A Closer Look at AFL Stats : Conversion Rates
Exploratory data analysis on conversion rates in AFL matches
Following on from my last analysis on clustering of team stats, it becomes more apparent that a model for match outcomes needs to consider both the generation of scoring opportunities in the midfield and the execution of scoring opportunities in the goal square.
In this article I explore the latter in greater depth — conversion rates — the final link in the possession to scoring chain. Specifically, I am looking to understand factors which affect conversion rates, whether it is just an “interesting” ratio and/or how can be useful in predicting upcoming matches.
Historical Context
Conversion rates have improved significantly since AFL was first played in the 1890s from around 40% to 52% currently. This path has not been completely linear; history shows the changes in conversion rates has been due to skill improvements in (1) the midfield via the number of scoring opportunities and (2) the goal square as the ratio of goals to behinds has changed. There have been four distinct periods -
- In the first 40 years to 1935, conversion rates increased as the number of actual goals increased.
- Between 1935 and 1960 conversion rates tapered off slowly as did total scoring shots and goals.
- From 1960 to 2000 the increase in conversion rates was again due to an increasing number of goals kicked; during this period the actual number of goals kicked exceeded behinds.
- Since 2000 there has been no significant increase in conversion rates while total scoring shots has fallen by 25%..
It is difficult to explain these changes without the context of history of game rules and evolution of team styles and performance, however it does appear that game play has evolved towards more accurate goal kicking over time.
In the current era, average conversion rates are around 51.4% on 22.3 scoring shots per match — that is, teams have scored an average of 11.6 goals per match. For the purposes of this article goal accuracy or conversion rates is defined as Conversion Rate = Goals / (Goals + Behinds).
- An issue underscoring the analysis on a match by match basis is due to the actual values being both small and discrete —the difference between converting one more behind to a goal translates to a 4.4% increase in the conversion rates. To get around this issue, and taking into account the variance of each of the underlying components, we look at conversion rates by season rather than by match in most cases.
- As scoring shots form the denominator of the calculation, it becomes more apparent in the analysis that we can’t look at conversion rates in isolation.
The remainder of this article focuses on conversion rates in the current era from 2000–2019 and excludes the finals series.
How consistent are conversion rates at home vs away games?
There are no significant differences in conversion rates between home and away games — in aggregate similar profiles can be observed for the distribution and cumulative frequency for conversion rates.
How consistent are team conversion rates?
Team conversion rates for a whole season smooths out the week to week variances from changing oppositions. Conversion rates do not appear to be a fully random process from one season to the next however some teams (1) do exhibit more consistency than others in the long run and (2) do have straighter goal kickers than others.
- Consistency of conversion rates — (1) above average — Geelong and Hawthorn (2) below average — St Kilda (3) completely random — Port Adelaide and Collingwood.
- Trend of conversion rates — (1) improving — West Coast (2) declining — St Kilda.
While Geelong and Hawthorn have higher conversion rates and have ended the seasons higher up on the ladder in the last 20 years, Richmond has had conversion rates over the 2017–2019 in line with season averages.
Are conversion rates affected by pressure?
As I watched the 2019 Grand Final between Richmond and GWS it was apparent to me that pressure is a factor in goal accuracy with final tally being 17.12 vs 3.7. Team conversion rates are lower than average for losing matches whether pressure comes from scoreboard tally or opposition tactics or a combination of both.
In general, the difference between conversion rates is around 7% between winning and losing matches for most teams.
Can conversion rates predict win percentages for the season?
There is a positive correlation between conversion rates and winning although this relationship is, at best, scatty. Honestly it looks like a dog’s breakfast. Teams with high conversion rates than average can still lose the season. How is this possible?
A closer investigation of the individual seasons reveals that teams that have a greater number of wins also have a higher number of scoring shots during the season. The size of the data points in the chart below are proportional to total scoring shots for the season.
What is apparent is that teams that have more scoring shots have a higher winning percentage — because they are creating more opportunities. When we rejig Figure 6 above such that each data point size is proportional to the conversion rates we see a stronger linear relationship emerge.
This is an interesting diversion. The analysis shows that teams are more likely to win when they generate more scoring shots than if they have a higher conversion rate.
Hence, winning percentages for teams can be modelled as a function of both conversion rates and scoring shots, however the latter carries a higher weight because aggregate scoring shots are a proxy for team rankings on the ladder.
Reflections
The exploratory data analysis points to the following observations for the 2000–2019 seasons :
- Conversion rates are similar for home vs away teams.
- Conversion rates for individual teams vary in terms of predictability.
- Conversion rates are higher when teams are winning matches.
- Scoring shots — the number of execution opportunities — matters more to a team’s performance over the season than conversion rates. Both components should be included in a predictive model.
Directions for future research :
- Exploratory data analysis of factors affecting scoring shot production.
- A baseline model which uses cumulative scoring shots and conversion rates to predict next round match winners.
References
(1) Data scraped from afltables (link)
