The month of March brings with it the return of Australia’s two most popular sporting competitions: the NRL and the AFL.
Both competitions aim to maintain a competitive balance to avoid becoming boring or predictable. This is achieved via schemes such as salary caps and revenue sharing. But the success of such initiatives varies between sports.
In a theoretical league where every team is evenly matched, success in a given season would purely be down to random chance. This would not provide any sort of guide to their chances in a future season.
This is an extreme and unrealistic situation. Try convincing Australian sports fans that Johnathan Thurston’s annus mirabilis in the NRL or Hawthorn’s sheer dominance in the AFL were purely down to a cosmological coin flip. Some teams are consistently better than others, regardless of salary caps. This will be reflected in the league ladders.
It is still true, though, that the more balanced a competition is, the less predictable the ladder will be from season to season. A good measure of competitive balance in sport is the extent to which a team’s performance in one season can be used to predict their performance the following season. So what predictions can we make about the coming football seasons?
We compared four prominent sports leagues: Australian rules’ AFL, rugby league’s NRL, basketball’s NBL and football’s A-League. The analysis covers the last decade – seasons 2006 to 2015 for the NRL and AFL, and 2005-06 to 2014-15 for the NBL and A-League.
We considered only performances in the regular season, which represent a more robust measure of teams’ comparative abilities. Playoffs or finals are designed to introduce a further element of chance, which means the best team does not always come out on top. Points deductions for issues such as salary cap breaches were ignored.
In some sports, the number of matches played per season, and thus the number of points available, has changed across the study period. To account for this, we compared each team in terms of the proportion of available points earned.
Fremantle picked up 68 points out of a possible 88 from the club’s 22 AFL matches in 2015. That means the Dockers earned 77% of the available points, giving them a score of 0.77.
Fremantle’s score in 2014 was 0.73. These scores are very similar, which suggests that Fremantle’s 2015 performance was quite predictable based on 2014.
However, it would be wrong to draw a conclusion based on just one observation. We need to compare Fremantle in each other pair of years (2013-2014, 2012-2013 and so on), and to do the same for each other club.
By doing so, we obtain a total of 151 pairs of scores for the AFL, 143 for the NRL, 81 for the NBL and 80 for the A-League. We can use this data to compute the statistical correlation between scores in consecutive years.
Correlation can be thought of as the level of dependence between two objects. If two variables are correlated, then a change in the value of one will correspond to a change of value in the other.
Pearson product-moment correlation coefficients were computed to compare the season pairs for each league. The larger this score is, the greater the positive correlation between the scores and the more “predictable” you can consider the league to be.
A score of one implies complete correlation. This would mean that you could exactly predict each team’s score just by knowing their score from the previous season. A value of zero implies no correlation, meaning the scores from one year to the next are completely random.
We also obtained a 95% confidence interval, which accounts for the uncertainty in our calculation. This interval outlines a range within which we can reasonably expect the true correlation to lie.
What we found
For the AFL, we obtained a correlation coefficient of 0.57, with a confidence interval of 0.45, 0.67. This means it is highly likely that the season-to-season correlation for the AFL lies somewhere in the range 0.45-0.67, with our best estimate being 0.57. This estimate implies a moderate to strong correlation between scores in consecutive seasons.
The NRL had a coefficient of 0.25, with an interval of 0.09, 0.39. The NBL had a coefficient of 0.47, with an interval of 0.28, 0.62. The A-League had a coefficient of 0.07, with an interval of -0.15, 0.29.
It appears the AFL has been the most predictable of the four leagues over the last decade. A glance at recent ladders suggests this may be true – Hawthorn, Sydney and Fremantle have performed consistently well in recent years.
The NBL also has a fairly high correlation. This may reflect the domination that the Perth Wildcats and New Zealand Breakers have had over the competition in the past decade.
The correlation is much lower for the NRL. This suggests a greater competitive balance within the league. Although the Sydney Roosters have been minor premiers for the last three seasons, the form of many other sides has fluctuated wildly. The Parramatta Eels and Gold Coast Titans are the only clubs that have not reached the playoffs in the last five years.
The most unpredictable of the four leagues appears to be the A-League. The confidence interval suggests that it is plausible that there is no relationship at all between points scores in consecutive seasons. The league has had five different premiers in the last six years; the title race tends to go to the wire each year.
Lessons to be learnt
All four leagues take similar measures to try to maintain competitive balance, but the success of these measures varies across sports. The AFL is the most predictable, while the A-League is highly unpredictable.
It may be that tradition plays a part in this difference. The A-League is still in its infancy and a “natural order” has not yet been established in the league.
The AFL may be more predictable than the NRL or the A-League, but that doesn’t mean that it is in any way easy to predict. Sport is by its very nature unpredictable, and that’s why we love it.