Predictive Analytics Applied To Rugby League: Looking At Try Scorers In The NRL

Predictive Analytics Applied To Rugby League: Looking At Try Scorers In The NRL

Over the last few months, we’ve built statistical models with the aim of predicting the likelihood of team & players actions in the NRL. One key area we’ve focused on is predicting try scorers. Looking at which players will score a try at any time in a match and who will score the first try. 

Now, if you’re a rugby league fan you will be used to the ads on TV throwing stats at you along the lines of ‘Player X has scored 4 tries in his last 5 games against this opposition’. Never mind two of those games were at neutral venues, or the opposition team has had a roster overhaul or they’ve changed coaches twice and a host of other factors. As fans, we’ve got a low baseline for the use of numbers in rugby league and more importantly, there’s just nothing better out there so we’re forced to put up with it. 

We thought of taking a closer look at this. Using publicly available data and using statistical modelling techniques, we have come up with results that consistently outline the inefficiencies in the market. We’ve also got interactive visualisations for you to compare your favourite players as well as betting & fantasy resources – feel free to check it out and follow us on social media for more of our content. 

Anytime Try Scorer Model

Firstly we calculate a form of weighted average called Exponential Smoothing for each player for the following stats: Tries, Line Breaks and Tackle Breaks. This allows us to track each player’s “form” throughout their careers.

For example, if we look at a player’s Exponentially Smoothed tries, we should see that when a player is in better try-scoring form, he should be more likely to score a try in the next game (shown in the plot below). Note in the plot below that the y-axis is a probability, so for example if you have an Exponentially Smoothed tries value of 1 per game, you are roughly a 65% chance of scoring a try in the next game.

Secondly, we find an optimal weighting between a player’s try scoring probability and the probability of scoring a try in that particular position, while also accounting for the number of games that the player has played. For example, the average Winger scores a try in roughly 40% of games and Player X scores a try in roughly 60% of games. How do we weigh between the two? And how much more would we weigh towards the player if he had played 100 games compared to if he had played 10 games?

Hypothetically if the optimal weighting for this scenario was 25%, then we would apply a 25% weighting to Player X’s try-scoring probability (60% for the player) and a 75% weighting to the position try-scoring probability (40% for a Winger), which would come out as (25% * 60%) + (75% * 40%) = 45% chance. We call this value the “Bayes Anytime Probability”.

Thirdly, we apply a logistic regression model to predict the probability of a player scoring a try, factoring for the following variables:

  • Position for that game
  • Minutes Played in that game
  • The player’s Exponentially Smoothed tries
  • The player’s Exponentially Smoothed line breaks
  • The player’s Exponentially Smoothed tackle breaks
  • The Bayes Anytime Probability from above
  • Total Points scored in the match
  • Team’s Margin in the match

Surprisingly, second rowers and halves have scored the first try of games at almost the same rate since 2015. Nowadays, fullbacks are some of the highest profile players in the game. However, the data shows their rates of scoring the first try has been the same as centres.

First Try Scorer Model

Our model for predicting the likelihood of a player scoring the first try of the match is relatively similar. Firstly we calculate a form of weighted average called Exponential Smoothing for each player for the following stats: Tries, Line Breaks and Tackle Made. This allows us to track each player’s “form” throughout their careers. For example, if we look at a player’s Exponentially Smoothed tries, we should see that when a player is in better try-scoring form, he should be more likely to score a try in the next game and therefore more of a chance to score the first try (shown in the plot below). 

We then find an optimal weighting between a player’s first try scoring probability, the probability of scoring the first try of the match in that particular position, while also accounting for the number of games that the player has played. And finally, we apply a logistic regression model to predict the probability of a player scoring the first try, factoring for the same variables as for predicting a player to score at any time in the match.

Result Since 2015

Tryscoring records from Round 1 2015-Round 17 2020, grouped by position

As most rugby league fans would assume, wingers are by far the most likely players to score the first try in a match based off historical records from 2015. Surprisingly, second rowers and halves have scored the first try of games at almost the same rate since 2015. Nowadays, fullbacks are some of the highest profile players in the game. However, the data shows their rates of scoring the first try has been the same as centres. These are obviously only to be used as a guide, we’re not saying Dean Whare is just as likely to score the first try of a given match as James Tedesco!

Conclusion

Hopefully we’ve shown you the difference between the information that’s out there, what we’re looking to do with our predictive models and some of the historical try scoring records that help clear up common misconceptions on an under analysed area of the sport. To finish off, we’ve attached a screen shot from ‘Edge‘ which shows the discrepancies between the odds bookmakers have available, odds based off a player’s historical tryscoring records & our predictive model’s projection for what their odds should be.

First Tryscorer table from our data hub, ‘Edge’.
Share on facebook
Share on twitter
Share on whatsapp
Share on email
Share on linkedin

Contact us

Book A Free Demo