Michael Caley has written an excellent series of posts recently on loking at identifying shooting skill in the Premiership, and whether there are specific players that an identifiable amount of it. One piece in particular that caught my eye is this one.
I think Michael and I think about shooting skill in a similar way – it’s something that exists (though the magnitude of it’s effect is open to debate) but over the course of a Premiership season it’s very tough to identify because players don’t take a large enough number of shots. Even teams take so few shots that their shooting percentage will regress ~60% of the way towards the mean from season to season.
In the piece linked in the introduction Michael has taken non-penalty shots from the ’09-10, ’12-13 seasons, adjusted the shots for each player based on the location of the shot, the type of shot, and the type of pass that assisted the shot. He’s then split players into two groups – shooters and non-shooters, with shooters being those who’d taken more than 38 shots over the course of the season (n.b. the original post says ‘more than a shot per game’ – I’ve confirmed with Michael that it’s >38 shots in the season, regardless of games played).
The results? Well Michael explains it best, so I’ve taken an excerpt directly from the post:
Now that’s pretty damn interesting, and I was wondering whether it’s something I can confirm. Fortunately I have a set of data provided to me by Infostrada (twitter, website) that provides me as independent a set of shots as is possible – penalties. This is something that I wrote about extensively in 2012 – a summary of which can be found here – but it’d be great to test this dataset against Michaels’ results.
From the numbers I have there are three seasons common to both mine and Michaels datasets – ’09-10, ’10-11, and ’11-12. Michael was kind enough to send me the list of ‘shooters’ in each season (for those interested they’re included in the appendix at the end of this post). I split the penalties in each season into those taken by the ‘shooters’ and ‘non-shooters’, as defined by Michael, and looked at the conversion rates of each of the two groups. The results are summarised below – first by looking at all players, then split into the two groups:
Now that’s interesting – in two of the three seasons the ‘non shooters’ have had a markedly higher conversion rate than that of the ‘shooters’, and in the third season it’s basically a wash. This looks to be showing the opposite of that found by Michael, but is it a significant finding? Below I’ve summarised the p-value of the two groups of penalties in each season, and across all three seasons combined:
So for those unfamiliar with p-values they determine how similar two groups of numbers are – and range from 0 (two groups that are entirely dissimilar) and 1 (two groups that are exactly the same). A p-value below 0.05 is considered to show that two groups of numbers are different to a statistically significant level.
As such, given the sets of penalties and their relative sample sizes this is something we’d expect to see happen about 12% of the time, and isn’t an effect that we’d call statistically significant. It doesn’t mean it’s necessarily something we shouldn’t be trying to explain though – can anyone suggest a reason as to why we’d see some players being statistically significantly better at converting almost all other types of shots, but not the one shot that is as damn close to a neutral shot as we’ll find?