Discover more from Criconometrics
Long ago I had developed what I had thought was a fairly unique way to visualise a limited overs cricket match. Now I find it everywhere!
First of all, welcome to all my new subscribers. After (I think) Jarrod Kimber recommended this newsletter on Substack, I’ve gained a steady stream of subscribers here despite not writing at all. So I feel like I owe something to all of you. And here is the something.
Story of a match
This morning I woke up to see that Punjab Kings had beaten Lucknow Super Giants in last night’s game. I was checking the scorecard on my phone, and all I saw was that Sikandar Raza (who had been plodding along when I turned off the TV and went to bed) had hit a 50, and M Shahrukh Khan had played a quick innings.
However, just by looking at the scorecard I found that I just couldn’t figure out the “story of the game”. Was it a close game? When did it turn? Did it go back and forth or was it one sided? If it was one sided, when did it become so? It would take a lot of effort to get all this information through the scorecard.
Thankfully, I didn’t need to struggle for this once I had opened my computer. The Cricinfo web page for this game had this helpful graphic on the top right.
This graphic easily tells what happened in the game. It was largely in balance, with Punjab having a slight advantage, till about the middle of the case. And then the game decisively swung Punjab’s way. There was a small setback in the middle, but they quickly wrapped it up.
Of course, there were more questions in my head - Sikandar Raza made a fifty but was it useful for his team? How important was Shahrukh Khan’s cameo? What was that blip in the chase? Nevertheless I was happy to see this graphic in Cricinfo.
I didn’t make it, but it was rather similar to what I had been building between 2012 and 2019 back when I had enthu for sports analytics. For example, this was I think my graphic on the first match of the 2018 IPL season.
Rather similar to what you see on Cricinfo today, but with “annotations”. Actually I realise that one of my old “apps” to tell the story of cricket matches is still “live” - except that it only covers ODI matches until 2019. A screenshot from that :
Thanks for reading Criconometrics! Subscribe for free to receive new posts and support my work.
My cricketing journey
My experience with cricket analytics began sometime in 2012 or so, when I decided to apply my learnings from pricing financial derivatives at Goldman Sachs to evaluate the “flow of a cricket match”. The idea was - if we can calculate the probability of each team’s win at the end of every ball, then the movement of the probability can tell the “story of the game”.
There were a few different models that I explored. One was a Monte Carlo simulator of the remainder of the game. Another was to model the cricket match as a random Markov Process, and then work backwards from all possible end states to the current state. Then I adopted this model called WASP that a couple of Kiwi scientists had built (and whose results would be shown on TV during matches played in NZ). I even had issues with WASP.
None of these models worked because I had fundamentally assumed that what happens in each ball is independent of what happened in the balls before.
There were other challenges, of course - how do you take into account the state of the pitch? How do you take into account who is yet to come to bat, and bowl? How to take into account the individual “matchups” (a concept that I think is grossly overdone in cricket analytics, but that is for another day)? The complexities are endless.
Nevertheless, through 2018 and 2019 IPLs, I regularly built and published (on my own blogs of course) the “story of each game” according to my algo. I built a web app where people could interactively track games. I found it hard to understand what happened in T20 games if my app hadn’t tracked it (then again not many others felt that way so I didn’t have ANY TRACTION). I even made some YouTube videos.
And then, as is common for me, I lost interest in cricket after the 2019 World Cup (I’d had too much of cricket, with the World Cup immediately following the IPL). I stopped writing this newsletter. I gave up my newspaper contract to write about cricket analytics. I took down my server where I would host this app. Since then, each big cricket event (IPL or T20 WC) I consider “resurrecting my server and restarting the graphics” but don’t muster the enthusiasm.
A rather common representation
In any case I don’t think I’m required any more to restart my server and produce graphics like this, for it has now become a rather common method to analyse cricket matches.
Cricinfo has it on the top of every match page now, for example:
Now I’m unable to find it but I remember seeing such graphics at the blog A Cricketing View. In 2018-19, I had met with an analyst who had served as head analyst for some IPL teams, and he too had some pretty good models to calculate winning probabilities after each ball.
Then, I recently came across the work of Tinniam V Ganesh who seems to have enhanced the model with “player embeddings” (which hopefully takes care of which batsman and bowler is yet to come). He has made an R package called yorkr where he has put together all his models and analyses.
I’m yet to really play around with the package - though I do plan to do it sometime before the current IPL is over. However, what gives me heart is that the sort of model and match representation I had built a decade ago is still alive and kicking.
And I was especially kicked to see one of Ganesh’s functions - he gives options for “interactive plot” and “static plot”, exactly like I used to do.
So I guess this is one opportunity for me to not-so-silently gloat. I had hawked around this model for a few years, trying to build a parallel career in cricket analytics. I’m a poor product manager so it never got traction. Then I got bored and moved on.
However, I’m happy to see that the representation I had come up with all those years ago still lives.
Unless something drastic happens, this newsletter is likely to continue in “passive mode” - where I write in it only occasionally when I really feel like writing. However, I encourage you to remain subscribed so that you can get this occasional piece in your inbox!