19. Twelve bowling strategies for a happy world
We can use clustering analysis to find the dominant ways in which bowlers are used by their teams
Traditionally, a bowler is classified based on which arm he uses to bowl, what speed he hurls the ball at, and (very broadly) what the ball does after it leaves the bowler’s hand. So you have “right arm fast medium”, “leg break” and “left arm orthodox” and so on. The problem with such classification, however, is that it doesn’t tell us anything about what the bowler's role in the team is.
A right arm fast bowler, for example, might be a great opening bowler and also bowl a few overs at the death. Another right arm fast bowler might be used by his team as a middle overs enforcer. One off-spinner might bowl a few overs at the beginning and finish off his spell in the middle overs. Another might come in to bowl only late in the innings.
So can we classify bowlers based on their role in the team rather than how they bowl?
For this purpose, we will revisit our old friend “K means clustering” (we’d last seen it in the edition on Balanced Lineups). For each bowler, for each ODI innings, we will look at the number of overs bowled in each 10-over slice of an innings. For example, in the game against England on Tuesday, Mitchell Starc bowled 3 of the first 10 overs, 2 of overs 11-20, 2 of overs 31-40 and 2 of overs 41-50 (overall only 45 overs were bowled in the innings. So 3-2-0-2-2 (the number of overs Starc bowled in Tuesday’s game) becomes his “signature” for the game.
We similarly collect vectors for all bowlers across all ODIs for which we have ball by ball data (starting in mid 2001, and the 1999 World Cup. Data is all from ESPN Cricinfo), considering only innings that lasted at least 40 overs. This gives us more than 22000 such vectors, and then we put them through a K-means clustering algorithm.
With some trial and error, we settle on 12 roughly equally sized clusters (the disadvantage with K-means is that you need to specify K).
Notice that not all opening bowlers (clusters 1,4,11,12) are the same. Not all middle overs bowlers (7, 9, 10) are the same. You have bowlers who bowl in the beginning and at death (11), right through the first 20 overs (4), in overs 21-40 (10) and only in overs 11-20 (5). There is also a significant enough cluster of part-timers who only bowl a couple of overs an innings (3).
Now let us look at the bowlers in the ongoing World Cup, and see what their dominant role in their team is, in terms of the number of innings played bowling a certain combination of overs. Some, such as Kemar Roach and Lungi Ngidi have been used in an equal number of innings in two different ways, and we will consider them in both uses.
Some pertinent observations based on this:
The three fast bowlers in India’s squad are identical, in terms of bowling at the top of the innings and at the death. There is no surprise that India plays only two of them in each game.
The World Cup has seen lots of leg-spinners, and they neatly slot into two clusters. Adam Zampa and Adil Rashid bowl mainly in overs 21-30 and then in overs 11-20. Rashid Khan, Yuzvendra Chahal, Shadab Khan, Imran Tahir et al are solid middle overs bowlers, bowling in overs 21 to 40. Kuldeep Yadav, interestingly, finds himself in a cluster with finger spinners and medium pacers
New Zealand, who have bowled with an unchanged lineup so far, don’t really have any death overs specialists. It is interesting that nobody has taken advantage of this so far. Maybe there is a case for bringing Tim Southee into the eleven.
Bangladesh have only one death specialist - Mohammed Saifuddin.
Mark Wood is an opening bowler but being used first change by England. There is a case for replacing him with Liam Plunkett, who is a specialist first change bowler. Maybe there is something to the correlation between Plunkett playing and England winning.
Australia has played each game with only four specialist bowlers. Again it is interesting that nobody has really taken advantage of their part-timers
The only death bowlers played by the West Indies are Kemar Roach (who hasn’t played that much) and Chris Gayle (who doesn’t bowl that much nowadays). This weakness has been well and truly capitalised upon by opponents
I have uploaded a “data puke” of all the bowlers to have played since 2001, along with the number of innings in which they’ve been used in different ways. You can find that here.
If you liked this, feel free to share this newsletter with whoever you think might like it. Subscribe if you haven’t already - this will come right into your inbox three times a week. And let me know if you have any comments, and if there is something you want me to cover in a future edition!