Australia's Relative Decline; The Distribution and Production of Talent
Russell Degnan

When I started gathering these statistics the aim was to put together a post on the difficulties of associate cricket. That will now be part three. Australia's recent, somewhat lamentable, performances have allowed me to retool and add to them to create a broader picture of how talent appears from a playing base, how it improves and declines with age, and the challenges in creating a competitive environment for much smaller teams.

Australia's challenges against England and before that India and South Africa need to be kept in perspective. They are still rated as an above average test team - let alone amongst the ICC's 106 cricket nations; they are the most successful, the equal oldest test team, the second or third richest in annual income, the third largest test team in gross domestic product - one of the best indicators of sporting excellence. The total playing base is 800,000 plus, of which around 500,000 play in club competitions, equal with England and South Africa.

There are good reasons to think Australia aren't adequately developing their talent - on which I will try and show in part two - but these are persistent problems, and problems not unique to Australia: talent must be spotted from within a large pool and this is very difficult; it must be kept in the game while it develops, which in the modern world means paying under-performing youth; and selecting the right players based on a handful of data points means talent and temperament spotting is often as important as analysis and results. The rejoinder that these weren't problems ten years ago is false; they were, but when you have Warne, McGrath, Ponting and Gilchrist, it doesn't matter if selectors choose a batsman with a career average of 35 over one of 45 for several years. The team is starting 100 runs ahead of average. That is not the case when a team is at its normal level.

Likewise, the current Australian team is not that dissimilar to the one that played in 1989. That team had recently, and continued to be until 1993, dismantled by the West Indies. It was youthful, and where it wasn't - Border aside - it wasn't performing terribly well. But it came up against an English side that only beat Sri Lanka and a terrible Australia over a 4 year period. A team most of the Australian batsmen of the era averaged 5-10 runs more than their career average against. That's the luck of the draw, but it is worth remembering.

Australia can't be great all the time. They draw on the far-end of the talent curve, (hopefully) picking the best six batsmen available, and then (hopefully) they perform near to or above their talent. Over the course of test history Australia's population has increased, as have several comparable nations. For the purposes of this exercise I'll compare New Zealand, a nation with roughly a fifth the number of cricketers as Australia, and therefore (at least in theory) they will have a fifth the number of batsmen at each elite level - meaning they'll select a lot of weaker batsmen.

In order to create blocks with sufficient number of data points, I've looked at performance by batting position, not individuals, split into three year blocks. I'll also be looking at the reciprocal of the average - more easily understood as the probability a batsman makes another run before being dismissed - because it better fits a normal curve. Below is the graph of Australia's talent distribution, for batting (bowling is much harder to analyse by position) dating back to the 1920s - when overall averages leveled out to approximately the current level.

The spread was much larger before the war, mostly because of fewer games being played, but the average is broadly consistent: 39.5

The current period is unusual in having two positions in the bottom 5% of historical performances and one in the top (number five, needless to say) but a t-test isn't significant at the 5% level. In fact only four lineups are: the invincibles of 1947-1949 (well above), the mostly all-rounders Laker-bait of 1956-58 (well below), the WSC years of 1977 (below) and 2001-03 (above). Expecting the present lineup to be anywhere near the last of those is crazy, and matches previously weak teams (the mid-80s) or weaker sides playing strong English sides (the 1950s).

New Zealand exhibits a similar graph, but it is shifted to the right. The mean batting average by position for New Zealand in test history is 28.0. They have had very good batsmen, but they draw from a smaller pool. This can best be demonstrated with this graph, also showing England:

This graph shows the historical distribution of three-year position averages for several teams, and some modelled distribution probabilities for different sized playing bases. Needless to say, the probability of getting Bradman is miniscule, but it is mostly interesting for what it says about the effect of playing base size.

Two things stand out on this graph. Firstly, modelling talent is hard, because variation in performance is huge. In theory the worst players in the side should be a mass of replacable pieces - that's not to say you should replace them, merely that most replacements are as good as each other - but players can fail to meet their true talent for long periods - Steve Waugh, after 52 tests, the same as Bradman and Ramprakash, averaged only 34. Hence the long swinging tail, accentuated by the log scale.

Secondly, the first modeled dotted grey lines is fitted to Australia's talent curve. The others, from 100,000 players (roughly New Zealand) down to 5000 (roughly the Netherlands), are calculated from Australia's line. Each average on the 100,000 line has one fifth the probability of occurring as on the 500,000 line. Australia and England have similarly sized playing bases, and both sit quite close to where they ought to be. More interestingly (and the fit is rather remarkable), New Zealand's historical distribution line meets almost exactly where the model predicts.

Any team can luck into having great players, but the bigger the playing base, the higher the probability. Every team is constantly rolling the dice and hoping to draw a collection of high performers - and even small playing bases can produce excellent teams, if intermittently. A nation can also fail to use its vast population - as India has failed to do until very recently, and which may fundamentally change international cricket, if they succeed in doing so. Or not have the money or opportunities to develop what talent they have.

But they are always slaves to their playing base. In the long term, that's the mean they regress to. Unless someone can show something has fundamentally changed in the production of Australia's cricketers, that breaks a hundred years of landing on or near that line pretty consistently, it is safe to assume Australia will find decent batsmen again. Given the comparative youth of this side it may have already, and we must merely be patient.

In the long term Australia and England are evenly matched. In the short-term, development and selection matter in order to get the (probable) best players in the team for each match. There are good reasons to think Australia does this badly which I'll explore in part two.

Idle Summers 28th July, 2013 02:59:15   [#] 

Comments

Troy
I've been pondering this as well, but hadn't got as far as formulating a methodology. But your results look to capture something I suspected; Australia are probably not historically woeful, but the current team is following a team that was historically good. Although not that runs are not coming as easily for #5, maybe they are slipping towards lower territory?
Troy Wheatley  29th July, 2013 16:24:37  

Historical comparisons
I think, even leaving aside the runs from five, they are drifting towards historically bad. But they've also played, in 2013, difficult conditions, against good sides. That's the difficulty with comparative analysis. it is also the down-side of CA organising so many series against India, England and South Africa. Losses are much more likely.

We'll see I guess. The administrative side can do better, which I'll get to next. Although it is much harder to prove as such.
Russ  29th July, 2013 18:38:14