clock menu more-arrow no yes

Filed under:

An Opinion On Football Statistics and Ryo Miyaichi

New, comments
Getty Images

For someone who uses statistical analysis (better known as sabermetrics) in baseball, you might think the increased use of statistics in football would please me. To an extent, I am pleased by their increased use, and the feeling that statistics aren't the domain of nerds in their mother's basement. When Joey "Shane Victorino of the Premier League" Barton called people who use statistics "wankers" there was a universal response of telling Barton that he was an ignorant fool. Sites like WhoScored and EPL Index give more access to statistics for fans, and these is some value in some of their statistics. The thing is, though, football statistics could do a lot better.

So far, development in statistical analysis in football has led to analysis mainly being based on the value of goals and counting stats, such as tackles and interceptions, and crude rate stats, like interceptions per game. There is value in those statistics, as long as any comparison is placed in context. Comparing Robin van Persie's tackles per game and Scott Parker is silly, because the two play completely different positions; comparing van Persie's shots per game with Wayne Rooney's shots per game is a more worthy form of analysis, as the two play somewhat similar roles, while comparing shots per game gives an idea of how much service they get and whether they create chances for themselves.

A majority of passing statistics also lend themselves to worthy analysis, though, again, many are just counting statistics. The OPTA definition of through balls, requiring a pass to go between two defenders, and the subsequent statistic gives fairly good evidence of prowess at providing through balls. Like many of OPTA's stats, though, it could be better. Not only could OPTA provide how many through balls per game someone like Alex Song or Aaron Ramsey completes, but they could also give a percentage, and then allow for better analysis.

One passing statistic, though, that is very flawed and widely used is "key passes" or "chances created". The premise behind this statistic is that every shot is a chance, and thus, the assister to this shot should be rewarded with a "key pass". It's not an entirely bad idea; the problem, as with many statistics in football, is that there is no further analysis to chances created. There is no differing in quality of chance between a 40 yard shot and a possible tap in, and thus, every key pass is treated as being equal when it quite obviously isn't. To not differ between quality of chances and then call results thorough analysis is inexcusable, because it isn't thorough analysis; it's like using RBIs to compare power hitters when wOBA is available. We're now at the stage in full video documentation of many matches that many game-states are repeated from previous matches. Thus, win probability for events like passes and tackles can be computed. It's hard to compute win-probability, and it'd take thorough documentation of many events, but it most certainly isn't impossible.

Because we don't have this stage of analysis complete yet, we're left with misguided opinions being made based on incomplete statistics. Arsenal loanee Ryo Miyaichi has been very impressive for Bolton, and has made 3.2 key passes per game. This is more than the best playmakers in Europe, leading some to say that Ryo is an excellent creator of chances. The problem, though, is that many of his chances have been created from crosses or corners. Generally, crosses have a poor goal conversion rate, thus, the quality of chance isn't as good as an Alex Song through ball. Without differing the quality of chance, though, OPTA would have you believe that Ryo is a better creator than Alex Song, which simply isn't true. With better analysis, we could have a much more meaningful comparison between Ryo and Song, or Ryo and Theo Walcott. If we truly want to have better statistical analysis in football, and more meaningful comparisons, ditching some of the counting stats like "chances created" and replacing them with win probability will go a long way towards better analysis.