In 2017 I completed my Masters degree under the supervision of Dr Brendon Brewer. My research looked to tell a more meaningful story behind a cricket player’s batting average. Using Bayesian statistical techniques, I explored more in-depth methods of quantifying a cricketer’s batting ability than the simple batting average. More specifically, I built statistical models which describe how well a batsman is playing at any given point in their innings, allowing us to quantify the cricketing idea of a batsman ‘getting their eye-in’. The primary focus was on Test match cricket, with wider applications to 4-day First Class cricket. Using these models, I explored the plausibility of popular cricketing superstitions from a statistical point of view, such as the commentator’s favourite, the ‘nervous 90s’.

ABSTRACT: Cricketing knowledge tells us batting is more difficult early in a player’s innings, but gets easier as a player becomes familiar with the local conditions. Using Bayesian inference and nested sampling techniques, a model is developed to predict the Test match batting abilities of international cricketers. The model allows for the quantification of players’ initial and equilibrium batting abilities, and the rate of transition between the two. Implementing the model using a hierarchical structure provides more general inference concerning a selected group of international opening batsmen from New Zealand. More complex models are then developed, which are used to identify the presence of any score-based variation in batting ability among a group of modern-day, world-class batsmen. Additionally, the models are used to explore the plausibility of popular cricketing superstitions, such as the ‘nervous 90s’. Evidence is found to support the existence of score-based variation in batting ability, however there is little support to confirm a widespread presence of the ‘nervous 90s’ affecting player batting ability. Practical implications of the findings are discussed in the context of specific match scenarios.

Click here to read thesis titled “The nervous 90s: a Bayesian analysis of batting in Test cricket”.

ABSTRACT: At a glance, data is more meaningful when presented in graphical form. This project explored innovative methods of automating the display of catch data for large-scale conservation projects. High priority was given to developing methods that allow users to interact with their data, affording them some control over the graphics that are produced. Two interactive applications were developed that allow conservation volunteers to select the data they want to view and how to view it. After a day in the field, volunteers are able to use these applications to see their day’s work summarised on a map or graphic. These graphics highlight the positive impact their efforts are having on the local environment, keeping volunteers motivated and engaged in their work. Various methods of improving the automation of these graphics are outlined, as well as other practical uses of these statistical applications.

Click here to read dissertation titled “Graphical applications for large-scale conservation projects”.

The batting rankings are based on the models developed as part of my Masters and PhD at the University of Auckland. The model accounts for a player’s recent form, venues of matches played in (i.e. home, away or neutral) and whether the player was batting in their team’s first or second innings of the match. The data support the general belief that players tend to score more runs when batting in their team’s first innings of a match, at a home venue.

The “predicted average” is the the number of runs we expect the player to score in their next Test innings, assuming their next innings is played at a neutral venue and it is unknown whether they are batting in their team’s first or second innings of the match. The official International Cricket Council (ICC) ratings (and world ranking #) are also provided for comparison. The ranking of players is generally similar between the two methods, although there are a couple of notable differences.

Firstly, our model rewards players who are able to overcome the “getting your eye in” process and remain on a not out score, while the ICC ratings simply provide not out innings with a “bonus” that we susepct is too low. For example, Rohit Sharma is currently ranked 11th in the world by our model and 54th by the ICC. Sharma has a large number of not out scores between 50 and 100, suggesting he frequently overcomes the difficult “getting your eye in” process, but for various reasons, has not had the opportunities to convert these not out innings into big scores.

Secondly, the ICC ratings tend to place more emphasis on recent innings compared with our models. Our general findings suggest that there is no evidence to suggest that recent form is a significant predictor of current batting ability for the majority of players. Instead, we believe a player’s underlying ability tends to change slowly over time, rather than erratically between innings as a direct result of recent performances. It is unclear whether the ICC ratings attempt to provide predictive accuracy of ability, or instead tries to formalise expert judgement about who is in and out of form. These two goals may not be entirely compatible.

Finally, while both methods provide a general indication of batting ability, by measuring underlying batting ability in units of a batting average, we are able to maintain an intuitive cricketing interpretation when comparing players. Instead of concluding “Steve Smith is 26 rating points better than Kane Williamson”, we can make more meaningful probabilistic statements, such as “we expect Steve Smith to outscore Kane Williamson by 9.2 runs in their next respective innings”, or “we expect Steve Smith has a 54.1% chance of outscoring Kane Williamson in their next respective innings”. In both statements we are assuming a neutral venue and it is unknown whether they are batting in their team’s first or second innings of the match. Of course, we can update these estimates to include specific match information, if we know the venue of the next match and whether the player is batting in their team’s first or second innings of the match.

Here you can find an application that allows users to visualise how the models estimate batting ability on two scales:

1. Short-term changes in ability that occur during an innings due to the “getting your eye in” process

2. Long-term changes in ability that occur between innings, over a playing career, providing an estimate of a player’s batting career trajectory to date, as well as a prediction for their current ability. These estimates are what are used to compute our batting rankings