South Africa 3, New Zealand 1

Watching this series as a fan rather than from inside a program has given me the time to enjoy the game for what it is, and to poke at the numbers behind it. Four tests between the All Blacks and the Springboks, arguably the two best sides in the world, is about as good a place to start as you get.

If you’re unaware, South Africa won the series three tests to one. New Zealand took the first at Ellis Park by 17, then South Africa won the next three. I went back through World Rugby’s match data for all four, along with every test between top sides since 2023, to work out where each team was better and how the games were won.

If you only read one paragraph, it’s this. South Africa won the collision up front in every test. New Zealand were better at beating a man in every test. Different teams, different coaching, different skill sets. At Ellis Park New Zealand’s work at the breakdown cancelled out South Africa’s edge up front and their finishing decided the outcome, and after that it was goal kicking and sin-bins.

How did the four tests go?

Ellis Park was the odd one out. South Africa were 16–12 up after 47 minutes and didn’t score again, and near the end they had two players in the bin at the same time. 13 v 15 is hard to come back from. The next three all went South Africa’s way. I’ve been having some fun visualising this, so below is how the scoreline moved through each match.

Where did South Africa win it?

As mentioned above, up front, at the gain line. The gain line is the imaginary line across the pitch where the last breakdown was. When a player carries, he either gets over it and his team is going forward, or he gets stopped behind it and they’re going backwards. South Africa’s front row and locks got over it with 61% of their carries across the series. New Zealand’s got over with 41%, and a normal top side sits around 49%.

To put that in a game: at Ellis Park, New Zealand’s tight five carried 30 times. Eight of those got over the gain line and 15 were stopped behind it, with the rest getting back to roughly where they started. South Africa carried 24 times, 15 over and four stopped. New Zealand still won that test, so the gain line isn’t the whole story, but over four tests it was the most consistent thing South Africa had.

This isn’t a knock on the All Blacks. A lot of it is South Africa’s defence. Tight fives only get over the gain line about 45% of the time against the Boks, second-hardest to carry into after England, and New Zealand’s 44%, 46% and 42% in the last three tests are about what anyone gets against them. Ellis Park, at 27%, was the tough day, and they won it anyway. In July the same New Zealand tight five got over the line 65% of the time against France and 64% against Ireland. It works both ways: New Zealand’s defence has been one of the easier ones to carry into for a couple of seasons, so this was as much about the defences as the packs.

It wasn’t only the tight five. Set every position against what that position normally manages between top sides and South Africa matched or beat it everywhere except the back three, where they were level with it. New Zealand came up short in four places. Their front row got over with 38% where 48% is normal, the locks 44% against 50%, the centres 48% against 60% and the back three 59% against 71%. Their back row was the exception, above par at 66%. So the pack that got talked about after the series was really the tight five, and the same problem was showing up out wide.

South Africa were bigger, too. Their pack was three to six kilos a man heavier than New Zealand’s in every test, and the five forwards on their bench were heavier again, heavier than their own starting pack in all four tests. New Zealand’s bench forwards were lighter than their starters in three of the four. I wouldn’t read that as the reason they won, though. Across top-side tests I can’t separate pack weight from how good the team already is, because a heavy pack is part of what makes South Africa good in the first place.

How much was each part of the game worth?

To put everything on one scale, using World Rugby’s API, I built a model from 226 recent tests, 110 of them between top sides. It works out how much each part of the game usually moves the scoreboard, then adds up what each side’s edge in each area was worth in these four tests. Think of it like pricing a house: a garage adds a bit, a busy road takes a bit off, and you add it all up to get what it should sell for. Here the features are parts of the game and the price is points. I’ve always been interested in what actually wins a match and what to spend the week on, and this is the same sort of weighing up as the hamstring risk factors piece, just with points instead of injuries. These four tests weren’t used to build it.

One thing to be straight about before the numbers. The model only sees how a side played. It never sees a line break, a try or a piece of silky skill, so it answers where each team was better, not who won. The two rows under each test show how that goes: it picked the right winner all four times and called FNB almost exactly, but it was out by eight to fifteen points in the other three. The same stat sheet can end in very different scores depending on who takes their chances, and taking your chances is what the chart leaves out. So there’s still work to do on this.

Contact went South Africa’s way in all four tests, about 34 points’ worth over the series. Set piece was worth six to ten points a test until Baltimore, where it hardly counted. Ellis Park is the one where the breakdown went New Zealand’s way, by enough to cancel out South Africa’s edges up front, so on the numbers that game was about level. Finishing is the name for the rest of it: five tries to one, including one while South Africa were down to 13, and none of that is in the bars.

One check worth knowing about. Blowouts can flatter a stat, the same way a side’s defence looks brilliant when they’re 30 points up and the other team has stopped trying. So I rebuilt the model using only games that were within seven points with 20 minutes left. Every number gets smaller, but the picture holds: contact (+8) and set piece (+8) are still South Africa’s biggest edges, and the breakdown is still the biggest thing in New Zealand’s favour at Ellis Park.

Where were New Zealand better?

Beating a man. New Zealand beat more defenders per carry than South Africa in every test, so while South Africa were winning the collisions, New Zealand were better at avoiding them. Will Jordan beat 20 defenders over the series and Cam Roigard 18, and nobody in green got past 12.

Digging into the individual players, size doesn’t explain it. Across every top-side player since 2024, heavier players win more of their collisions, but they don’t beat more defenders or make more metres. Age might explain some of it. Players beat about 15% fewer defenders per carry for every five years of age, and that’s measured against others in their own position, so it isn’t forwards lasting longer than backs. South Africa’s ball carriers were on average three years older, 31 to 28, and the gap holds inside both units: their backs were 30 to New Zealand’s 28, their forwards 32 to 28. I wouldn’t put a number on how much of the gap that covers, though. It’s a snapshot of lots of different players rather than the same players followed over the years, and the older ones still getting picked are the good ones, so it tells you what older and younger players tend to look like, not that anyone is slowing down.

Kicking from hand was the other area New Zealand did well. Their kicks went further in three of the four tests, and distance is worth points. South Africa got more of their own kicks back though, and over the series those two things roughly cancelled out.

Which stats aren’t what they look like?

A few numbers that get quoted after every test don’t tell you what you’d think. The side with less of the ball won three of the four. That fits what the research finds at this level. Bunker and Spencer looked at the 2019 World Cup and found that dominating possession, carrying more and winning more rucks separated winners from losers in the group stage, but none of it held up in the play-offs. Carrying over the gain line and total metres gained mattered at both. And South Africa missed more tackles than New Zealand in three tests, which sounds bad until you realise it’s the same number as New Zealand beating defenders. It’s like a bowler’s wicket and a batter’s dismissal in cricket: one moment, written down in two columns. In 98% of top-side games since 2023 the two counts match exactly. South Africa still won two of those three.

How much a side kicks is another one, and this is kicks from hand, in play, not shots at goal. The side that kicks more wins most tests, and I’d have told you it was one of the big levers. But the kicking follows the score as much as the score follows the kicking. Every time you score, the other side kicks off to you, and more often than not you put it straight back down the field, so the team that’s ahead ends up kicking more. Once I took those restarts back out of each team’s count and compared it with its own normal games, how much a side kicked barely moved the result. What does matter is how far the kicks go and, to a lesser extent, whether you get them back.

Ruck speed is similar. Quick ball mostly goes up when you’ve got the ball and you’re chasing the game, so on its own it doesn’t tell you much. What costs you is losing the ruck altogether, about a point each time once you allow for the score at half-time.

What decided the tight ones?

Ellis Park was the breakdown. New Zealand won 14 turnovers that day. No side has taken more off South Africa in 28 tests against top sides since 2023, and Jordie Barrett won five of them on his own. South Africa weren’t being driven backwards either. Their rucks behind the gain line were among the lowest they’ve had. They were having the ball taken off them going forward. That levelled the game, and then New Zealand scored five tries to one.

After that it was small things. New Zealand kicked 13 from 18 off the tee over the series and South Africa 20 from 23. In the two tests New Zealand lost by seven and by five, they missed nine points, seven of them at FNB. South Africa missed some too, five of their seven in Baltimore. And the sin-bins hurt. A yellow card between top sides costs about four points on average. The four full ones here cost seven each, though the two at Ellis Park overlapped and shared the same try, and South Africa copped three of the four. New Zealand’s one yellow at FNB came with them 12–5 up, and a penalty try had South Africa level a minute later.

What’s changed under Dave Rennie?

Rennie’s All Blacks look different in a couple of clear ways. He’s reshaped the backline, with Damian McKenzie at 15, Will Jordan on the wing, Quinn Tupaea at 13 and Ruben Love at 10, and he’s brought back experienced forwards like Luke Jacobson and Asafo Aumua. The biggest change on the field is who does the kicking. Under Scott Robertson the 10 took more than a third of New Zealand’s kicks. Under Rennie it’s under a fifth, with Cam Roigard and the centres doing more of it. That fits the 10 he picked. When Love was named for Rennie’s first test, the AAP report described a player whose instinct is to take the defensive line on, where Barrett tends to kick or pass before it.

Love had a hard time off the tee at FNB, and he missed more tackles than you’d want from a 10, but he did the job he was picked for with the ball. He beat eight defenders in his three tests, getting on for half as many again as a typical top-side 10 in the same minutes, and top quarter for that since 2024.

Beyond the kicking, I can’t yet show you much that has changed. The team stats I’ve got are totals for the whole game, things like passes and rucks, and they can’t see attacking shape, width or who’s standing at first receiver, which is where most of a new coach’s work goes. On those totals nothing has moved further than the normal ups and downs. Seven tests is early, and four of them were against the world champions.

Who did the jobs?

I’ve left a single player rating out on purpose. I tried a couple, and they either rewarded the backs who made line breaks or turned into a count of mistakes that punished the busiest players, and neither could see what a prop does at scrum time. So these are straight counts you can check against the tape.

Jasper Wiese got over the gain line more than anyone, 37 times. Jordie Barrett won the most turnovers and made the most tackles, 55 of them. Pieter-Steph du Toit made 47 but missed 15, which is what happens when you’re the one throwing yourself at everything. The penalties are nearly all props, and most of those came at the scrum.

If I had to sum it up

Ellis Park was levelled at the breakdown and won by New Zealand’s finishing. The other three came down to small things like goal kicking and sin-bins. It’s four tests, so none of this proves what caused what, but it’s a fair picture of where each side was better. Same two things every week: South Africa up front, New Zealand in space.

How this was put together, and what I had to fix

The data is World Rugby’s match-centre stats (Opta) for these four tests and for every men’s test since 2023 with a full stat sheet. The models were built without these four tests and allow for how strong each team was that year. Heights, weights and ages came from the players’ current profiles, matched on date of birth.

I rebuilt the model twice. The first version was mostly measuring the scoreboard, because stats like line breaks and kick counts go up when you’re scoring, so it looked great and told you nothing. The second still used total metres per carry, which one long break can inflate, so the final version counts carries in normal phase play only. I also ran it on close games alone, and on games between the very top sides.

With four tests, none of this proves one thing caused another. And I’ve left the model’s uncertainty ranges off the figure on purpose: they came out narrower than I believe for a fit on 226 games.

If you coach or analyse at this level and think I’ve read something wrong here, I’d like to hear it. Four tests is a small window, and anyone who watched these games properly will have seen things the stat sheet can’t. Tell me what I’ve missed.

Reference

• Bunker R, Spencer K (2022). Performance indicators contributing to success at the group and play-off stages of the 2019 Rugby World Cup. Journal of Human Sport and Exercise 17(3), 683–98. The source for the possession finding above: what separates winners in pool games stops working once the knockouts start, while carrying over the gain line keeps mattering. arxiv.org/abs/2012.02099