Conversion Points Added

This past weekend I posted a simple table of individual conversion stats gathered from World Rugby. It was a complete list along with each kicker’s conversion success rate. World Rugby keeps tabs on the leaders in total conversions, but we can’t tell who are actually the best kickers since the quantity and difficulty of the attempts are not equal.

Fortunately, I track the width of each conversion attempt. With this, I ran a logistic regression for success rate dependent on the conversion width. This regression allows us to plug in the width of a conversion to get the average success rate at that width. For example, a dead-center conversion is expected to be made around 97% of the time while a conversion from the touchline is only successful about 18% of the time.

In this fashion I calculated how many points each kicker’s profile of kicks is worth if averaged across all kickers. For example, let’s say a kicker had two attempts, one from dead-center and one from the touchline. Using the percentages from above, the average (across all kickers) would earn .97 x 2 = 1.94 points on the centered conversion and .18 x 2 = .36 points on the wide kick for a total of 2.3 points.

Again, this is on average. So some kickers do better than others. If Kicker A made both kicks to score 4 points, he actually scored 1.7 more points than the average. And if Kicker B missed both, his 0 points are 2.3 points less than the average. Using this method I tallied how many points above or below the average each kicker has added through all of his attempts.

Player Team Cons Attempts % Pts+ Pts+ per Att
Gaston Revol ARG 59 88 67.0 % 17.8 0.20
Scott Wight SCO 95 126 75.4 % 14.8 0.12
Ethan Davies WAL 82 113 72.6 % 12.5 0.11
Terry Bouhraoua FRA 43 57 75.4 % 11.3 0.20
James Stannard AUS 74 117 63.2 % 10.2 0.09
Rocky Khan NZL 40 53 75.5 % 9.6 0.18
Tom Mitchell ENG 79 114 69.3 % 9.5 0.08
Katsuyuki Sakai JPN 45 62 72.6 % 7.5 0.12
Nathan Hirayama CAN 71 105 67.6 % 6.4 0.06
Branco Du Preez RSA 67 108 62.0 % 3.7 0.03
Beaudein Waaka NZL 30 46 65.2 % 3.2 0.07
Isaac Te Tamaki NZL 9 12 75.0 % 3.1 0.26
Javier Rojas ARG 23 36 63.9 % 2.2 0.06
Joe Webber NZL 3 5 60.0 % 2.1 0.42
Tila Mealoi SAM 72 108 66.7 % 2.1 0.02
Osea Kolinisau FIJ 74 104 71.2 % 2.1 0.02
Folau Niua USA 4 10 40.0 % 2.1 0.21
Kevin Kermundu UGA 3 3 100.0 % 2.0 0.67
Vatemo Ravouvou FIJ 39 62 62.9 % 2.0 0.03
Tom Lucas AUS 16 21 76.2 % 1.8 0.09
Amenoni Nasilasila FIJ 11 15 73.3 % 1.7 0.11
Lloyd Evans WAL 1 2 50.0 % 1.6 0.80
Billy McBryde WAL 10 12 83.3 % 1.3 0.11
Rosko Specman RSA 2 2 100.0 % 1.1 0.55
Igor Genua ESP 4 5 80.0 % 1.1 0.22
Han Gumin KOR 2 2 100.0 % 1.0 0.50
Sam Caslick AUS 4 6 66.7 % 1.0 0.17
Alivereti Veitokani FIJ 6 10 60.0 % 1.0 0.10
Alexis Palisson FRA 0 1 0.0 % 1.0 1.00
Stephen Parez FRA 10 14 71.4 % 0.8 0.06
Leon Ellison JPN 12 16 75.0 % 0.7 0.04
Jonathan Ruru NZL 2 5 40.0 % 0.7 0.14
George Horne SCO 1 1 100.0 % 0.5 0.50
Luke Treharne WAL 5 8 62.5 % 0.5 0.06
Jean Pascal Barraque FRA 21 31 67.7 % 0.5 0.02
Domingo Miotti ARG 3 5 60.0 % 0.4 0.08
Madison Hughes USA 92 142 64.8 % 0.4 0.00
Selvyn Davids RSA 1 1 100.0 % 0.4 0.40
Vladimir Ostroushko RUS 1 2 50.0 % 0.4 0.20
Will Edwards ENG 5 5 100.0 % 0.3 0.06
German Klubus ARG 1 1 100.0 % 0.2 0.20
Pablo Fontes ESP 2 2 100.0 % 0.1 0.05
Julien Candelon FRA 2 2 100.0 % 0.1 0.05
Alatasi Tupou SAM 9 13 69.2 % 0.1 0.01
Mikhail Kudinov RUS 1 1 100.0 % 0.1 0.10
Emosi Mulevoro FIJ 1 1 100.0 % 0.1 0.10
Jamie Henry JPN 1 1 100.0 % 0.1 0.10
Ilya Babaev RUS 1 1 100.0 % 0.1 0.10
Tomasi Alosio SAM 1 1 100.0 % 0.1 0.10
Nick McLennan SCO 1 1 100.0 % 0.1 0.10
Laaloi Leilual SAM 1 1 100.0 % 0.1 0.10
Masahiro Nakano JPN 0 1 0.0 % 0.1 0.10
Patrick Stehlin JPN 1 1 100.0 % 0.1 0.10
Andrew Durutalo USA 1 1 100.0 % 0.1 0.10
Josua Vici FIJ 1 1 100.0 % 0.1 0.10
James Johnstone SCO 1 1 100.0 % 0.1 0.10
Dewald Human RSA 1 1 100.0 % 0.1 0.10
Stephen Tomasin USA 17 23 73.9 % 0.1 0.00
Emmanuel Guise PNG 2 3 66.7 % -0.1 -0.03
Joaquin Riera ARG 4 10 40.0 % -0.1 -0.01
Kosuke Hashino JPN 6 9 66.7 % -0.3 -0.03
John Porch AUS 2 3 66.7 % -0.3 -0.10
Justin Geduld RSA 27 45 60.0 % -0.3 -0.01
Gavin Lowe SCO 10 16 62.5 % -0.3 -0.02
Nicolas Menendez ARG 0 1 0.0 % -0.4 -0.40
Phil Burgess ENG 0 1 0.0 % -0.4 -0.40
Eugene Tokavai PNG 0 1 0.0 % -0.5 -0.50
Gregor Hunter SCO 2 3 66.7 % -0.5 -0.17
Harry Jones CAN 0 1 0.0 % -0.6 -0.60
Brandon Quinn AUS 0 1 0.0 % -0.6 -0.60
Blair Kinghorn SCO 1 2 50.0 % -0.7 -0.35
Yoshiaki Tsurugasaki JPN 0 1 0.0 % -0.7 -0.70
Scott Curry NZL 0 1 0.0 % -0.7 -0.70
Vincent Inigo FRA 1 2 50.0 % -0.7 -0.35
Pierre Popelin FRA 0 1 0.0 % -0.8 -0.80
Francisco Urroz CHL 0 1 0.0 % -0.8 -0.80
Cecil Afrika RSA 30 46 65.2 % -0.8 -0.02
Liam McNamara AUS 0 2 0.0 % -0.9 -0.45
Matias Ferro ARG 0 1 0.0 % -0.9 -0.90
Martin Iosefo USA 0 1 0.0 % -0.9 -0.90
Tate McDermott AUS 1 2 50.0 % -1.0 -0.50
Kitione Taliga FIJ 8 10 80.0 % -1.2 -0.12
Jamie Hood HKG 4 8 50.0 % -1.2 -0.15
Fernando Luna ARG 2 3 66.7 % -1.2 -0.40
Lee Jaebok KOR 0 1 0.0 % -1.3 -1.30
Morgan Williams WAL 1 3 33.3 % -1.8 -0.60
Mike Fuailefau CAN 3 5 60.0 % -1.9 -0.38
Tim Mikkelson NZL 1 4 25.0 % -2.0 -0.50
Yury Gostyuzhev RUS 1 4 25.0 % -2.1 -0.52
Jeremy Aicardi FRA 11 15 73.3 % -2.3 -0.15
Mike Teo USA 2 6 33.3 % -2.5 -0.42
Brian Tanga KEN 8 14 57.1 % -2.6 -0.19
Philip Wokorach UGA 11 19 57.9 % -2.6 -0.14
Ruhan Nel RSA 1 3 33.3 % -2.7 -0.90
Oliver Lindsay-Hague ENG 0 2 0.0 % -2.7 -1.35
Arthur Clement PNG 5 8 62.5 % -2.9 -0.36
Waisea Nacuqu FIJ 7 11 63.6 % -3.2 -0.29
Lautaro Bazan Velez ARG 1 5 20.0 % -3.2 -0.64
Roman Roshchin RUS 0 5 0.0 % -3.6 -0.72
Marcelo Torrealba CHL 6 12 50.0 % -3.9 -0.32
Jamie Booth NZL 0 3 0.0 % -3.9 -1.30
Eden Agero KEN 27 39 69.2 % -4.1 -0.11
German Davydov RUS 4 10 40.0 % -4.2 -0.42
Daniel Bibby ENG 25 41 61.0 % -5.0 -0.12
Dmitry Perov RUS 3 10 30.0 % -5.4 -0.54
Lugonzo Augustine Ligamy KEN 16 25 64.0 % -6.6 -0.26
Pat Kay CAN 18 30 60.0 % -8.4 -0.28
Vilimoni Koroi NZL 19 40 47.5 % -9.7 -0.24
Samuel Oliech KEN 33 53 62.3 % -14.2 -0.27
Dmitry Sukhin RUS 24 52 46.2 % -16.4 -0.32

So while Scott Wight has the best success rate, Gaston Revol has added the most points above the average for his profile of kicks. This is due to Revol’s more difficult set of kicks (average width of ~20m compared to Wight’s ~14m). If given Wight’s set of conversion attempts, Revol would likely have a higher success rate.

The last column shows our points added on a per attempt basis. This helps us compare kickers who may have significantly different quantities of kicks. Terry Bouhraoua has many fewer attempts due to injury, but his points added per attempt is as good as Revol’s.

It should be noted that both of these metrics are influenced by the profile of kicks. If a kicker was given 100 kicks from straight in front of the posts and made them all, they could only add .06 points per attempt for a total of 6 points added. But all of the regular kickers above avoid this sort of extreme kicking profile.

Application
While not perfect, I think this method does a better job comparing kickers than success rate alone. Also, quantifying each kicker’s ability in terms of points helps frame their influence on an individual match or tournament, where one extra touch-line conversion can greatly influence the outcome.

Further Visualizing Meters Gained

A reader (who claims to be my biggest fan) requested a box plot for the recent data on differences in meters gained. The reader wanted to more clearly see how teams vary in their performance.


Here we see the distribution of each team’s game to game meters gained difference ordered by the team’s average difference. (Since box plots use quartiles, the bar within the box is the median, not the mean.) A taller box or set of whiskers means the team has a larger variance across games. I think we’ll all be happy to see that France varies greatly. As does England while Japan looks to have the largest spread. Conversely, New Zealand, USA, Argentina, and Samoa have some of the narrowest distributions.

I’m not too surprised by NZ considering they typically utilize a methodical offense, a strong defense, and limit the number of possession opportunities in a game. I’m more surprised by the USA who can win kickoff after kickoff versus a poor team but also had their struggles in early tournaments.

This plot may also give us our first taste of team style. Japan appear to be erratic with high risk play while NZ and Samoa appear to use a more systematic approach. Though more investigation would be needed to truly label teams.

 

Meters Gained and Margin of Victory

In yesterday’s post I showed each team’s season long trends for meters gained and conceded by game. As an introduction to considering the application of meters gained, I broadly mentioned how gaining meters is good for a team. This makes sense and still holds true but to illustrate this statement I created the plot below.

The chart plots each game’s difference in final score as it relates to the game’s difference in meters gained. The relationship is clear, the more meters a team outgains their opponent by, the more points that team should win by. Outgaining your opponent doesn’t guarantee victory but the R² is .81 and nearly every team that gained more than 100 meters than their opponent has won the game.

Perhaps the concept isn’t incredibly enlightening to you. But seeing the data lets us know a few things. First, from a stat-taking perspective, it’s clear that tracking meters gained is worthwhile. Measuring meters gained from certain game situations could even help evaluate the success of set pieces, individual players, or tactical changes.

Second, the slope of the regression line, around .147, tells us what to expect from gaining meters. So for every meter a team outgains their opponent, you’d expect an additional .147 points on the scoreboard. For example, if you outgain your opponent by 100 meters, you’d expect to win by 14 to 15 points.

Knowing that expectation can guide decision making. If a team is outgaining their opponents but losing, they may be the victim of some bad luck. That bad luck could come in the form of turnovers near either tryline or some uncharacteristic missed conversions. A team could easily lose their two Sunday matches while outgaining their opponent, and without the above knowledge the coach may deem the tournament a failure and make knee-jerk changes. Whereas if they continue to play as they did, and continue to outgain their opponents, in the long run the team will be successful and win similar games in the future.

I should make it clear that “bad luck” events do matter and definitely affect the outcome of the game they occur in. But there’s a good chance these events are not as indicative of future results as other metrics, such as meters gained.

Plotting Meters Gained and Conceded

Motivation
I’ve been interested in visualizing the relationship between meters gained or conceded and overall team success. My data shows that the best teams gain the most meters and concede the fewest. This makes sense even without the numbers; teams need to move the ball towards the opposition try line to score.

I also was interested in seeing how teams were trending in meters gained and conceded. Do these trends help illustrate personnel or tactical changes within teams? I went ahead and plotted the figure below.

Explanation
There’s a lot going on so I’ll make sure things are clear. First, I’ve created the same plot for each core team with the same scales for easy comparison between teams. I eliminated non-core teams since their low number of games created noisy and truncated graphs. The dots display meters gained (blue) and meters conceded (green) for each game played this year. The lines show the teams’ trends across the games. Games are displayed in chronological order left to right; Dubai is on your left and the most recent games in Hong Kong are on the right. The colored vertical shading represents individual tournaments.

Analysis
Take Russia for example. Their meters conceded in defense (green dots and line) were very poor in Dubai. But they improved and from Sydney on, have mostly remained the same. On the other side of the ball, in attack, they were poor in Dubai as well. But they improved, culminating in a Challenge Trophy victory in Sydney, before dropping off again.

England and Fiji have been notably consistent in their success. Both have maintained what looks to be an average 50-75 meter per game advantage over their opponents. Meanwhile, the series-leading Blitzboks were incredible to start the season, but have seen a consistent increase in meters conceded and more recently, a decline in their meters gained. Their loss of advantage in meters gained and conceded correlates with some of their worst results of the season. In their Hong Kong semi-final they needed overtime to get by the USA before suffering their worst defeat of the year, 22-0 to Fiji in the final.

Australia is another interesting case. En route to bronze in Hong Kong, they managed victories over England, Argentina, and the USA all while being out-gained in meters. Considering the wide variation in meters gained and conceded from game to game, their wins could be the result of pure luck, rather than some team tactic that involves scoring without gaining meters. I would not have seen this without the chart above and it’s something to keep an eye on in Singapore.

Summary
I think this chart has promise and I’ll likely reproduce it after future tournaments. It helps illustrate overall team success and failure through the series and provides a quick comparison between teams. Trends can be noticed quicker than with tabular data and this same display with different metrics could provide similar insight.

P.S. – Technical Mechanics
I do nearly everything in Python and the above chart was created using FacetGrid and RegPlot from the Seaborn library. My first attempt was with Seaborn’s lmplot (which is a combination of FacetGrid and RegPlot) but I was having trouble creating the colored vertical bands. Not saying it’s not possible to do it with lmplot but I solved it by building the FacetGrid myself.