SoT=Shots on Target, Gin = Goals in Box, SOTin=Shots on Target in Box, Sin = Shots in box, SOTout=Shots on target from outside box, F3rd= Final 3rd Touches |

**UPDATE:**It was pointed out that I missed 'Total Shots' from this analysis so there is an updated post to include this data here.

The above table is a correlation matrix for almost 100 midfielders and forwards from 2011/12 season showing the correlation between their FPL attacking points scored and their key attacking stats. The 100 players include all the main men you'd expect from the 2011/12 season, and Stewart Downing.

The data was obtained via the MCFC Analytics project Thank you kind sirs.

When I first did this analysis there wasn't any really strong correlation, with the highest factor being the number of minutes played (0.61 correlation). Of course this is true. Half of most player's points comes from appearance points. So I stripped the appearance points out. I also eliminated defender's from the analysis as their clean sheet points would confuddle the data in the same way.

The results are meaningful and demonstrate why I am so focused and maybe a tad overzealous of the Shots on Target metric, and indeed why I have named this site such.

After goals scored and goals scored in the box (Gin) the strongest correlation with FPL points is Shots on Target (SoT), stronger even than Shots on Target In the Box (SOTin).

The correlation between Assists and FPL points is not so strong but really this is due to the vast number of points obtained from goals, particularly the extra point scored for a midfielder, and of course bonus points. If I do a basic subtraction of 5pts per each goal scored from the total points per player the correlation for assists increases to 0.75 (midfielder clean sheet points remain in the total, and 5pts per goal is a very simplistic value given the differnt range of bonus points awarded). It's nice to see a good correlation between key passes and final third touches (0.80) , however there is not a great deal of correlation between key passes and assists (0.65). I think this could be improved by looking where key passes are made - i.e in the box or from crosses, but this is for further work.

I don't think there is much more to say on this to be honest. I believe the numbers speak for themselves. Clearly goals mean fantasy points, and shots on target means goals.

I've actually taken this analysis further by combining a few of these individual stats to improve the correlation and produce a profile of the optimum FPL point scoring player. Stay tuned.

Again, thanks. This is so helpful.

ReplyDeleteHow about just numbers of shots? Is the correlation there not good? Do we need to be looking at Shots on Target or some combination of both?

Gummi. I missed shots off accidentally and realised this earlier so re-ran the numbers. The correlation of Shots and FPLPTS is 0.73, with Goals it's 0.73, SOT it's 0.88, and Sin it is 0.80.

ReplyDeleteI'll update this post, or new post, to add these important missed numbers in.

If we are only looking at one metric then it should be SoT. But of course we have access to more than jsut one. I am working at combining the data into a hopefully improved metric.

Excellent. Thanks for the reply. It seems that it must be time to start crunching some numbers myself.

Delete"If we are only looking at one metric then it should be SoT. But of course we have access to more than jsut one. I am working at combining the data into a hopefully improved metric."

ReplyDeleteI am confident we can find a model that regresses to a single season with a correlation in the low 0.90's. I am hopeful, we can find one that regresses over multiple seasons to the low 0.90's.

Once we find that piece, we need to start trying to uncover what leads to the driving activities of said model. In other words, if SOT is the basis of your model, what things drive SOT? Is it penalty area touches factored in with final third passes received? Is it key passes received? Is it something innate in each individual player that lends him to have more SOT (or whatever) in certain formations?

Once we get the basic model down, the real fun begins!

But do we need more? If SoT has such good correlation isn't that most of what we need? We could use other numbers to find diamonds in the rough, perhaps?

DeleteThere is need to dig deeper in order to provide a faster reacting metric. The worst thing, or one of the worst things, about relying on statistical analysis to make FPL decisions would be ignoring an obvious player because you're waiting for the data to become reliable. The further down you go in th data the faster it will "react" to changes.

DeleteHi Shots on Target,

ReplyDeleteJust wondering if you could split out the 'FPL Points - Start'

(Not sure why it's call that though, thinking this means 'Total Game Week FPL points' rather than 'FPL Points - Start', as they wouldn't correlate strongly with most of the elements i.e. it doesn't matter if you score a goal you still get starting points? Please can you clarify what aggregation we are considering for the points.

Anyhoo back to my point of splitting out the 'Total Game Week FPL points,' If possible would be good to see how Bonus Points correlate to the different measurement elements.

Further to this you could split the positions out, and then run correlations for each position to see what drives points for each position.

Wishing I had the data myself now... :)

Cheers

Matt

That is FPL Pts "minus" Starts, i.e. FPL total points minus appearance points. I explain the rationale in the 4th paragraph starting "When I first did this..."

ReplyDeleteBonus Points I do not have data for from last year although it's available on the FPL website for this years so I could do this analysis. You could do it too. Basic shot/appearance data is on http://www.whoscored.com/Regions/252/Tournaments/2/Seasons/3389/Stages/6531/PlayerStatistics/England-Premier-League-2012-2013

Aha, now that makes complete sense, totally miss read the ' - ' instead of 'minus' much clearer.

ReplyDeleteThanks for pointing out the website, will run some numbers myself, intend to split out the positions and bonus points. Once complete will post the results.

here's the update....

ReplyDeleteI collated (through vlookups) the Stats that were on WhoScored.com & The Bonus points from the FPL website. There are a few players that don't overlap between the two websites, because only players with total appearances greater than the average number of appearances are displayed on whoscored. Unfortunately both of those sights don't hold data such 'Shot's on target', 'Key Passes' etc info. But thankfully we can use this site, (I take it these types of data cost money)

Anyhoo, with these few exceptions I've completed correlation on different positions for BPs for a range of metrics....

GK BP Defenders BP Midfielders BP Strikers BP

Goal - 0.689716 0.609090 0.786234

Assists 0.623610 0.254777 0.438051 0.324136

Yellow 0.270868 0.005562 -0.064479 0.142719

Red - -0.004030 0.020359 0.015589

Shots pG 0.378620 0.413454 0.392049 0.368207

Pass Succ% 0.126229 0.059663 0.051673 -0.014996

Aerial Wins -0.155566 0.134920 0.108682 0.244276

MoM 0.644383 0.428543 0.573696 0.721021

Total points 0.331597 0.754325 0.812197 0.858451

I definitely need to expand the metrics to find more relevant ones. i.e Saves, Tackles, Forward passes etc etc (Not to mention the ones in the article above, but that may cost hence not too sure that will be any time soon.) From the above data it would seem that the people at the FPL are mostly basing Bonus Points on Goals and then MoM's & a tad surpisingly total amount of points accumulated that week rather than something else i.e. Assists. For GK's I don't think I have the correct metrics to measure the way BP's are being allocated, as there are very few BP's that have been awarded so far, hence a larger data set will help.

Cheers.

ha, Pity the table didn't publish nicely...

ReplyDeleteHi Mattm, thanks for this. Can you email me please? shots_on_target@hotmail.co.uk Ta.

ReplyDeleteHi Shot's on Target,

ReplyDeletePlease see this article I wrote inspired by this article.

http://footballcubed.blogspot.co.uk/2012/11/analytics-correlation-article-1.html