In round 4 of the 2020 Tri-Nations tournament the Argentina Pumas came from behind to draw with the Australia Wallabies 15-15. In the previous blog post we looked at team performance using some advanced analytics. For this post we will look at player evaluation. You may have already seen hints of this type of analysis on Tabai Matson’s segment on The Breakdown from Sky Sport NZ. Since he only had ~30 seconds to explain, it’s likely that most viewers probably didn’t understand the inner workings of the player ratings. This post will go into further detail around how the ratings work.
Here are some traditional individual statistics that were used to summarize the players for both teams (Tables 1A/1B):
TABLE 1A: Individual game statistics for the Wallabies from the Round 4 fixture between Australia and Argentina in the 2020 Tri-Nations tournament (Source: Opta).
TABLE 1B: Individual game statistics for the Pumas from the Round 4 fixture between Australia and Argentina in the 2020 Tri-Nations tournament (Source: Opta).
These tables summarize what are typically referred to as counting/descriptive statistics. These are characterized by identifying specific game actions (e.g. tackles, ball carries, penalties conceded, etc.) and counting them. By the end of the match you should have a total for each player across all categories measured. There has been a recent trend to start counting more detailed aspects of the game (e.g. positive tackles, OOA, LQB, BIG, etc.) but does more necessarily mean better? The challenge that you will have when evaluating players “statistically” this way is that it is difficult to ascertain who had the best game overall. Counting statistics by nature assume that all measures have equal values of +1, but does a ball carry necessarily equal a turnover won via jackal?
The limitations of counting statistics were discussed in previous Rugby 101 posts, while the concepts and associated advantages of Expected Points (EP) and Expected Points Added (EPA) were introduced (Article 1, Article 2). Here are the same metrics summarized in Tables 1A/1B but converted to EPAs.
TABLE 2A: Individual game statistics converted to EPAs for the Wallabies from the Round 4 fixture between Australia and Argentina in the 2020 Tri-Nations tournament.
TABLE 2A: Individual game statistics converted to EPAs for the Pumas from the Round 4 fixture between Australia and Argentina in the 2020 Tri-Nations tournament.
Conceptually, here are what each of the EPAs mean and how they were determined:
Carries: Change in expected points resulting from a gain/loss in territory from the start of a carry to the end of the carry
Pass/Offload: Change in expected points resulting from a gain/loss in territory of a teammates’ ball carry who received a pass or offload from a player. A value of 0 means a pass/offload was completed but no territory was gained.
Kick: Change in expected points resulting from a gain/loss in territory from the start of a kick to the end of the kick. This did not take into account the resulting kick return if the ball was kicked in play.
Tackles: Change in expected points resulting from the absence of a missed tackle weighted to the number of tackle attempts
Missed Tackles: Change in expected points resulting from a missed tackle weighted to the number of tackle attempts. This is based on the expected points of a missed tackle based on where it happens on the rugby pitch. Because this impacts a team adversely, the value is negative.
Jackals: The expected points gained from the turnover won from a successful jackal.
Penalties Conceded: The expected points lost from a penalty conceded. Because this impacts a team adversely, the value is negative.
Turnovers Conceded: The expected points lost from a turnover conceded. Because this impacts a team adversely, the value is negative.
One advantage of using an expected points model to evaluate players is that all game actions can be converted to the same units (EPAs). Another advantage of using EP/EPA is that the values for all measures are based on actual game-derived values rather than arbitrarily selected values like those found in fantasy rugby. (Disclaimer: This is not to disparage fantasy rugby leagues as they are a ton of fun, but ultimately serve a different purpose). In addition, expected point curves are unique to each competition so the EPAs for players will reflect the relative weightings of game actions specific to each competition (e.g. the value of a missed tackle on the 40m line will differ from Super Rugby to Gallagher Premiership). Since all measures of interest have the same units for all players, we can add everything up to a player total and compare them. In Tables 2A/2B each players’ numbers should sum across to the total denoted as “epa_total”. Please note that since all values in this column were rounded to 2 decimal places, the arithmetic may not be perfect in the tables provided. Since players play for different amounts of times, we can normalize their raw epa_totals to an 80 minute game. These values are found in the “epa_80” column but did not include the “finishers” (a.k.a bench players) since normalizing totals for players with less than 40 minutes to a full 80 minutes is generally less accurate and prone to distortions.
This model highlights the importance of 9s/10s due to the fact that they have such a big influence on territory gained via kicks, carries, and passes – they both dominate the epa_80 scores for both teams. For this reason it might be more appropriate to isolate and compare halfbacks to each other in a separate analysis rather than compare to the rest of their teammates. In addition, the current iteration of this model under-reports the contribution of tight 5 players as metrics related to setpiece won/lost have yet to be incorporated into the model. Obviously as this is an evolving process, we hope to incorporate the setpiece contributions of tight 5 players into future iterations of this model. We also hope to incorporate both attacking and breakdown metrics into our player rankings in the near future. Stay tuned!!
How did your favourite player compare in these rankings?