JQAS Article on Quality of Skill Performance and Winning Points

A recent issue of the Journal of Quantitative Analysis in Sports(Volume 6, Issue 2) contained an article by Michelle Miskin, Gilbert Fellingham, and Lindsay Florence entitled "Skill Importance in Women’s Volleyball." Access to articles is by subscription, but the journal has guest-visitor privileges for single articles.

Miskin and colleagues analyzed data for a particular women's Division I team (not identified by name) during the 2006 season. When the team played at home, play on its side of the net was videotaped and later coded. Serves, passes, and digs were rated by judges on quantitative scales (e.g., 0-to-5), sets were evaluated in terms of their distance from the net, and spike attempts were coded by area of the court from where they were hit.

Essentially, the authors appear to be looking at correlations (or associations) between characteristics and quality of skill performance, and likelihood of winning the point. As they state on page 2:

The importance score incorporates not only the impact of a specific skill..., but also the uncertainty associated with the performance... Thus, a skill whose association with scoring a point is less certain will be penalized when using this metric when compared to a skill where performance at a given level is more closely associated with a positive outcome.

The article throws a barrage of statistical terms at the reader (e.g., Bayesian analysis, Markov Chains, Dirichlet prior, Gibbs sampling, gamma distributions), some of which I was familiar with, but many of them not. Fortunately, the authors translated the complex statistical results into plain English recommendations for the team that was investigated:

1. Keep sets and passes away from the net.

2. Force the attack to the middle and right side if at all possible.

3. Devote a considerable proportion of practice time to transition offense.

4. Get to blocking positions more quickly following a serve.

Presumably, if a team wanted to apply the analytic tools described in the article in their full glory, it would need to hire a pretty high-powered statistical consultant (in addition to acquiring the videotaping and coding resources). Perhaps similar analyses could be done via more basic correlational and regression techniques, but I suspect that the resulting conclusions may be somewhat imprecise, compared to those from the fully sophisticated analyses.