Why hitting is more optimal than batting in T20

I've recently become interested in the dominance of power hitters in T20, and the idea that constructing innings and investing in bowling may not be the most optimal way to play in T20. The distinction between hitting and batting seems to me to be foundational to the difference between T20 and cricket. In this article, I present evidence that shows that in T20, hitting (the willingness to risk dismissal readily in search of quicker scoring) is more optimal than batting (the desire to build an innings and score runs with certainty).

Last year I built a model that predicts the outcomes of limited-overs games. The result of the model is a win probability for each of the two teams based on the individual records, at the start of a match, of the players involved in the match. The approach used is that of a simple Monte Carlo simulation. A complete description of the method is available here. Readers ought to be able to build the model (or their own variant/development) based on the description.

The graph below shows the results of the model over 3750 ODI games. For each game, the model produces an expected win probability for each team as well as the expectation of a tie in the match. The expected win probabilities range from 0 to 100%. In the chart below, teams have been grouped by expected win probabilities within a range of 2% (the first group represents a group of teams expected to win 0-2% of the time, the second 2-4% of the time and so on). "Actual Win %" is an answer to the question "How many teams out of the teams expected to win within this range of probabilities actually won?" The graph below represents the expected win probabilities in comparison with the actual win percentages.

The orange dots represent the actual win percentages (on the Y axis at left) for the group of teams in each expected win probability (the horizontal axis). The black dots represent the number of games (see the second Y axis, along the right vertical edge) in each grouping.

For instance, in ODI games, of the 106 ODI teams given a 20% chance of winning the game, 24 (or 22%) teams actually won. Of the 128 teams given a 22% chance of winning the game, 37 (or about 28%) won. Of the 300 teams that, according to the model, have a 50% chance of winning, 156 (52%) actually ended up winning.

I did the same thing for 5564 T20 games (internationals, franchise leagues, domestic T20). The results are shown in the graph below.

According to the model, 50% of the 5564 T20 games are 60-40 (winning probability) or closer. The comparable figure for ODIs is 38%. In all, 63 ODI teams have a 20% chance of winning, and 174 have a 50% chance of winning. In contrast, 43 T20 teams have a 20% chance of winning, while 362 teams have a 50% chance of winning. T20 match-ups tend to be closer than ODI ones as a rule. This is partly to do with the fact that a large share of T20 games are in franchise leagues, in which teams tend to be more equally matched than is the case in the more traditional representative game.

Further, while the generally predictive trend holds, teams given very little chance of winning in the model still win about 20% of the time. This is because of the shortened T20 contest, which increases the effect of chance occurrences.

The model is a reasonably reliable predictor of the outcome of single-innings limited-overs games. It uses information that is available at the start of the game. The expected win percentage (accurate within a range of 2%) predicts the actual win percentage reliably. What this means is that when the model predicts that a team has a 39% chance of winning in one game and a 23% chance of winning in another, the record shows that the expectations are valid.

In the remainder of this post, this model is used to test how four different types of T20 XIs* perform. Each team is built to represent a different attitude to T20. Roughly speaking, each attitude towards T20 represents a view about its relationship with cricket as the sport has been conventionally understood for over a century.

As a guide, to pick batsmen, we used a batsman's career strike rate (runs scored per 100 balls) and dismissal rate (balls faced per dismissal) for all T20 batsmen who have made at least 1000 career T20 runs. The career records of T20 batsmen show that with a very small number of exceptions, T20 batsmen with high career strike rates tend to be dismissed more frequently, while batsmen who are dismissed less frequently tend to have lower career strike rates.

The Hybrid team below is intended as our best approximation of what today's conventional wisdom about T20 cricket would posit as a high-class team. Batsmen and Specialists are teams heavy on representatives of batting and bowling. Batsmen include four specialist bowlers, while Specialists have five. Given the quota of four overs per bowler, the Specialists team exists to test the possibility of having five specialist bowlers over 20 overs.

Here are the four teams. The bowlers are listed in brackets. The rationale for choosing bowlers is below the lists.

Hitters: (nine hitters, two bowlers) Chris Gayle, Virender Sehwag, Jos Buttler(wk), Luke Wright, Glenn Maxwell, Kieron Pollard, Yusuf Pathan, Andre Russell, Shahid Afridi, (Sunil Narine, Dwayne Bravo)

Batsmen: (seven batsmen, four bowlers) Quinton de Kock (wk), Virat Kohli, Joe Root, Shaun Marsh, Rohit Sharma, Kevin Pietersen, Brad Hodge, (Sunil Narine, Dwayne Bravo, Muttiah Muralitharan, Yasir Arafat)

Hybrid: (four batsmen, four allrounders, three bowlers) Brendon McCullum, David Warner, Suresh Raina, AB de Villiers (wk), Andrew Symonds, David Miller, Shane Watson, David Willey, (Sunil Narine, Dwayne Bravo, Muttiah Muralitharan)

Specialists: (six batsmen, five bowlers) Martin Guptill, Matthew Hayden, Michael Hussey, Aaron Finch, Owais Shah, MS Dhoni (wk), (Sunil Narine, Dwayne Bravo, Muttiah Muralitharan, Yasir Arafat, Dale Steyn)

If we consider the 30 batsmen and allrounders listed above, then the bowlers in the table below have bowled most frequently in T20 games involving these 30 players. Pollard, Russell and Afridi also make this list, but since they are already in one of the sides, they have been excluded from consideration for the bowling slots.

The bowlers are organised in the table above by their economy rate and then by their ability to take wickets. Since these are the bowlers who have bowled most frequently at the 30 batsmen in question, and since they are also some of the most prolific T20 bowlers in the history of the format, the high economy rates of Bravo, Arafat, Morkel and Mahmood is more likely to be a function of when they bowl in an innings than of their quality as bowlers. In order to achieve balance, where only two bowlers needed to be picked, the bowlers with the best and worst economy rates are chosen. Where three were required, the two bowlers with the best economy rates and the one with the worst were selected, and so on.

A league involving the four XIs described above produces the following results:

For the league as a whole, the Hitters win 57% of the time, Hybrid 55%, Specialists 45% and Batsmen 42%.

One could also select bowlers purely based on their economy rate and set aside the idea that economy rate has something to do with when a bowler bowls during the innings. The choice of bowlers can also be made simply based on wicket-taking rates. The bowlers with the best wicket-taking rate could be chosen irrespective of their economy rate. The table below gives the simulation results in each of these cases.

The Hitters come out on top overall. It is significant that the players considered to be hitters still get dismissed only once every 18 balls. In an average T20 innings, only seven of them would be dismissed.

In ODI cricket, this is not the case. If you look at the career records of some of the more adventurous contemporary ODI players, you will find that their terrific career strike rates also mean that they are dismissed once every 25 balls or so in an ODI innings (Maxwell, Corey Anderson and Luke Ronchi are examples; Buttler hovers around the 30 mark as well). This is significant because it means an ODI side made up exclusively of Maxwell-style hitters, would, on average, last only 40 overs.

Currently three distinct types of T20 batters exist. The first lasts more than 28 balls per dismissal and scores 115-130 runs per 100 balls. The second lasts 20-28 balls per dismissal, and scores in the 130-140 range. The third scores at nearly 150 runs per 100 balls and is dismissed once every 15-20 balls faced. It is a matter of time before a fourth kind - an uber-hitter - emerges. This player would last about eight to ten balls per dismissal, but score at more than 200 runs per 100 balls faced.

A recent piece by Ed Smith suggests that batting is evolving towards some form of "total" play. It is difficult to see how exceptional players like Kohli or de Villiers are emblematic of larger shifts. It is tempting but ultimately a mistake to treat exceptional individuals as prototypical inventions. Garry Sobers did not inspire an steady cohort of brilliant allrounder polymaths in the West Indies.

Batting of the quality and range of de Villiers and Kohli is similarly unlikely to become the norm. On the other hand, their respective approaches could be said to be exemplary of the past (Kohli) and the present (de Villiers) of T20 play. The changes in hitting technique in T20 come with more frequent dismissals. De Villiers was dismissed once every 26 balls in T20s in 2016. Kohli, who is approaching the peak age for elite international batsmen, was dismissed once every 64 balls in 2016, while scoring 21 runs fewer per hundred balls faced than de Villiers. The simulation results presented in this article show that while de Villiers' approach is more optimal than Kohli's it does not quite represent the cutting edge of play in T20. Currently the World T20-winning West Indians embody this cutting-edge approach best, and the evidence suggests that T20 could easily accommodate even more belligerence.

*The four teams were compiled with the help of Subash Jayaraman