Monday, 15 December 2014

Context Blindness

Although the thread has now been removed from the Betfair Forum, Shapeshifter posted some thoughts on our old friend Jonny Grossmark.

I know some excellent sports data analysts, some who have worked with firms both here and america.  It is a whole different head-space.One I have known since high school, originally he dove heavily into the stock market then taking his love of numbers and incorporating it into sports.Conversations with him have taught me the one thing that is of utmost importance: each piece of information has to be put into context.
Six months ago, I came across a blog by Jonny Grossmark.  His name had come up in other circles that he had an approach he wanted to bring on-line
Always open-minded, I followed what he was up to, reading his blogs.  I passed this on to a couple of people I know (all published) at the same time.
In a nutshell, he looks at time of the goals and, by half time of a football match, has percentages of outcomes.
An interesting hypothesis but upon reading it, it falls into a couple of theories I have about data:
- As mentioned, context needs to be attached to the data
- Regardless of the context, sometimes too much information can actually make you ‘blind’ when compared to using ‘some’ of the information.
- (in relation to betting) A mistake that can be done is taking data and “molding” it to look like it offers an edge when, in fact, it is either one of two extremes: too general OR overthought.
I am not singling out Jonny Grossmark.  In fact, there are others who have not taken it as far as he has with their blogs (Someone named Peter Nordsted is offering a “click and win” approach based on the odds (i.e. Lay if between X and X in Y league).
But Jonny Grossmark’s approach ticks all the boxes that I think actually hinder research. That’s not to say that it won’t return a profit.  But:
- because he has focused in such a way, by “tightening the criteria”, it will find few opportunities.
- The relationships between some of the support data is unconnectible (leagues, team quality, and, as mentioned context (there is a huge difference between a team in a critical H2H versus two teams where one is already relegated while the other is resting players after securing top 4.  Yet they appear in the same data).
- and the fixation that once in the dressing room at HT, a team trailing 0-1 that was scored against at 9 minutes is different than a team that was scored on at 23 minutes.
Now the information has been incorporated into a web site that does “minute by minute” analysis of a teams performance.
As one that trades in-running, minute by minute can make you “slave” to the data.
I can’t let historical results over three years be factored into my decisions on value, pre-kick off or in-running.
You have to deal with the moment. The data might say otherwise IF YOU BET ON EVERY MATCH THAT FALLS UNDER THAT CATEGORY, but it is about being selective rather than being guided by numbers. I’ll take context within a moment in time any day of the week.
Subsequent comments suggested that Jonny G is banned from the Betfair Forum, an assertion which, based on my own brief experience with him, sounds entirely credible, but unfortunately after adding some thoughts of my own, the update failed due to the thread apparently being removed.

Context is a topic which many people overlook. Statistics such as Team x have won their last y games are, out of context, useless. Who were the opponents? What were the competitions? What was the criticality of the match to both teams? 

One recent example from football is Southampton, who recently won four EPL games on the trot without conceding a goal. That could be an impressive statistic, but unfortunately for anyone falling in love with Southampton's strength, the teams they had beaten were not Chelsea, Manchester City, Manchester United and Crystal Palace but Sunderland, Stoke City, Hull City and Leicester City. 

There's a big difference, and when Southampton did play three of those former teams, they lost all three. I'm hoping it'll be all four come Boxing Day. I jest a little, but the point I am making is a serious one. 

Points or goals scored always need to be weighted before they can be used in any meaningful way. I've written before that League Tables are, from a betting perspective, useless. In fact, worse than useless, and the same goes for form. L-L-L-L can be better than W-W-W-W because all Ls and all Ws are not the same. 

My Elo based ratings sometimes reward a losing team with a higher post-match rating and 'award' a lower rating for a winning team.

No comments: