A long time ago I took a look at a range of statistics to determine which was the best predictor of future performance. I did this by looking at the performance of a team one season and seeing which statistic showed the least regression towards the mean.
I now have a much larger data set (702 sets of back-to-back seasons) on which to base a more concrete conclusion and the results are as follows. The initial post demonstrated that using ratio’s (such as shots on target ratio) regress far less to the mean than differential or pythagorian values and so these are what I will report. The updated table appears as follows:
| Factor | R2 | Regression towards the mean |
| Goals | 0.332 | 42.4% |
| Shots on target | 0.462 | 32.1% |
| Total shots | 0.473 | 31.2% |
| Sh% | 0.109 | 66.9% |
| Sv% | 0.128 | 64.2% |
| PDO | 0.067 | 74.1% |
To give these values a bit of context, the points a team earns regress ~47.8% towards the mean over consecutive seasons (it’s important to note that I didn’t include seasons where teams were promoted or relegated so the inevitable drop off/improvement the following year). Therefore if we want to know how a team will perform in the future then using points is less accurate than goals or shots.
Two of the ratio’s clearly stand out as being much better at predicting future performance than the rest. In this case it turns out that the best predictor is in fact the ratio of the total shots in the games a team plays. From now on (as in the few posts that remain) I’ll switch to use total shots ratio to predict future performance.
Finally I threw sh%, sv% and PDO in there to once again reiterate just how much they fluctuate from season to season. All three of these are vastly more luck driven than skill driven.
Pingback: Winning matches, is it luck or skill? | 11tegen11