How useful or accurate are player stats

[brumsfan]

I like looking at player stats, even though I know they don't tell the whole story. The metrics for some positions are very misleading and often make similar reading for players that have had vastly different impacts on games.
With this in mind I have started keeping some of my own, on players and positions I have an interest in.

For example, stats have nowhere to record pressure applied, cleanouts made, ball slowed down, loose balls gathered, or turnovers forced.

One player I recorded this weekend, had average published stats, but according to my very close review played an excelent game, had 2 more runs than recorded, made 6 more tackles, gathered 5 loose balls (not a measured stat) made 3 tackles that resulting in knock ons (also not measured)

I'm not sure how many people are tasked with recording stats during games, but for those of us that quote them as indicators beware, as they are probably wrong and as I said before a poor indicator of some positions.

 

[Scott Allen]

brumsfan - agree with your sentiments - here's an article I published last year on the same subject http://www.greenandgoldrugby.com/video-statistics-how-accurate-are-they/

 

[RedsHappy]

Not so much to the issue of individual player stats, more their usefulness in aggregate in terms of final team yield and game outcome, I find grave limits in the whole stats platform for rugby (and potentially just as much as per other codes and sports).

In that this: there is no work I have ever seen that does the type of sophisticated statistical modelling that would analyse the degree to which certain underlying patterns of statistical player or aggregated team statistics correlate over time with aggregate points scored per match, and, far more importantly, with winning match outcomes per a particular team's and opposition's total stats profile for a match. Namely, its not what necessarily happens in an isolated 'cell' of data per any one individual, it's the pattern relationship between multiple cells' interactions over time or by trend and time, that will reveal what underlying data cells (of say player behaviour) are actually the most valuable and important to achieve in team performance.

Stats taken in isolation mean little without the type of deep correlation analysis that is the heart of cutting edge data mining technologies and mathematical modelling today. Firms like the UK's Autonomy that HP has just purchased for $11bn do this work to exceptionally high standards and it's now in very high demand (partly as the internet is throwing up huge quanta of data regarding internet users' actual patterns of consumer behaviour over time). For the required deep correlation analysis, masses of accurate data is required over multiple time horizons, and the results can be utterly compelling as they highlight the really critical characteristics of behaviour (or player outcomes etc) that are proven to correlate over time with success outcomes. This is the real 'money ball' stuff that matters most. The existing conventional rugby stats profile is just someone's historical intuition of what matters most, but it's likely to be wrong if not subject to rigorous correlation analysis over large data samples. (That's not to say that individual player stats are useless, they can help assess and compare useful individual capabilities, but that should not be mistaken for surety about whether these stats in aggregate across a whole team and game are really the ones that matter the most, or in fact don't matter at all.)

Ultimately, the purpose of most stats is to be able to more reliably predict desired (or any defined) outcomes that matter to a particular constituency that will presumably benefit from those outcomes. Just taking the 'traditional rugby TV' stats profile of possession etc, went nowhere to properly explaining how, say, the Hurricanes thoroughly beat the Force in Perth recently, and that's just a very crude example of my point above.