Expected Points (Expt Pts); numbers to use

What is Expt Pts (Expected Points)?
Expt Pts is very loosely aligned to Expected Goals (xG) in soccer. You take every shot and assign it a value linked to the likelihood of its success based on the observed historical averages.

Essentially you track every shot and once you have enough data points you work out the average return for each shot. The average then becomes the “expected” and you are up and running (worked example and current outputs below)

Before we set out I am acutely aware that Expected Points is not the answer to anything. It is a tool to quickly highlight if there might be issues. In its current form it is very raw. We know from other sports that the game state is very important (teams chasing a lead in soccer tend to have more shots for instance) whilst the quality of opposition can also be a factor (football outsider’s DVOA v VOA). As such the values herein will be referenced as “Raw Expt Pts”. Other variables need to be overlaid onto the raw averages, whilst tighter sectors also need to be worked upon, to elevate it from its raw state.

These deficiencies are known and acknowledged. But as a quick reference point it is still a better guide to a teams’ returns than pure Conversion Rates – or worse still a count of the number of wides (don’t get me started!)

Working out the average
Unlike soccer those working in the GAA do not have a company, such as Opta, collating shot location data – it has to be done manually. The area on the pitch from where shots emanate from is also much greater.

Rob Carroll over at gaelicstats has produced a detailed return graph showing how often shots are converted from various pitch co-ordinations. The Expt Pts herein uses a more simplistic sectoral approach. Whilst Rob’s shot location chart is much more granular we will run into sample size issues were we to use something akin to it. Instead I started tracking shots using the below grid to (a) ensure I overcame any sample size issues and (b) it enabled me to use the pitch markings so as to be consistent with the shot location seeing as I was doing it manually off TV pictures (the picture quality of which can vary)

As well as shot location the type of shot is noted – 45, free, point attempt from play, sideline, penalty – as well as whether it was a point attempt or a shot on goal.

So we end up with three questions that need to be accounted for on every shot – location (which is then converted into one of the above segments), type and whether it is a goal attempt.

The type is fixed and whilst there is some subjectivity on the location generally speaking we have a lot of latitude given the larger segments we are working with. There is even less subjectivity on whether a player is going for goal though you will occasionally need to be a mind reader!

Expected Points
Fine. I have the above tracked for a game and realise its limitations – now what? Now you start to (a) produce data and (b) use it to review. Below is the Raw Expt Pts return for every combo mentioned above (pitch position, shot type and goal attempt or not).

Should you wish you can start to use these immediately as they are based on a dataset that comprises well in excess of 4,000 shots. That dataset comprises entirely of intercounty Championship games but I have used the outcomes on league & club games as a starting point and taken the results with this in mind.

Take a point attempt, from play, from sector 4. The Raw Expt Pts for every shot from this sector is 0.39. Every shot is thus expected to produce 0.39 points. Now we know the binary outcome for this shot is either a score (1) or not (0). Thus if the shot is converted the Expt Pts outcome is 1 (scored 1 point) minus 0.39 (the expected value) for a return of +0.61 (1 -0.39 = 0.61). If the shot is missed, for whatever reason, the Expt Pts outcome is -0.39 (0 – 0.39 = -0.39).

Do this for every shot using the appropriate “Raw Expt Pts” value. You can now add up an accurate, appropriately weighted, value for each shot.

Concrete example
Below is the Mayo shot chart from the 2017 All Ireland defeat to Dublin.

Mayo’s Raw Expt Pts come in at +1.84 which means that, at a macro level, they scored about two points more than they should have (were expected to) going on how often inter county teams historically converted those attempts. Their shooting was very good.

Breaking the shooting down we can see that the “very good” shooting was not universal. Two goals attempts returning 1 – 00 is just above what is expected. 21 point attempts returning 0 – 12 (57%) with a +2.39 Expt Pts is exceptional … but the excellent shooting was restricted to the central Sector 5 channel. Deadballs were poor.

By using the Expt Pts – with all its inherent flaws – you can quickly highlight areas that need addressing (or alternatively you can get bland results across the board and move on to something other than shooting). Its real power comes in over longer timeframes where the returns from multiple games can be reviewed.

Build upon the data and you can get a sense of the team’s relative strength/weaknesses/tendencies. Do it for the opposition and you start to build up a picture of the defence.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: