A Review of the 2023 IPL

Runs Above Expected Runs is a measure of the outcome of a delivery relative to the average expected runs from a delivery. There are several ways to estimate the average expected runs from a delivery. The measure used here used three variables:

1. Scoring rate at the start of the delivery (e)
2. Legal balls Remaining in the innings (b)
3. Wickets in hand (w)

An extended discussion of the comparative merits of different ways of measuring expected runs is for a separate post. But, essentially, there are two broad approaches. The first seeks to identify the simplest, smallest set of factors which includes all the necessary features, the second seeks to build a large set of factors, making everything explicit. In cricket, given that the laws specify what is to be counted and how, the first approach is preferred here.

The problem with the second approach is that it becomes very difficult to keep track of what is being measured how often when one seeks to consider every imagined aspect of the context explicitly. For example, if one considers separate, explicit measures for things like ‘pressure’ or ‘pitch’ or ‘bowling quality’, then it is impossible to know how the exact weight of the difficulty of scoring or bowling conditions in the model, since each of these measures contribute. A further problem is that making such measures explicit invariably involves introducing arbitrary (albeit well-considered) thresholds, constants and/or weights, making the model even more contingent on these assumptions. The second approach is easier to ‘sell’ (i.e., its easier to satisfy the casual reader with the second approach because the reader sees that most of her concerns are apparently satisfied), but the first approach is significantly more rigorous, because it relies on the measurements available in the game.

Using the first approach, If the wicket is difficult to score (or easy to score), this will be reflected in (1) and (3). If the scoring rate is high and wickets are not falling, then the conditions probably favor batting. (2) and (3) are the two finite resources in a T20 (or ODI) match. The basic idea in this approach is that the score (and the match state) reflects the conditions.

The chart below shows a comparison for the expected runs estimation using two methods. The first is the misbah (which is used in the calculation here), which is a naive average of the record. Essentially, the average runs scored from all balls which fall in a particular (e,b,w) category (e is taken as the runs per ball upto one decimal point precision), gives the runs expected from such a ball. The second is a linear regression of the record. The linear regression is unsuitable for modeling this type of record because scoring rates do not have a linear relationship with all three factors (b,w,e) considered. The 4th, 5th, and 6th overs produce quicker runs in T20 than the 7th, 8th and 9th overs due to the powerplay rule.

More generalized estimators, such as a neural network regression model, or a decision tree like xGBoost, could also be used. Each involves making assumptions about the record. The assumption in the linear regression shown here, that the relationship is linear, is a bad one. The advantage of the naive average is that it reflects the available historical record. The disadvantage is that it tends to be noisier than a regression.

The two matches shown in the chart are both matches in which Virat Kohli made hundreds. On May 18th, he made 100(63) balls. The normalized expected runs (if the x runs are expected from a delivery, and r runs are scored, the then normalized expected runs from that delivery are given by (r-x)/x) from those 63 balls was 99.1. Kohli did 0.9 runs better. On May 21st, he made 101(61) balls. The normalized expected runs from those 61 balls was 95.4. Kohli’s score was 5.6 runs better than average.

Using this measurement, bowlers (arranged in descending order of balls bowled) and batters (arranged in descending order of runs scored) in the 2023 IPL are given in the charts below. The expected runs is given in two forms. First, the raw expected runs are given. This is simply the raw aggregate of the expected runs from the deliveries the bowler delivered and the batter faced. The misbah and jogi measure gives the aggregate normalized runs above expected runs. For a batter, the higher this number, the better. The a bowler, the lower this number, the better.

misbah: aggregate normalized runs above expected runs scored by a batter in an IPL season.

jogi: aggregate normalized runs above expected runs conceded by a bowler in an IPL season.

The names are a tribute to the final over of the 2007 World T20 final.

jogi: aggregate normalized runs above expected runs conceded over the course of the season. For a bowler, the smaller this number, the better. sr: runs conceded per 100 legal balls bowled.

misbah: aggregate normalized runs above expected runs scored over the course of the season. For a batter, the larger this number, the better. sr: runs scored per 100 balls bowled (a wide is not counted as a ball faced by a batter).

The top performing bowlers in the 2023 IPL are Matheesha Pathirana (-47.7 normalized runs above average), Mohammed Siraj (-30.8), Ravichandran Ashwin (-26.5), Axar Patel (-26.4), Naveen-ul-Haq (-22.4), Kuldeep Yadav (-22.3), Ravi Bishnoi (-21.8), Krunal Pandya (-21.0), Yash Thakur (-17.2), Varun Chakravarthy (-14.1), Rahul Chahar (-12.9).

The top performing batters in the 2023 IPL are Suryakumar Yadav (+89.3 normalized runs above average), Heinrich Klaasen (+70.2), Glenn Maxwell (+63.7), Yashaswi Jaiswal (+62.1), Shubman Gill (+43.5), Ajinkya Rahane (+42.1), Cameron Green (+39.5), Prabhsimran Singh (+39.5), Nicholas Pooran (+38.5), Rashid Khan (+35.0, over only 60 balls faced, an extraordinary return) and Faf du Plessis (+32.8).

T20 though, is a contest of provocation. The lowest scoring overs in a T20 innings are the opening couple of overs, and the 3-4 overs immediately following the power play. Readers will perhaps have noticed that the bowlers who top the aggregate list tend to bowl in these overs, because these are the overs in which batters are most likely to be content to take whatever the line and length of the bowler offers them, instead of playing the field.

So, if we consider all batters who faced at least 100 balls in the 2023 IPL, and all bowlers who bowled at least 100 balls in it, and consider the average normalized misbah per ball for the batters and the average normalized jogi per ball for the bowlers (as opposed to the aggregate misbah and jogi figures for the whole season as shown above), then, an IPL XI for the 2023 league is as follows:

Jason Roy (+0.161 misbah per ball)
Phil Salt (+0.156)
Suryakumar Yadav (+0.263)
Glenn Maxwell (+0.283)
Ajinkya Rahane (+0.218)
Heinrich Klaasen (+0.268) (wk)
Cameron Green (+0.138)
R Ashwin (-0.086)
Krunal Pandya (-0.083)
Naveen-ul-Haq (-0.128)
Matheesha Pathirana (-0.154)

This is perhaps not as well balanced an XI as possible.

There are six players who bowled at least 60 balls and scored at least 60 runs in the 2023 IPL. These are listed below. Six of them, from Rashid Khan to Cameron Green have made net positive contributions.

If one were to follow the example of the current World T20 Champions, and build an XI with all-rounders at its core, then, based on the full record, the IPL 2023 XI would be:

Phil Salt
Jason Roy
Suryakumar Yadav
Glenn Maxwell
Heinrich Klaasen (wk)
Cameron Green
Marcus Stoinis
Axar Patel
Rashid Khan (surprisingly, for his late order hitting as much as his bowling)
Matheesha Pathirana
Mohammed Siraj

I leave it to readers to construct their own preferred elevens keeping in mind the requirement for 7 domestic players in the comments section.