# Statistician picks winning cricket teams for World T20

A Canadian statistician has picked optimal lineups for England and South Africa’s Twenty20 World Cup cricket teams.

Twenty20 is the shortest form of cricket and on March 8, 2016, sixteen teams will compete in India for the World Cup title. 11,500 kilometers away, however, a statistics professor and his students already have a good handle on what will give each team the winning edge.

Professor Tim Swartz from Simon Fraser University in British Columbia, Canada, devised a data-heavy strategy to optimize international team lineups.  Using a match simulator, he uses batting and bowling probabilities to estimate the best outcome. His results were published the Journal of Statistical Computational and Simulation in January, 2016. His earlier study brought in a hypothetical 300 percent betting return in the last Twenty20 World Cup in 2014.

As international Twenty20 sports teams get ready to take to the crease, we ask Swartz for an explanation – and sample – of his 2016 results.

ResearchGate: What got you interested in this field of research?

Tim Swartz: I have been interested in cricket analytics for a long time. Cricket is by some measures the world’s second most popular sport, but I believe it has been underexplored in terms of analytics when compared to soccer, baseball, basketball, and American football. The Archive section of the www.espncricinfo.com website gives ball-by-ball commentary on all international games dating back for decades, so there are analytical opportunities in cricket with this detailed and comprehensive collection of data. We are the first to make use of detailed ball-by-ball data, which demonstrates that despite the money in Twenty20 cricket (especially the Indian Premier League) its analytics lag behind other major sports.

RG: What are the current methods or metrics used to evaluate Twenty20 team players and lineups?

Swartz: Team lineups are chosen by team selectors and I believe that every country has its own approach. Past history and current form are analyzed, as well as many of the standard measures reported in cricket, including batting and bowling average, for example. It is likely that special circumstances such as pitch conditions are also considered.

RG: How does your research on optimal lineups improve these decisions?

Swartz: Our approach provides the most modern, data comprehensive and systematic strategies available. And by implementing our strategies, teams may discover ways of improving performance and providing their team with a competitive edge. We ask how much better is it to have a specified player in the lineup compared to a standard player. It isn't batting average or bowling average that directly causes a team to win or lose; it is run differential. Run differential is the total number of runs scored by one team in comparison to the total number of runs scored by their opposition. By focusing on run differential, we believe we are measuring what is truly important in this complex sport.

RG: Can you explain how you determine optimal lineups?

Swartz: Our method relies entirely on data. Each batsman has a set of characteristics that describe the rate that they are bowled out, and their run scores (0's, 1's, 2's, for example). Similarly, each bowler has bowling rates. These rates are modified according to the game situation (opposition, overs completed, batsmen caught or bowled out, and score). We then search over the space of potential lineups (team selection, batting order, and bowling order) to find an optimal lineup (i.e. one which provides the greatest run differential). The search algorithm requires extensive simulation of matches, and we use the same simulation approach to assess players.

RG: Have you used your research results to place bets in cricket? Did it work?!

Swartz: I have never placed a bet on a cricket match. However, in our paper “A simulator for Twenty20 cricket,” there is an example of potential betting during the 2014 Twenty20 World Cup. Here, we wagered a hypothetical \$100 on 15 specified matches. The exercise resulted in a hypothetical profit of \$399.

### South Africa

The numbers in parentheses are the number of overs bowled.

Swartz: This new lineup should be better (in terms of run differential) by 21 runs on average.

RG: Based on your knowledge of the game, would you make further changes?

Swartz: We do not believe that JA Morkel is being utilized to his full advantage, and he would help South Africa greatly.  He was not included in our optimal lineup because South Africa did not include him in their list of players in the February 2016 England in South Africa tour. Not only would we advise that JA Morkel play, but we would have him bat much earlier in the lineup than where he is typically used. The reason for this is because getting a batsman is out is less important in Twenty20 than in the other forms of cricket (One Day International and Test matches). Therefore, batsmen that score runs at a high rate should be given more opportunity to bat.

### England:

This new lineup should be better (in terms of run differential) by 17 runs on average.

RG: Are there any surprises in your optimal lineup? Would you make further changes?

Swartz:  There is obviously information that is not available to us. For example, a player could have a niggling but unreported injury. If I were a team selector, I would use our data driven methods to obtain an optimal lineup, and then I might modify it according to the special knowledge that is available.

Feature photo courtesy of Marc