Follow-On Strategies in Cricket

Test Cricket has returned to the international stage after a brief hiatus due to COVID-19, with England taking on West Indies. Before this interruption in the Test Calendar, World Champions England had just returned home after a 4–1 drubbing of South Africa. In the final test, despite establishing a first-innings lead of over 200 runs, England Captain Root decided against making the Proteas follow-on. Instead, England took the crease and quickly built a commanding lead, before bundling out South Africa to win the Test Match.

A couple of decades earlier, it would’ve been considered blasphemous to not enforce a follow-on. However, this strategy seems to be increasingly popular these days, with captains frequently preferring to forego a follow-on advantage. This makes one wonder about what changed the mindset of international teams so drastically. As any follow-on verdict weighs crucially on the outcome of a test match, I decided to investigate this further by broadly answering the following questions:

· Have the follow-on strategies really changed in the past few decades, or is this just a mistaken belief?

· If Yes, what was the driving factor behind this change? Can we identify any specific inflection point in the timeline?

· Can we establish any patterns which indicate the best strategy for dealing with follow-on advantages?

So let’s dive right in!

Data Acquisition and Cleaning

The data for this analysis was scraped in its entirety from ESPN Cricinfo, using the Requests and lxml libraries in Python. The raw data contained team-wise information on every official International Men’s Test Match ever played, at an innings level granularity. This was later wrangled into a convenient format for further analysis. Without going into much detail about the painstaking cleaning process, it should suffice to say that the main parameters extracted for each test match were:

· Innings Scores

· Innings Wickets

· Innings Run Rates

· Declaration (Y/N)

· Option for Follow-On (Y/N)

· Follow-On Enforced (Y/N)

· Teams, Venues, Dates and Results

To make a fair comparison, I’ve only included the matches played since the introduction of 5-Day Tests in 1980. This encompasses nearly 40 years of Test Cricket experience, which should be good enough to derive meaningful conclusions.

Q1. Have the follow-on strategies really changed in the past few decades, or is this just a mistaken belief?

Let’s examine the variation of follow-on strategies with time in Test Cricket:

Since the frequency of Test Matches has increased significantly in the past 2 decades, we observe a concomitant rise in the number of follow-on opportunities. Note that the gap between both curves has widened considerably in the new millennium.

Clearly, teams are hesitating to enforce follow-ons in modern cricket. In fact, on average, teams have rejected nearly 50% follow-on advantages handed to them in the past decade!

Q2. What was the driving factor behind this change? Can we identify any specific inflection point?

A sudden change in any playing strategy can usually be traced back to an example gone dreadfully wrong. Mistaken follow-on decisions instantly reminds one of the legendary Eden Gardens Test in 2001, in which the Indian team came of age with a heroic effort to beat Australia. This remains one of only 3 instances in Test Cricket (only one in the modern era) of any team winning a test match after following-on. This also seems to conveniently split our dataset into equal halves of ~20 years each. I tried using this match as a benchmark to compare its effect on follow-on strategies.

Aha! Turns out that until the Eden Gardens Test Match, 92% of follow-on advantages had been enforced. However, this figure drops considerably to 56% in its aftermath. It can be safely said that this Test majorly influenced the approach of international teams to follow-on strategies. While one could argue that the introduction of T20 cricket, renewed attention to player fitness, and improved data-driven decision making has contributed to this pivot as well, the impact of this test match is certainly significant.

Q3. Can we establish any patterns which hint towards the best strategy for dealing with follow-on advantages?

This part was slightly tricky. I tried constructing a logistic regression/classifier model with my limited knowledge of data science. However, it was hard to identify any visible trend. Which sort of makes sense. The decision to enforce follow-on depends on several factors, a lot of which are very specific to the circumstances of the match being played.

The personality of a captain heavily bears on this decision, and captains themselves keep shuffling around every few years. While Kohli typically adopts an aggressive strategy, by rejecting follow-on advantages to quickly set a fourth-innings target, Williamson would be more likely to enforce the advantage and let the game play itself out. The dynamics of the test series directly influences this decision, as it could determine whether a team is looking for victory at all costs, or is it content with playing out a draw. Rain could significantly reduce the playing time in a test match, thereby changing the input variables for any such strategy. Most importantly, the expected condition of the pitch plays a huge role in making this choice. Taking guard on a deteriorating final day pitch is a challenging prospect for the best of batsmen. The team batting first would prefer to preserve the order of play if it means transferring this risk to the opposition.

Due to the abundance of such dependencies, and the difficulty in quantifying them, I’ve tried to keep this analysis simple and focus on just 2 parameters which majorly influence the follow-on decision:

· The first-innings lead established

· The time consumed in the test match

Since the dataset does not contain timestamps for the match duration, I’ve taken the cumulative first-innings overs bowled as an indicator of the time consumed in the test match. However, this does not account for weather interruptions and other such random events which cut short the playing time. As a result, it wouldn’t be the best way to capture the time remaining in a test match. Nevertheless, on most occasions, the overs bowled gives us a fair approximation of the time spent and shall be sufficient to observe any rough trends in the data.

I’ve ignored the pre-2001 matches for this analysis since teams would almost always enforce this advantage prior to the Eden Gardens Test. Also, I’ve only considered test match victories for the subsequent evaluation, as the goal is the come up with effective follow-on strategies. A draw or loss would not be a successful outcome of our decision.

On trying to plot our control parameters against each other, this is what I found:

I shall refrain from defining any decision boundary here, as a significant overlap exists, and any attempt to delineate the orange and blue classes would unnecessarily torture the data. Also, an overlap should exist, as there would be several scenarios where either option, enforcing or foregoing a follow-on advantage would ultimately result in a victory. Being granted the option to enforce a follow-on in itself implies that the team is in a dominant position to win that test match.

A few trends are distinctly visible in this plot.

· Successful strategies generally involve enforcing the follow-on, when a significant lead has already been established. There is only 1 documented instance of any team foregoing a follow-on advantage after building a 400+ run lead. Given that the highest 2nd innings run chase in test history is 418, it would be safe to assume that at a 400+ target is extremely unlikely to be chased down. Irrespective of the time remaining in the test match, it would make more sense to enforce the follow-on advantage, instead of compounding the lead even further.

· Also, teams prefer to forego the follow-on advantage when a lot of time has been spent in the test match, or when there is little time remaining. This represents the aggressive follow-on strategies adopted recently and is a testimony to their successful results. The main motive behind such a strategy would be to quickly set an achievable target, allow the opposition to sniff a chance of victory, entice them into playing high-risk cricket, in the hopes of seizing quick wickets and securing a win. A side benefit is that it gives the dominant team’s bowling attack an opportunity to rest, and therefore restore their stamina for a fourth innings kill.

Conclusion

Evidently, follow-on strategies have changed over the past few decades. A turning point can be traced back to the Eden Gardens Test in 2001, after which teams preferred to forego this advantage on several occasions. Trends post-2001 suggest that teams prefer enforcing the follow-on when armed with a huge lead, and rejecting it when there isn’t much time remaining in the game. The results of this analysis align with conventional wisdom and provide statistic validation for frequently observed trends in modern cricket. However, several practical factors affecting this decision have been ignored in favour of simplicity. A few of them have been listed in this article, while several others could be established on probing further. The primary objective of writing this is to bring the attention of the data science community to patterns in follow-on decisions. Hopefully, this initiates a discussion on the subject, eventually leading to the deployment of more advanced analytical techniques to anticipate the best follow-on strategy for a given situation.