Language of Advanced T20 Analytics

One of my favourite T20 articles is Jarrod Kimber’s piece on the new vocabulary of T20. He explains how “back-room staffs and coaches are already looking for alternate ways of evaluating players” and then goes on to offer a few solutions. Some were brilliant, and I fully intend on stealing them – True Economy and True Strike Rate are much more elegant ways to describe many of the metrics that I already use. Others I remain sceptical – it is hard to see Unconsolidators catching on

Within American sports coverage, a person can lose themselves in a swarm of new-age metrics and acronyms. Baseball’s acronyms make it almost impenetrable for the uninitiated. Basketball statisticians have created True Shooting, Real Plus-Minus, Effective Shot Quality. Football analysts (soccer) now have expected goal models (xG) to reduce the noise around scoring stats

Long-form cricket has been ahead of other sports in this regard. Batting and bowling averages have always been sensible ways to assess player performance and have been in use since 1867. With so much time available in each match, the objective is almost always to maximise/minimise the number of runs scored per wicket

Then one-day matches forced us to also pay attention to strike rates. But it is the rapid growth of T20 for cricket to finally realise that more advanced metrics are needed

My Twitter feed frequently features a wide range of new advanced cricket metrics. Some are better others but the primary issue with each one is a lack of transparency. Everyone knows exactly how to calculate strike rate but, no matter hard we try, measures like True Economy will always remain something of a black box

But that should not disqualify them from being used. In any domain, a new advanced metric is valuable as long as it passes three criteria: it should tell us something useful, it should tell us something that existing metrics do not, and it should be easy for it's intended audience to understand

Venn diagram showing the three criteria that any new metric should pass before being introduced

The metrics listed below probably pass on all three accounts. If any fail, it is likely on that third criterion that people need to understand what is being measured. The hope is that this will change as analysts and decision-makers become more familiar with the terms

True Economy and True Strike Rate - @ajarrodkimber

The objective here seems transparent enough – we are trying to assess how a bowler’s economy or a batsman's strike rate compares to what we should expect. This allows us to compare middle over spinners and death specialists side-by-side with a single metric

The question is how to define what ‘expected’ is. In his article, Jarrod suggests that we should be adjusting based on the over number. But there are plenty of other factors that we could consider: required run rates, wickets lost, venue, opposition. All of these affect run rates

Expected T20I average / strike rate - @SAAdvantage

Here we don’t have the same problems. Dan uses metrics that are extremely clear on what they are trying to measure. And whilst he often refers to a mysterious “unique algorithm” he is open about what that algorithm involves... he basically takes as much data as he can get his hands on and then adjusts according to the difficulty of the competition

Batting Impact - @CricViz

I'm slightly resistant to the name but, in context, it is usually clear what this is measuring: the overall impact of a batsman on the final run total

However, it is hard to find any detail on how it is calculated. I have no reason to doubt its validity but it is hard to interpret a metric when you don't know what goes into it

Runs Added - me

Like Batting Impact, the objective of Runs Added is to identify the impact of a player on their team’s final run total. I try to be as transparent as possible in how I calculate it – regarding both my expected runs model and how I combine data to create a single player rating. As with Dan Weston’s metrics, the final player rating is expressed as what I would expect their Runs Added to be in one particular competition (IPL), after adjusting for the level of difficulty in other leagues

But I don't limit myself to Runs Added. That would be foolish. For example, it is often useful to explicitly talk about strike rates in certain phases of the game. Teams should look for bowlers with excellent Powerplay strike rates and excellent Death economies. Communicating in those terms helps to convince decision-makers that the analysis makes sense

However, I am sceptical that differentiating between the phases is entirely sensible. With such low samples, we are in danger of isolating noise in the data and missing the bigger picture. I believe that there are just five metrics that I need to understand the overall value of batsmen and bowlers to teams and franchises. They might miss some small nuance but they get me 90% of the way there

The five key metrics for bowlers and batsmen that I need to judge player value: sample size, True Economy, True Strike Rate, Runs Added

My intention is to focus future articles on these metrics. Simple measures like strike rates and averages will obviously get used too but I want to set myself a standard for how I communicate my work. This is the language that I plan to use

To read more about what is included in my Runs Added model, see here. For individual matches it is based on ball number, wickets lost, and required run rate. When creating player ratings, I also adjust for venue, competition, sample size and recency
To read about what I include in my “True” models see here