Weightings. A good comparison tool but not without its faults

With a bit of a lull in games to review now seems as good a time as any to post this.

I received a text from Brendan Coffey, a journalist with the Kildare Nationalist who is well worth a follow on twitter (@coffeybrendan), after the Dublin-Donegal game querying how it was that Ryan McHugh’s weighting was greater than Paul Flynn’s. Both converted four from five from play yet McHugh’s attempts were much closer to goal. With Flynn’s attempts being harder to convert surely his weighting should be higher?

And there is the rub. The weighting is not a finely tuned precision instrument but rather a roughly hewn prototype. I believe it gives a better sense of a player’s accuracy than any other measurement out there but due to its rough nature there will be inconsistencies.

A weighting is essentially a value for a shot based on historical Success Rates for similar shots. It is broken into four main components
• Is the shot from play or a deadball?
• What type of deadball is it?
• Is the shot for a point or a goal?
• Where on the pitch was the shot taken from?

Is the shot from play? This is easily classified but there will be differences of opinion as to what constitutes a shot. Anything classified as a shot gets a weighting. This is irrespective of game scenario (Dublin’s two goal mouth scrambles at the end of the Donegal game are counted as shots on goal as if they were one-on-ones in the 15th minute).

When creating the weighting originally the aim was to have no subjectivity. As time passes it is obvious that not all shots are created equally. Colm McFadden’s goal in the semi final is not the same as Connolly’s attempt in the 66th minute. To give them separate weightings however it would be necessary to (a) create subjective definitions over how hard an attempt was, the angle and distance of the shot etc and (b) collate enough examples to create a meaningful Success Rate in each new sub category.

Type of shot? Again this is easily classified however the weighting to attribute to each is not. Take an attempt from a sideline. Ideally we would have an individual weighting for this type of shot but in the database there are less than 20 shots from sideline balls. A weighting could be created from these twenty but given the low volumes the Success Rates could be way off the “real” returns. Instead I have decided to treat attempts from sidelines the same as frees from that sector. Now it stands to reason that the majority of frees are easier than a sideline therefore when a player misses a sideline attempt their weighting will be quite harsh.

A subjective decision (to treat sideline attempts the same as frees from that sector) has an impact on the objective outcome (the weighting). In hindsight perhaps it would be best to go with the objective return (take the Success Rates from the 20 sideline attempts and just go with that. Something similar is at play with penalty attempts where I have treated them as goal shots from play rather than use the weighting from their actual Success Rates.

Is the shot for goal? There is some subjectivity here but I would say that 90% – 95% of shots at goal are easily identified as such. The others can be harder to discern i.e. when the ball scoots over the crossbar – was he drilling it for a goal or drilling it high hoping for a goal? Currently whether you get a goal or a point the weighting is treated the same for a shot on goal. The player got a score. Again this is probably being generous to those players who aim high and get the point as opposed to those who try for the goal only.

Where on the pitch? This is relatively straight forward but again you can have different camera angles, poor camera tracking, a low lying sun obstructing your ability to see the lines on the pitch (and gauge where the shot was taken from). 99% of shots are easily placed on a pitch diagram but there will be that 1% that requires some guesswork. And that doesn’t touch on the subject of how big the sectors are. Only having six sectors inside the 45 is too few – the sectors need to be refined. Once they are the weightings will be adjusted accordingly.

So there you have it. The weighting is a good tool for shooting accuracy comparisons between players, teams and games. It is robust for 90% of shots and scenarios but when we get to the margins (sidelines & penalties, grading goal attempts) it is not fool proof.

And so we come to the original query – how was McHugh’s weighting better than Flynn’s? Three of McHugh’s shots were from Sector8 but all three were shots for a goal. He may have scored 2-01 but as outlined above these are all treated the same for weighting purposes. Those three shots gave him +1.778 (+0.5927 for each score). His first shot was from Sector5 which went wide (-0.4987) whilst his other point was from the left in Sector6 which has a weighting of (+0.6408). In total +1.920.

What of Flynn? His long range shooting was exceptional but unfortunately (for him) the sectors are too big to account for this. Thus his four points came from Sector4 (2 x 0.6407) and Sector 5 (2x 0.5013) whilst he also missed one from Sector5 (-0.4987). A total of +1.785.

Over the years players have converted point attempts from Sector5 more frequently than they have goal shots from Sector8. As such, from a weighting perspective, McHugh’s conversions were harder (taking all the above points into account) than Flynn’s were.


One Response to “Weightings. A good comparison tool but not without its faults”

  1. Review round up | dontfoul Says:

    […] of the deficiencies in the weighting were outlined earlier in the year (here) – especially when it came to less frequent events like penalties, attempts from sideline balls […]

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 )

Facebook photo

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

Connecting to %s

%d bloggers like this: