Forecasting goals

The other day I was invited by a friend to fill out a prediction bracket for the upcoming World Cup for a pool he had organised. In many ways relating a one off sport tournament to the analysis of  climate systems is a bit of stretch. However, while doing so I noticed that there were many parts of the exercise which felt, well, familiar.

Without some background, available information can be worse than meaningles:

I haven’t been following the World Cup qualifiers this year, so I started with a (to me) arbitrary list of names and numbers. The numbers, current FIFA standings and gambling odds (set by who?) were pretty opaque at first. The country names were however, of course instantly recognisable. There are all sorts of associations that could influence selection criteria, vague memories of previous World Cups, national economic standings, personally held cultural associations (essentially culinary preferences), military histories, and so on, each more arbitrary than the last. An hour of familiarising myself with this year’s teams and other peoples’ expectations for them transformed all that and gave me some actual context to interpret the rankings and to start to assess what seemed likely based on current situations. If when initially handed the piece of paper to fill out I’d been told to return it in the next five minutes, well, I could have easily have used various associations to make decisions, but they would be pretty well unrelated to the facts of this year’s tournament. The same is true when working with climate and weather, the subjects address phenomena we encounter and hear theories about every day. Useful analysis though often pertains to very specific and nuanced cases, which may produce results different than we may expect from a general sensibility. It can be very difficult to work out what of our general knowledge is applicable to the questions as hand, In part that’s because…

The relevancy of the statistics is crucial:

The information I found on the FIFA website was a mixed bag for predicting outcomes. There are long lists of stats of how different national teams have fared against each other over the last hundred years. That sounds useful, except many of these matches happened before any of the participants in the 2014 World Cup were born, let alone playing for their respective teams. That Portugal’s Ronoldo is injured but recovering, that the USA is coached by the same man who gets a lot of credit for the strong German showing in 2006, that Bosnia and Herzegovina has become of major exporter of talent for top professional teams across Europe, are all relevant. The amount of money various countries are willing to invest in their national teams, especially given Brazil’s reported difficulty in consistently preparing infrastructure for the games could easily be a telling factor.  However, the result of a playoff in 1962 is just story telling… except of course that those are the games that would be studied by the current generation of players and coaches.  So not only are you asking how similar are current and former situations, but also how much memory is in the system becomes an important factor. Similar questions apply to the study of climate and weather. How well do obtained results describe different regions or time periods?  Is the current climate system an appropriate benchmark for describing what an altered climate will look like a hundred years from now?  What past climates are useful as analogues for the present or future?  What key factors make the current climate “the way it is”? To what degree is our present or future climate predetermined by past states?  Finding ways to investigate this an be difficult because…

The big is made from the small:

So one team is favoured over another, why? Really it comes down to individual players, the situations they may be presented with and how they are expected to respond to them. When you consider the intricacies that determine the events of a few minutes of play, and then sum those up to produce a match with a decided outcome, then consider multiple matches with constantly shifting participants and strategies, and then allow for the influence of these changing standings on the moment to moment play (downward causation?)… well that sheet of paper with the tournament bracket on it starts to seem pretty divorced from what it refers to. That sheet of paper, however, defines the shape of all the events that the world will observe over the next few weeks. We need to be careful not to stretch the metaphor and say that the bracket is the ‘climate’ of the World Cup. The World Cup is only a single realisation of a chaotic event, describing the climate system is more a question of what sort of narratives would dominate if the World Cup could continue indefinitely, and how those would change if the rules of the game were to be altered over time as well. What’s relevant here is that in filling out a bracket for the World Cup I wasn’t asked to predict the scores of the individual matches, or who would score those goals, or at what point during the game, or under what circumstances. The exercise isn’t writing a script for the entire multi-week event, no one would expect you to do that, just say who will win each match-up, despite that who wins/loses is determined by exactly the set of occurrences that we casually assume is  beyond prediction. Similarly, producing a description of every weather event for the foreseeable future is not realistic. Rather we’re forced to consider a representative sample of all possible sequences of events and their implications.  The question, like in the case of the World Cup prediction, is not narrating minute by minute events (“He shoots, he scores!”), but by looking for indicators of the their potentiality (team schedules, tactics, health of players, etc) and how this in turn affects the potentiality of different outcomes (winning/losing a match). This is very intuitive behaviour when discussing common subjects such as athletics, but is sometimes baulked at when evaluating climate and weather. Perhaps this is because it implies…

There’s no one right answer:

I have pretty limited expectations that the actual tournament results will have much in common with what I have predicted.  Being forced to fill out just one set of predictions was a rather frustrating experience.  There are a lot of unknowns. Sure, I doubt very much that some teams will leave the opening round (Paddy Power is currently offering twelve Rand for every one you bet on Australia progressing) and it’s expected that Germany will perform very well until they meet a South American team that emphasises ball control over position tactics (some things are just givens), but outside of that I’m not sure how a lot of the matches will play out. Much of that is a result of my own lack of information, but there are many match-ups where the height of professional expertise is measured as having the  ability to recognise that it could go either way. There are many potential narratives, some more “reasonable” than others, some more informed than others, but no a priori correct answer. Fans fill out brackets, professional analysts give odds. This is not a failure of the deductive process, Sherlock Holmes can’t retreat to his “mind-palace” and return with the irrefutable result. Even with a game of chess the most that can be done is to compute all the possible scenarios, you can’t know which ones will take place until they do, and a chess board is a very closed and simple system compared to the “beautiful game”.  Or climate for that matter, getting a bit excited about the football, starting to forget that this is supposed to be a metaphor. The situations are similar though, analysing and predicting climate isn’t about one fixed imperative narrative, its about the types of things that can happen, and how this potential evolves with time. We can’t actually know all these things, but we do have the tools to discuss where our limitations are, and to hedge our bets appropriately.

Of course the popular 1998 film Lola Rennt (Run Lola Run), that looked at similar ideas of the multiple possibilities we face on a daily basis, caveated its discussion in terms very appropriate for the current discussion: “Ball ist rund. Spiel dauert 90 Minuten. Soviel ist schon mal klar. Alles andere ist Theorie. – Ball ab!”