Community Submissions

One Stat To Rule Them All

Jonty often brings up on the Scoutcast about how he loves his shots in the box stat, or chances created for midfielders – doing what midfielders should. Of course there is community regular Doosra with a stat table always at the ready to answer every question. Statistics are clearly a staple part of many Fantasy managers psyche, but which single stat is the best to follow. Here I’ve sought to find the best correlation between goals and assists over the last four seasons, crunching some numbers to find out what is the best stat to use to predict Fantasy returns.

Results

stats1

 

The above chart shows the average R squared per stat over the last four seasons. R squared is basically how well the stat explains the differences in performance (a more technically correct definition is how it charts the differences from the mean, in this case a linear trend line joining the stat to goals and or assists). I worked out how well the shot based stats per player correlate to goals scored over the season, the touches and passes received to goals + assists and the chances created and throughballs to assists.

What the data shows over four seasons (2011-2014) is at least over a season that Jonty and other shots in the box fans are better off looking at shots on target. Also touches in the area and passes received type involvement based statistics are pretty poor predictors of goals and assists relatively (especially when used in isolation) and chances created and really good chances created (throughballs) are really poor at explaining assists.

We can also expect reversion to the mean, which means players that start out with lots of goals but poor shots on target numbers, such as Anthony Martial, over a whole season will revert closer to what the shots on target numbers predict. Conversely, players with high shots on target numbers but a low goal tally look to be good acquistion targets if they can maintain their shots on target statistics.  This bodes well for Sergio Aguero owners in particular.

This Season

Looking at the season so far for the best shots on target and big chances (two highest R2)

Shots on Target:

soTarget

 

Big chances:

bchances

Of course you can and should use stats in combination and small sample sizes for this season caveat but I thought it was an interesting exercise and is hopefully useful for some people. It is clear that with shots on target there really is one stat to rule them all.

rakkhi Love my football, love my stats, hoping to improve each year. Go the gunners! @rakkhis on Twitter Follow them on Twitter

113 Comments Post a Comment
  1. Numb
    • 9 Years
    8 years, 5 months ago

    Just to play devil's advocate - isn't this more the case that to score goals you *have to* have shots on target, by definition? I.e. The more goals a player has scored the more shots on target they must have had, rather than having shots on target being a predictor of future goals - which is what we're really looking for from stats.

    1. Numb
      • 9 Years
      8 years, 5 months ago

      Same for "big chances" - most goals result from a big chance.

    2. Camp No No
      • 10 Years
      8 years, 5 months ago

      Well, it works in the way that if player gets constantly shots on target with good rate, he will score more goals than if he didn't. Of course, if the SoT conversion rate or the shot accuracy are unsustainably high, then that number for that player is, given long enough time, going to regress towards the mean.

      The number of big chances may in the first place explain high conversion rates. It is a rather blunt tool, but does that work. I think you have to take a look at how the team gets their chances rather than the player. It certainly requires more close look, to passing game in particular. That need of a closer look is generally the case when you make individual decisions based on stats.

      1. Numb
        • 9 Years
        8 years, 5 months ago

        Let me give you a very recent example.

        In the 4 GWs before his hattrick, Sanchez had a total of just 4 shots on target - that's why he wasn't scoring the goals - past performance.

        The stats he excelled at in those four weeks however were pen box touches and shots in the box - he was very high for both of these, however not many of his touches / shots were on target.

        But maybe these were a better predictor of future returns - and then he got his hattrick.

        1. Woy of the Wovers
          • 13 Years
          8 years, 5 months ago

          There's a danger in reading too much into small volumes of data and hat tricks, being rare, present a bit of a trap for the budding statistician if they focus on the players who get them. They are highly likely to be outliers in the overall data set. I think if you look at Sanchez overall stats this season, he's now at or about is "expected goals" based on his SoT - before GW7 he was well below that figure while GW7 gave him more than he should expect.

          You are certainly right in saying that a player with lots of goals will generally have more SoT so there is the worry that this isn't a predictive stat but rather a confirmation that the player scored - and everyone can see that data. I think we can more readily accept the premise that a player who has previously got into scoring positions, will do that in future so high SoT historically indicates that the player will do that in future. The figures in this article then shows that this will be a good indicator of future goals.

  2. Jose's Magic Omelette
    • 10 Years
    8 years, 5 months ago

    The main limitation of this analysis is that, as I understand it, it's comparing shots on target and goals over the same time period. It shows that previous shots an target are, of the variables chosen, most closely correlated to goals previously scored.

    This doesn't mean that shots on target will be the variable that is most closely correlated to future goals scored. As an illustration if you consider the association between previous goals scored and previous goals scored across the same period it will of course be a perfect correlation, better than any of the variables in the graph above, but we all know from experience goals scored previously are no guarantee of scoring future goals.

    What we really want to know is which variables including previous goals and assists scored were the best predictor of future goals and assists in past seasons. I'm going to try do some retrospective longitudinal analyses to see if there are any variables which are consistently the best predictor of future goals, if it throws up any interesting results I'll post an article.

  3. gooberman
    • Fantasy Football Scout Member
    • 12 Years
    8 years, 5 months ago

    On this subject, I've noticed that there doesn't seem to be any correlation between player stats and points projections. There seems to be a glaring anomaly in respect of Pelle. Taking key stats such as touches in the box, shots on target, goal attempts, chances created and shots inside the box, Pelle has the most impressive figures out of the strikers. He comes out on top in 3 of the above indicators and is 3rd and 4th for the other 2. However, he is way down in 7th in the strikers next 6 gameweek points projections. Makes no sense. Rooney is even above him whose above mentioned stats are significantly worse than Pelle's and Man Utd also have worse fixtures over the next 6 according to the season ticker. I just can't think of any other statistical indicators in favour of Rooney that would over ride the above mentioned stats to put him above Pelle in the points projections. Rooney may well out point Pelle over the next 6 games, but at present, there doesn't appear to be any statistical basis to place him above Pelle.

  4. tm245
    • 12 Years
    8 years, 5 months ago

    This is a great article. Reading the comments is also very instructive and I was wondering if the statistics could be weighted to actually come up with a predictive metric by looking at how many goals were scored from the two stats and then applying it to what the player produced.

    What percentage of SiB became goals in the past?
    What percentage of SoT became goals in the past?

    Then look at what players have actually done so far this season, assign a weight to those shots that have gone in so their 1-1 correlation doesn't influence the stats too much, and then combine them. Kind of an Explosion Imminent plus Explosion Permanent, if you will.

    Or is that what Expected Goals already is?

    1. tm245
      • 12 Years
      8 years, 5 months ago

      Going to use this article as a place to store some thoughts. Thanks again for the spark from this article.

      Goal Attempts is a raw total.

      What if you could break it down a bit into subcategories of expectation:
      S Outside the Box x % likelihood of scoring +
      S Outside/On Target x % likelihood of scoring +
      SiB x % likelihood of scoring +
      SiB/On Target x % likelihood of scoring?

      You might come up with some aggregate of expectation vs actuality in terms of goals scored...

      Don't think it is quite there, but it might be headed there.

  5. Concrete
    • Fantasy Football Scout Member
    • 9 Years
    8 years, 5 months ago

    WC team ... Which one do you prefer?

    A

    Butland Gomes
    Bertrand Bellerin Francis Toby Souare
    Sanchez KdB Payet Walcott Mane
    Lukaku/Bony Pelle Ighalo

    B

    Butland Gomes
    Bertrand Bellerin Francis Toby Souare
    Sanchez KdB Payet Mane Cabeye
    Costa Pelle Ighalo