Mo Bobat and County Cricket

Cricinfo has this interview with ECB "Performance Director" Mo Bobat.  Bobat makes an interesting claim about county cricket, "Take something like county batting average. We know that a county batting average does not significantly predict an international batting average, so a lot of the conventional things that are looked at as being indicators of success - they don't really stand true in a predictive sense."  And later in the article there is a graph, showing county averages plotted against test averages for 13 English test batsmen.  This is reproduced below.

better than random?

raw data suggests no meaningful link between championship and test averages

Test

County Championship

Sam Curran

England players' batting averages in Test and Championship cricket, 2016-19 (min. 20 Test innings) 

As you can see the line drawn through a best fit of the points is (pretty much) flat.  Showing that the average in county cricket  has a weak link to the average in test cricket, maybe.  I'm going to come back to the graph, but first wanted to think a little more about how county cricket might help to produce a good England test side.

I think county cricket can help produce test players in two ways.  Firstly it can develop skills that will be used in test match cricket.  Allowing the best county players to adjust to the demands of test matches.  There is a general consensus county cricket is not effective in this job. Although there is a division between those (like Mike Atherton) who believe county cricket has never developed test cricketers and others who would argue the recent expansion in short form cricket and the ECB's hostile environment for the first class counties has undermined a system that previously worked well.

But I'm going to avoid that debate and look at the second function of first class cricket, which is to act as a grading system for picking test teams.  That is can we look at the performances of first class cricketers and conclude the players with the best county averages will have the best chance of eventually converting to test match cricket? Even if we feel county cricket hasn't prepared them for the step up. But Mo Bobat, ECB Performance Director, and the chart are telling us that even on this criteria county cricket isn't much use.  This seemed odd to me and I started to have a closer look at the chart.....

If you look at the chart on cricinfo you can hover over the points and see the names of the players concerned.  I noticed someone had a county championship average of almost 60 and wondered who it was.  Checking, I found it was Moeen Ali, which was surprising.  So surprising that I went back to cricinfo and pulled up the stats for Moeen Ali; by taking his test match figures away from his first class figures you can, pretty much, get a county championship average (although there might be the odd tour game as well.)  And Mo has a first class average a smidgen over 40 not the 60 shown on the cricinfo graph.  Once I'd spotted the first mistake others started to jump off the page, Ben Stokes' average in first class cricket 25?  No, more like 35.  The chart doesn't work.

Quite why the numbers are all jumbled up is a bit of a secondary issue.  It might be because ESPN journalist, Matt Roller had a brain freeze, it happens.  The chart does refer to performances between 2016 and 2019 so perhaps only takes first class performances between those dates into account.  But if that is the case the chart is still meaningless, test players play very little county championship cricket once established in the test team  (I reckon Jos Buttler played 6 county games between 2016 and 2019) and so the small sample size means their averages will be all over the place.  Of course if you base the first class statistics on limited data you are going to get inconsistent results, but that just proves the data isn't useful rather than anything about the underlying relationship between county championship and test performances.

Using cricinfo's data it is possible to calculate a test and an "other first class" average for each of the 13 players shown on their chart but this time using all of the available data rather than whatever they were doing.  The result is shown below.

 

Once the data is improved a relationship is established between county and test performance, the line of best fit slopes upwards showing the players with the best county averages tend to have the best test averages and most players are close (ish) to the line of best fit.  I don't want to overplay this, I'm not saying county performance always leads to test success.  Mark Ramprakash and Graeme Hick are good examples of players who were dominant at county level but couldn't adjust to tests and, sadly, it looks as if Jonny Bairstow may be a a modern day example. Conversely Jos Buttler and Ben Stokes are both better test batsmen than their county record would suggest.  But there is nothing in the data to support Mo Bobat's analysis, " A county batting average does not significantly predict an international batting average."  

A quick look at the data suggests the relationship between county championship and test performance, holds for bowling as well as batting, with my chart below showing respective averages for England players from recent years. 

Again the line on the graph shows a clear relationship between county performance and test outcomes, although now the best performers are in the bottom left of the graph rather than top right (as you want your bowling average as low as possible.)

As with batting, bowling performances in the county championship aren't test match destiny.  Although some bowlers are on the line of best fit others are outliers.  Moeen Ali has a lower test bowling average than you would predict from his county average (interestingly his test batting average is lower than indicated by his county batting average).  Chris Woakes has a higher bowling average than predicted and Sam Curran has better test outcomes for batting and bowling than are indicated by his county championship statistics. 

But I think the evidence supports a conclusion that performances give you a general feel for how a player will go in test matches although there will be surprises.  This I think is the result you would expect (after all it would be a surprise if success at one level of the game didn't corelate with performance at a higher level.)  The surprise is that Mo Bobat and ESPN cricinfo were prepared to go to quite a lot of effort to come up with data which seemingly supported a rather unlikely hypothesis.

I should come clean and say I'm no statistician, but this is perhaps indicative of a wider problem with the selective use of statistics in cricket.  For instance we are sometimes told a player with a good overall record is no good because his average in away test matches, or test matches in India, or test matches on a Wednesday in February don't match his career average.  But so what?  The whole point of an average is that it averages out individual results which will fluctuate and the whole point of looking at data is to test our preconceptions, not buttress them.  Isn't it?