Skip to content

Goal Kicking Accuracy Maps – ExpScore What's a Shot Worth?

Posted in ExpScore, Game Analysis, and Scoring Map

What this early article refers to as “ExpScore” is now usually referred to as”xScore”.

A key component of my FFSS ratings, which are used to calculate the winning chances of each team ahead of each match, is the concept of “Expected Score”. Expected Score, or to save precious character space, “ExpScore” is a basic concept, and one I’m surprised to not hear more about in footy coverage.1 It’s simply the score that a team would have kicked had they converted each of their Scoring Shots at the competition average based on the difficulty of each shot.

Teams that create many, higher-quality scoring shots, which we would expect the average player to kick, have a much higher “ExpScore” then teams that create only a few, lower-quality shots.

This is a way of valuing a team based on the amount of good chances that they create, opposed to just looking at how many goals they kick. I will show in another article that this is actually a much better way of differentiating good teams from bad then looking at final scorelines is. It turns out, that even the best teams tend to convert their scoring shots at roughly the competition average, what differs is how many, and what quality chances they give themselves.

But that’s for another time. Today I thought I would, for the first time, try and pull back the curtains and visualise how FFSS rates a range of different shots. How exactly does a shot with an ExpScore of 5.9 differ from a shot with an ExpScore of 2.5?

Well, first off, after a Scoring Shot is recorded, we use two criteria to determine the quality of that shot and it’s contribution to ExpScore. Firstly we look at whether the shot was a Set-Shot or not, and secondly we look at where on the ground the shot was taken from.

Before explaining how ExpScore is calculated, we need to look at the historic accuracy for each type of shot.

Set-Shots


Set Shot Accuracy Map

A Set-Shot is simply defined as any shot which was preceded directly by either a mark or free-kick, and in which “play-on” or “advantage” was not called. Note that neither time spent on the shot, nor whether there was a man on the mark is considered. You can see a map above of the empirical conversion rates for a set-shot from any spot in the forward half.

To create this map I have looked at every set-shot taken from the 2012 season until now, a total of 21,452 shots. For each spot in the forward half, I look at all shots taken within 5m of that spot and and calculate the percentage of those that were goals.2

This methodology creates some strange results around the boundary lines and at the edges of the players’ kicking range, about 55-60m out. These are areas where few, if any shots are ever taken from.

You can also see a odd little splotch about 60 out to the left of the goals. This is due to a smallish sample (20-25 shots) being skewed most likely by a few shots into an open square with nobody home. However you can see a slight preference to the left side of the ground from shots taken further out. This is probably due to the fact that right footers can hook their boot into the ball and shape it into the goals a bit easier from this side. There are more righties than lefties so we shouldn’t be surprised to see a bit of a preference to this side.3

Shots from Open Play


Accuracy map for open play shots

Any shot not classified as a set-shot gets lumped in this category. In total, there are 21,883 of them.

As you can see, the accuracy rates for shots from open play are much lower than those from set-shots. A shot from 25m out on a slight angle is converted about 84% of the time if it’s taken as a set shot, compared to around 50% if it’s taken from open play.

Of course, it should be kept in mind that not all “Open-Play” shots are created equally. The differences in level of pressure on the goal-kicker can vary greatly. Strolling into an empty 50 and taking a drop-punt is theoretically a much easier kick than hooking around your body through heavy traffic. I haven’t found a way to be able to algorithmically distinguish between these shots, so for the time being I am lumping them all together.

What is ExpScore?


To calculate the ExpScore for a shot we look at it’s type (set or open) and position and find the competition-wide average accuracy of all shots, up until that point in time, of the same type, from within 5m of the spot the shot was taken from (which corresponds exactly to the images above for a match in round 15).4 This probability is multiplied by 5 and then 1 is added to give us our expected score from that shot.

For example, a set-shot from the top of the goal-square5 has about a 99.5% chance of being a goal. The ExpScore of such a shot is then:

0.995 x 5 + 1 = 5.975

Or essentially 6 points. On the other hand, a shot from open-play, about 40m out, near the boundary might have closer to a 32% chance of being a goal. The ExpScore of this shot is:

0.32 x 5 + 1 = 2.6

Calculating the ExpScore for a game is as simple as calculating the ExpScore for each Scoring Shot taken and summing it all up along with any rushed behinds registered throughout the game.

An Example of ExpScore


Let’s have a quick look at the Western Bulldogs v Geelong game from a couple of weeks back. The score finished WBD (5.13) 43 to GEE (16.4) 100. A big win for the Cats, but seemingly pretty even in the amount of scoring shots each team created. The Bulldogs created 18 and the Cats 20. In fact, the Bulldogs created 2 more inside 50s over the course of the game, a stat which is often said to show who is on top.

However, when you look at where and what types of shots each team had you can see why the Cats won so easily. First lets have a look at the Dog’s output.

Scoring Shots Map for Western Bulldogs Geelong HT

The Bulldogs created very few chances in those “high percentage” areas. Their set-shots were taken mostly from the 40-60% areas and the fact the they converted only 4 out of 9 should not surprise us at all. On a different day they might have perhaps kicked another, but they certainly did not butcher any high-quality chances.

Similarly, their open-play scoring shots were all from weak areas and the fact that they converted only 1 out of 9, while also being a little under what we would expect on average, is nothing too unexpected. They didn’t manage to create a single scoring shot with anything greater than a 60% chance of being a goal. From 46 inside 50s this is very impotent.

When adding up all the expectations we get an ExpScore of 64. On an average day they would have kicked a score of 64 from their 18 chances. Certainly nothing to celebrate, but better than the 43 they kicked. Poor kicking on difficult shots let them down.

Now let’s compare this with Geelong’s chances.

Scoring Shots Map for Western Bulldogs Geelong AT

The first thing that strikes us is the amount of high-quality shots from in and around the goal-square the Cats took. Unsurprisingly, they kicked all of these.

The next thing we notice, is how few misses they had. Outside of those run into the goal-square and a few set-shots in higher-quality areas, Geelong took 8 shots from 40-60% areas and converted 7 of these. If they had these shots again, it would be highly unlikely that they would kick this many. It’d be easy to say, “Well Geelong are a better team, and better teams kick better”, but anybody who has followed the Cats this year knows that is not true. They have been all over the shop in accuracy in front of the big sticks. Look at another game, from another week and it would be easy to make the argument that Geelong kick worse.

The truth is, they most likely kick at about comp average, and we expect to see some days where they are up and others when they are down. This is just how averages work. In this game they kicked better than their ExpScore of 86, but still deservedly won the game with all of those high-quality chances the Bulldogs failed to create.

One Big Blindspot


It’s likely by now you’ve realised that a shot doesn’t have to end in a goal or a behind, it may well not register anything.

Unfortunately, data on shots that did not result in a score is hard to come by. The AFL does not make all the data it collects through Champion Data publicly available. A big shortcoming of ExpScore and these accuracy maps in general is that the data set only includes “Scoring” shots. That is, intended shots that go out of bounds or otherwise fail to trouble the scorers by falling short are not factored into these percentages, nor do they count as “chances” towards an ExpScore for an individual game.6

Factoring in these other “missed” shots, the true conversion rates are probably much lower than those shown in the picture above. Particularly on tougher shots.

Similarly, the ExpScore for a game is not a “true” ExpScore as it does not factor in point expectation from those shots that miss entirely. But by using the higher probabilities, these two cancel each-other out somewhat and ExpScore still remains useful.

A Final Note


Looking at the quality of chances a team creates, rather than just looking at the amount of chances or the final score, is a logical step forward in analysing AFL and determining the quality of each team.

I will write more on how we can effectively use ExpScore as the season goes on, as well as disecting individual games I find interesting using the ExpScore framework. However, as of now I hope you can see it’s potential.


 

  1. A similar concept has been used by blogger/analysts in soccer for years and it is just starting to gain traction in the mainstream media
  2. providing at least 20 shots have been taken within 5m of that spot
  3. I should point out that I actually only have 100% correct positional data for seasons 2015 and 2016. Before that, I know the distance and angle to the goal for each shot but am not entirely sure if it’s from the right or left of the goal. This requires some tedious hard-coding which I haven’t really got the time for. If anyone is willing to help, do get in touch. No particular skills are necessary.
  4. If there have been less than 20 shots from within 5m of this spot, we give the shot a flat 30% chance of being a goal. This is basically all shots from more than 60m out. We assume an empty square from this range, so the 30% assumption is probably pretty good.
  5. meaning the man-on-the-mark is likely on the goal-line
  6. Shots that are touched on the line count as misses, but those that fall short and are then “rushed” through in a marking contest do not. Sometimes it can be tough to differentiate between the two.

7 Comments

  1. Blake
    Blake

    Great article and touches on a very important area which is often disregarded when analysing matches.

    I’m currently doing some research in this area and am curious to know where you obtained the data from for this. Was it purchased from Champion Data, or do you have other contacts/methods to obtain this?

    Cheers.

    October 7, 2016
    |Reply
    • The data, unfortunately, aren’t freely and openly available at this time. But there are some sites out there that you can glean some information from.

      November 23, 2016
      |Reply
  2. Thomas Wilson
    Thomas Wilson

    Hi Robert, My name is Thomas and i am currently in year 12. I am trying to find data to begin writing my maths internal assessment and i would love to write my maths essay on a similar topic to this. I know it is a big ask, but by any chance would you be willing to send me some of your raw data? My teacher has said that i cannot complete my essay on this topic unless i have some of the raw data you have collected. I understand that this is your intellectual property and because of that i would give you complete credit for all of your work within my internal assessment

    April 18, 2017
    |Reply
    • Hi Thomas, that sounds interesting, how can I help you out? Contact me at ryounger@ this website with specifics of what you need.
      Cheers,
      Rob

      April 18, 2017
      |Reply
      • Thomas Wilson
        Thomas Wilson

        Fantastic, i’ll send one through now!!!

        April 27, 2017
        |Reply
  3. Ethan Gates
    Ethan Gates

    Very cool. The visualisation is great and it really tells a story. I’m wondering if it would be possible to do similar analysis for left footers vs right footers? I think the commentators get it wrong sometimes when they say a certain position suits a left footer etc. and it would be cool to see what the data says.

    May 4, 2017
    |Reply
    • Thanks! Unfortunately I don’t have any data on which foot these shots were kicked with. If I had a resource showing the dominant foot for each player, I could probably do something, but I’m not sure if this exists.

      I do remember seeing a graphic done by Champion Data at one point that suggested that “footedness” does have a small influence. Unfortunately, I don’t remember much else and I can’t seem to find it now.

      May 5, 2017
      |Reply

Leave a Reply

Your email address will not be published. Required fields are marked *