Customers

New Zealand Cricket

Cricket lovers the world over are amongst the keenest of amateur statisticians. Nearly 150 years ago, however, Englishman and Sussex fast bowler John Wisden published his first annual Almanac and turned cricket statistics into a business.

Ball-by-ball data abounds, and while most of it is used for the game’s followers and commentary for cricket writers and broadcasters, New Zealand Cricket (NZC) has turned data into powerful ammunition, using software from SAS.

Dr. Peter Mayell, who is Cricket Technology Manager at NZC’s High Performance Centre in Christchurch, says that, “A few staff members at New Zealand Cricket over the years have flirted with looking at big sets of numbers to process and see if they [could] provide an insight but this hadn’t led anywhere.”

Before Mayell joined NZC in 2005, there had been no real effort to scientifically extract value from match data. With his appointment, however, it was decided to develop the means to do so for One-Day International Cricket.

New Zealand Blackcaps logo As with any modelling, the bigger the history that goes in, the more valuable is the reporting that comes out, and while NZC had collected high level video analysis data for six years, New Zealand’s BLACKCAPS typically play fewer games each year than other international teams, and so the data volume was not large.

The initial data was, in fact, taken from just 180 one-day matches. This was supplemented with data from another 80 games which had not involved New Zealand and is now continuously added to, series-by-series. Mayell says that by the close of the 2007 World Cup, “We had enough data for statistical significance.”

The software used is SAS Enterprise Miner, and Mayell secured the services of Scott Brooker – an economics graduate who is now pursuing his PhD at the University of Canterbury – to develop the models.

The first steps were taken by Offlode, a Wellington-based consultancy and member of the SAS Alliance with offices also in Auckland, as well as Sydney and Melbourne in Australia. Offlode specialises in data mining and analysis for marketing and fraud detection applications.

Offlode had been approached by NZC in 2004 and knew that a lot of match data had been collected but that it was neither in an organised state nor suitable for analysis purposes. A number of separate databases had been created, whereas the obvious need was for a single data source, added to match-by-match over time. Offlode built a single data mart for NZC and created the model concepts in a way that would make them easy to use for coaches and others with no previous experience of sophisticated data analysis.

Offlode recommended the SAS tools and suggested training a student to do the programming work in order to minimise costs. Offlode generously provided Brooker with six weeks of training in SAS Enterprise Miner.

The models are designed to help team coaches and the captain determine their match play strategy in real-time; to make the decisions – at any point during the innings – most likely to produce a win. Such decisions might include changing the batting order at the fall of the next wicket or sending a message to the middle suggesting a bowling change.

Mayell explains, “While the initial data was limited to just how each ball was bowled, by whom and what happened to it, the database is being incrementally enriched with other important data such as pitch condition, venue, and players’ strengths and weaknesses. The early models were at country level only; matches in India, matches in England and so on.

“For example, there was no ground-by-ground data available for New Zealand’s first use of the models during the 2006 Champions Trophy in India so an allowance was made for the hypothetically ‘typical’ Indian ground. This was subsequently upgraded to groups of similar grounds.”

The ultimate is detailed time-of-day and time-of-year data for individual grounds, fine tuned with real-time details such as humidity and light; which of the 50 overs is currently being bowled; and the records of the players of both sides involved at that time.

Mayell gives an example. “Let’s say you’re batting second and chasing 250. You are 30 for two after 12 overs and the model – having considered all the variables – is telling you what your final score is likely to be; whether you look like winning or being a certain number of runs short of your target. Based on that intelligence, the team coaches will determine the best short-term strategy. It might be, for example, to change the batting order and advise the new batsman what he and his partner should do for the next five or 10 overs.”

“Different models apply for when you are batting first and when you are bowling. In each case, the system predicts an outcome based on the point-in-time situation and all the variables, and the coaches use this prediction to determine the best course of action.”

He adds, “There is a lot you can do. We’ve seen aspects of data mining that we never even envisaged as being possible when we embarked on this exercise.”

By the end of the 2007 World Cup, New Zealand was probably ahead of other cricketing nations in its use of data mining. Mayell is aware that the Australians and Indians are “doing something” but modestly admits, “I think it’s fair to say that we are at least up there as a leader.”

Given the fierceness of international competition, NZC is naturally unwilling to reveal too much detail about its system but concedes that for every single one of the 600 “good balls” in each one-day match, there are at least six variables fed into the database. This is ball-by-ball data, additional to the many high-level factors such as pitch condition.

NZC is now thinking about the new Twenty20 version of cricket and while there is not yet much data available, Mayell expects the data set to be of a meaningful size in time for the next Twenty20 World Championship in 2009. Modelling for the “real version of the game” – the traditional five-day Test Match – is, says Mayell, a much bigger challenge that NZC will address in due course.

While adopting SAS software was primarily intended to give New Zealand a competitive edge in matches, it is also helping to develop players’ skills. The day after each game, players sit down with the BLACKCAPS coaches and specialist coaches from NZC’s High Performance Centre and address the “player contribution” part of the system; examining each player’s performance against expectation. Mayell says, “Holding yourself accountable to the team is a big part of BLACKCAPS culture and data mining now helps players better assess themselves and each other.”

Appropriately, the last word belongs to John Bracewell, BLACKCAPS Coach and former Test and One-Day International player of the 1980s. “Thanks to data mining, technology is helping us to improve our international competitiveness for today and better develop our skills for the future. The fact that we have won games as a consequence of adopting this technology speaks for itself.”