The surprising importance of lineout steals
IN THE SPOTLIGHT: The number of lineout steals is the poor cousin in rugby analytics.
But it may prove to be the difference between winning and losing. I think there is a debate to be had over lineout steals.
Judging from the stats that receive constant attention in rugby blogs, newspaper columns and match commentating, lineout steals looks like a rather neglected metric in rugby analytics. It is no surprise either. Matches rarely produce more than two steals, so there isn’t much to say. However, I have come across a study that indicates this little metric might just be the most important. It might predict the outcome of matches and tournaments better than any other.
What the stats say about lineout steals
In 2004, a group of researchers in the UK published a study in the International Journal of Performance Analysis in Sport. They looked at the correlation between 22 key performance indicators and match results. The study looked at 20 matches played by a national premier division team in a particular season. Here’s the surprising finding: of the 22 performance indicators, lineout steals provided the closest correlation with winning or losing. By a huge margin.
Here are the values for the indicators:
(Note: For some obscure reason the study also measured the correlation between tries scored and the result. It was no surprise to be the best predictor of winning. Nonetheless, I ignore the points indicators in the study due to their obvious correlation with winning and losing.)
Explaining the numbers
I’ll explain the numbers briefly for the sake of clarity. When we say an indicator correlates with the match result, it means that it is higher or lower than the median/average for all matches.
Looking at the top indicator, Lineout success (opposition ball) %, we can see that the team analysed in the study managed to steal roughly 8.1% of the lineouts across all matches. Looking at only those matches that the analysed team won, it stole 14.6% of the lineouts. This is a whopping 79.6% jump up from the rough average across all matches. The statisticians calculated that the Lineout success (opposition ball) % indicator is the only one that’s “statistically significant“.
Statistically significant, meaning that it cannot be explained by chance.
The following graph illustrates just how much lineout steals stands out from the other indicators concerning correlation with match outcome. Including the third most significant correlation. Penalties conceded. Who would ever have thought that?
Now, I know you could argue that lineout steals occur so rarely that the numbers could quickly be inflated by one or two matches where the opposing side is merely inept at securing its lineout. I would counter that the matches in the sample were all played by quality teams and that the vast scale of the deviation most likely negates possible impact from anomalous games (if there were any).
It’s worth noting here, before we move on, that the things we usually assume are decisive for winning and losing, like successful tackles and attacking the rucks, actually seem to have very little relevance to the outcome of matches. Unless, of course, you fail to maintain decent playing standards.
Rugby World Cup stats
Now, let’s get back to the stats about lineouts. I found an interesting stat in the official analysis of the Rugby World Cup 2015. Looking at the sources of possession that led to tries being scored, the lineout is THE top source of possession for scoring tries. By a wide margin.
Here are the stats:
Now, guess which team at the RWC2015 had the highest lineout steal rate in the knock-out phase of the tournament?
New Zealand, the ultimate winner!
The stats are telling us something, but what exactly, I hesitate to guess at this stage. Part of me still feels that the steals are so rare that we should be careful to attach too much statistical importance to them. I am also a bit worried about using winning and losing as our test of correlation. We all know that the scoreboard often does not reflect the actual performance of the teams.
In tight contests, critical referee decisions, momentary player lapses or brilliance, and luck can and do swing matches one way or the other. The only way to mitigate these types of potential data distortions is to have large samples, which we just do not have.
Still, we cannot ignore the sheer size of the correlation between lineout steals and match outcomes, as well as the considerable gap between lineout steals and all the other indicators.
So, what is it about lineouts that give them this special significance in match outcomes? I have a few hunches, but I think it’s better that the rugby practitioners, rather than the statisticians, answer that question.