the search for a magic formula continues.

At this point in the season there are more pressing issues to think about (like trying to field a GW29 lineup!) but I want to keep these questions in the background to make sure they are ready to be discussed over the offseason. On that issue, I will be guest writing in a new fantasy publication called InsideFPL which launched last week. At this stage it is still a developing idea so any suggestions or comments would be welcome but I encourage everyone to sign up and give it a read as it could represent an exciting forum to discuss fantasy football and there are some excellent writers already on board.

7 comments:

CDI said...

Hey Chris, I was wondering if you had any plans to update the player dashboard?

2ndMan said...

Hi Chris, thanks for the article. I'd be tempted to remove all attempts from set-play (including those headers from free-kicks) which would eliminate shots from free-kicks, but also give a measure of shots from open play which may be more informative. Although this is ofcourse whittling the sample size further and makes correlation even less likely.

stooshermadness said...

Chris - I guess my take is that while you might come up with a better predicter of run-of-play goals/assists by removing set pieces from the calculus, in the end set piece takers are essentially getting a "free" chance to score/assist above their run-of-play chances. So that - even assuming the ratio of goals/assists from set pieces is significantly less than, say, 20 yard run-of-play shots or run-of-play crosses (which I'm not convinced is true), it's still a "free" chance (a chance no one but the set piece takers get) to add to FF points. So, while it may squew the overall 'normal', for the individual player I think it's still an important factor in selecting 1 player over another.

Ryan said...

@Stooshermadness

I think the point is though that when assessing the correlation of SIM/SOT, set pieces skew the results. Its a way of leveling the playing field somewhat in having a clearer understanding of these key stats. Especially if set pieces do indeed have a much lower success rate.

stooshermadness said...

Thanks Ryan. I guess my bigger point is that I think we can get carried away with combing the correlations ever so fine. A big part of why I come to Chris (and this is my go-to FPL site) is his narrative via the weekly line-up analysis and his brilliant 'Fanning the Flames' pieces. My comment above is just a plea not to get too caught up in the weeds. It's how Chris reads these stats, in conjunction with actually watching the games (anybody who has watched Taraabt play, for instance, that despite his great numbers, his penchant for shoot-on-site-from-30-yards would never result in a lot of goals and took a lot away from his assist potential) that keeps me coming back. I suspect there are mathematicians who are developing an algorithm approach (to take a stats approach to the nth degree) for selecting a team but I am not sure that would be too much fun. I like Chris's efforts to refine his approach to keep the numbers honest, but I suspect the PPI or whatever the FFS Member stats are based on gives a pretty credible take on player 'value'. So, what I like here is Chris's analysis of those numbers, not getting ever deeper in the weeds.

Chris Glover said...

CDI - yes sorry, I fell behind on that but I will be doing it tonight.

Stoosher/2ndman/Ryan - yes I think that's an important distinction to make. The chances from set piece skew the predictive nature of a model, but shouldn't be ignored altogether when assessing a player's value as, as Stoosher notes, they essentially extra 'free chances'. What we need to do is distinguish further between shot types, as we do with total shots. We don't throw SoB totally, but do concede that they are turned into goals at something like a 4% rate and thus Taarabt's value needs heavily discounting.

As for the wider point, I like messing around with the data just because I'm a nerd, but I totally agree that in isolation it isn't always too helpful. I am trying to get my spreadsheets etc to be as automated as possible over the summer so that that side of the blog should essentially just post itself every week, freeing up time to get into the secondary analysis which gives the data life.

Thanks for the comments guys

Helen Evans said...

let's be frank, even if you are only in a fantasy football league to have fun, it's more pleasurable to WIN and to more, visit at fantasy football targets