Written for shauna@social.coop on Mastodon

Please don't read any further if you want to maintain the belief I know what I'm doing

I was original entranced by this post - https://gephi.org/users/quick-start/

So I downloaded Gephi and then looked for some data to interrogate. Initially, I wanted to look at how the players at Euro 2012 were interconnected - https://fulltimesportsfan.wordpress.com/2012/06/08/finalised-diagram/

It's pretty much been the same process ever since, with some learnings that have been incorporated in the oh my goodness 13 years!

Information source - I use the Wikipedia squad pages. This is why you'll occasionally see notes about "delayed because of" and a warning that the data is taken from Wikipedia so may not be accurate. The club team a player plays for is the one most likely to be wrong, especially for competitions that are between seasons. (The strange places I've ended up at on Wikipedia because of edit wars about which club someone plays for.) It is, to an extent, worse for the rugby union ones because players move at the end of the season, not the start of the next.

Because I do the input (and the removals when teams are out) by hand, I go through the national teams starting with group A and move downwards. This is why, if there is a delay in the teams in group A naming their squads, it causes a huge knock on effect. (The time Italy delayed their squad announcement till ~ 7 pm on the night of the deadline for a game show caused significant cursing because they were the first team in group A and I couldn't start till they did.)

The larger the event and the bigger the squads obviously, the longer it takes.

If you can python (I can't, one day I will learn etc), James Ashford wrote a really nice post on how to do all of this with Python - https://james.ashford.phd/2023/08/25/analysing-the-2023-fifa-womens-world-cup-with-graph-theory/

Me, I manually add things into Gephi. Sometimes this means I notice things (like the number of Zambian players playing in the Kazakh Women's League, or that there is a Saudi Women's League - https://fulltimesportsfan.wordpress.com/2023/07/22/womens-world-cup-2023-group-stage-network-diagrams/)

I use the player as the source and the national or club team as the target. I've experimented with using both directed and undirected links, and it doesn't make much of a difference.

There are other layouts, but I like the way force atlas looks. (Force atlas works okay with this size data set. For significantly larger ones, Yifan Hu is easier)

I start with the pre-set values, increasing the repulsion strength if the teams crash into each other.





I like to add each player individually, because one of the things I enjoy is seeing the shape and positions change with each addition, but I'm sure making more links at the same time would make it go quicker.

For the colour and the size of the circles, I keep it really simple and stick to number of degrees. For size, I set minimum at 10 and maximum at 50. I find it's large enough to see the small changes with each player and clear enough when I've oopsed and not attached a player to the right national team. Or the Wiki page is wrong and hasn't taken off the players that didn't make it to the final squad. Or France have decided to only pick 25 players when they could have picked 26 for who knows what reason. I also like it for the rugby union one (because they have unlimited replacements for squad injuries) because it creates a subtle gradation for each injured player.

Colour is slightly more complicated. I like to try and use either the tournament colours or the colours of the host country flag but it's not possible to change the Gephi pre-sets (or at least not as far as I can find out) so sometimes I have to go with nearest to those. I know to not complain about free software but it's the reason I keep looking to see if there's a way to do something similar in R.

I keep link width at 1.

For closest to the centre, I use the zoom function, and the degree function to get the number of players.

When teams get knocked out, I remove the players manually, hence why I try to keep the teams in order when I add them. Again, you get some fun shape and pattern changes that way.

Finally posting another "100 Great Scenes In Not So Great Films" entry means I can update the charts too.



This is one where I feel bad because the people who are linking Ad Astra to other films are not the problems with this film or the others.

3 people link Ad Astra to 4 other films.

Ad Astra is linked to Assassin's Creed by producer Arnon Milchan, who is the pink circle underneath the blue A.

Hoyte van Hoytema (the pink circle to the right of the blue B), who is an excellent cinematographer, and did I mention how good his work was in Ad Astra, links Ad Astra to Spectre and Tinker Tailor Soldier Spy.

Robert Alonzo who works in stunts and is the pink circle to the right of the blue letter C, links Ad Astra to League of Extraordinary Gentlemen. That was the link that really changed the shape of the overall diagram pulling League of Extraordinary Gentlemen in from it's point on the far right of the diagram into the centre.

The figures are obviously very simple now that there are only two teams.



The same graph, but labelled:



Gotham FC are the club team closest to the middle.

Barcelona, unsurprisingly, have the most players left, with 10. The teams with the next most players remaining are Chelsea, Arsenal and Manchester City with 6.

Manchester City, Arsenal and Gotham FC are guaranteed to have a player on the winning side.

The community view gives less information when we're down to two teams so I haven't shared that.

It's a final to look forward to.

The group stage diagrams and predictions based on them can be found here - https://fulltimesportsfan.wordpress.com/2025/06/29/euro-2025-group-stage-network-diagrams/

How did the group stage predictions go?

From group A, I predicted Norway and Switzerland would qualify for the quarterfinals, and I was right, even if it involved extra time goals in the decider.

Group B, I predicted Spain and AN other, which I am aware is a bit of a "the sky is blue" sort of prediction.

Group C, I predicted Germany and Sweden

Group D, I predicted Netherlands and one of England and France, which was oh so wrong.

So out of 8, I am willing to call that about equivalent to 5/8.

With that in mind, here are the quarterfinal network diagrams

( Four figures under the cut )

Sweden are the national team closest to the centre (just, vs. Norway). Either Lyon or Bayern Munich are the club team closest to the centre.

Barcelona are the team with the most representatives left in with 14, then comes Bayern Munich and Juventus with 13 then Chelsea with 12.

Predictions for the quarterfinals:

These are quite difficult because Italy and Spain are pulled out by how many of their players play for Juventus and Barcelona respectively, while Bayern Munich, and the players that play for them, are pulling together Sweden and Germany, and Arsenal and Chelsea are holding together England, Sweden and Norway.

Sweden vs England - Diagram says Sweden

Norway vs Italy - Diagram says Norway, plus every single women's football pundit keeps bewailing how often Italy somehow manage to screw up. On the other hand, Norway trip over their own feet also.

France vs Germany - Diagram says Germany just. Football fan says "ooooh".

Spain vs Switzerland - Remember how I said Barcelona pulled Spain out of the diagram. This really reflects that. Switzerland are far closer to the centre. On the other hand, there is no way I can see Switzerland beating this Spanish team.

Now in a slightly different format.

The Making Of:

UEFA saying teams had to announce their teams by the 25th of June meant that I have had plenty of time to make these diagrams. That the women's teams are limited to only 23 players also sped this along.

Interestingly, while coverage and interest in women's football had increased hugely, the Wikipedia pages are still updated much more slowly than the equivalent men's pages. I was making the diagrams and spotted that Italy had a much bigger and darker red circle than the other teams, and when I checked, it was because the Wikipedia page hadn't been updated following the cut from 27 players to 23.

The Diagrams:

( Four diagrams below the cut )

Some Observations Based on the Diagrams:

Every country has at least one player playing for their home league, except Wales. Every country has at least one player playing abroad - for Italy it is literally only the one (Arianna Caruso for Bayern Munich).

It is not clear if Germany or Switzerland is the country closest to the centre. It is clear that Bayern Munich are the club team closest to the centre.

The club teams with the most players present are Barcelona with 17, Bayern Munich with 16, then Juventus and Chelsea with 14.

For most teams, the players are spread over several teams. The exceptions to this are Italy, Spain, Portugal and Germany, although with Germany it's less obvious when looking at the diagram because a number of non-German players also play for them. England are more weakly like that, with a lot of players from Chelsea, Arsenal and Manchester City.

The number of links between Italy and Denmark is due to the large number of Danish players playing in Italy, which might be related to similar patterns seen in the men's game. Danish acquaintance of mine used to complain players on had to be signed by Italian clubs to walk into the national team.

Having done this for women's tournaments before (see the 2022 version of this here - https://fulltimesportsfan.wordpress.com/2022/07/08/womens-euro-2022-network-diagrams-group-stage/), I'm slightly saddened that we seem to be losing those teams where the main team is the women's team; there's no players for Turbine Potsdam for instance, and only a few from London Lionesses, Paris FC and Madrid CFF. I don't want the increased interest in women's football to cause it to lose its history. (I still don't forgive the English Women's Super League for screwing over Doncaster Belles.)

Some of the old divisions still remain; Liverpool not giving a flying curse about it's women's team is reflected by there only being two Liverpool players present, both for Wales, compared to eight for Everton. Everton have always supported their women's team - when I was young, the only chance women in my area had of playing football properly was in their women's team.

One thing that might be affecting the clustering is the number of players who play for US teams. A country with a lot of club teams represented but not present itself doesn't usually happen, unless one of the big guns doesn't qualify (looking at you, so often, Italy).

I will be keeping an eye on Poland, particularly Emilia Szymczak who is 19 and plays for Barcelona B. If you're good enough to be picked up by Barcelona at that age ...

In terms of the community view, the national teams that group together are Sweden and England (because of Arsenal and Chelsea) and Switzerland and Germany (because of Eintracht Frankfurt, Bayern Munich and Wolfsburg). Switzerland and Germany being joined is really interesting given Germany's circle overlaps with Norway's, and yet those two aren't linked.

Predictions:

L likes me to try to predict the outcome of the games from this, and there is some correlation between closeness to centre and connectedness and doing better in tournaments. However, previous experience (https://fulltimesportsfan.wordpress.com/2023/08/03/womens-world-cup-2023-last-16-network-diagrams/) has shown that it doesn't work as well for women's football.

Because half the teams will be gone after the group stage, it's a lot harder to predict, and it makes the games so much more tense. Limiting it to 16 teams also means some of those groups are stacked - like group D - England, France, Netherlands and Wales. So that's the present World Cup holders, the winners of Euro 2022, the winners of Euro 2017 and the lowest ranked team in the competition.

Bon chance, Wales.

(Actually the diagram doesn't have them as separated from the other teams as I would have expected.)

Spain are the team that really confuses the diagram. I would expect them to be a lot more central, given I expect them to do well. It could be the number of players that also play for clubs with Portuguese players that is pulling them out there or potentially a sign that they may not do as well as expected.

Running purely off the diagram, the teams most closely clustered are Germany, Norway, Switzerland, Sweden, Netherlands, England, France, and then one of Denmark and Spain.

You'll notice that's 9 teams and only 8 go through.

If we take Switzerland and Norway from group A, that leaves 3 more groups where it is unclear.

For group B, it is unclear because only Spain are in that central core, and they're barely in it.

For group C, as Germany and Sweden are closer to the centre than Denmark, I will predict that these are the teams that will get through.

For group D, D for death, that logic can't work because England and France are a similar distance from the centre of the cluster. My best prediction - that group is going to be tight.

Game 26 was Saints away at Huddersfield, which Saints won (https://www.saintsrlfc.com/matches/2024/first-team/huddersfield-giants-v-saints-2024-09-01/?swcfpc=1)

The biggest news to my mind was Morgan Knowles coming back. I was not alone in this opinion - https://www.sthelensstar.co.uk/sport/24556400.morgan-knowles-brought-saints-return/

The St. Helens Star, a biased source I grant, said, "He missed the best part of three months with a groin issue – a period that coincided with the beginning of Saints’ picking up other injuries and then subsequent run of defeats." (https://www.sthelensstar.co.uk/sport/24556400.morgan-knowles-brought-saints-return/)

It also points out he then missed 3 games due to a ban for a high tackle.

Saints lost 7 of the 11 games Knowles missed.

The really terrifying thing is that 2024 was Knowles's 10th year with Saints. Time flies, eh?

While the game was a victory for Saints, it also highlighted a worrying trend for yellow cards (although I forgive Noah Stephens entirely).

Bennison having to do the kicks reassured me in the "there is another" with regards to kicks if Percival is off the pitch.

None of match-specific pictures are all that interesting so I'll move on to the season to date diagrams.

Seeing Bennison shoot up the "who scores for Saints?" diagram after just one game shows how important the kicker is.

( Under the cut )

Robertson is now on the list after scoring his first ever try for Saints. Overall 24 different players have scored either a try or conversion for Saints in 2024.

When do Saints score?
( Under the cut )

Who is present when Saints score?

( Under the cut )

The matrix of who is present when Saints score is interesting:



Normally it would go darkest (most often together) in the bottom right hand corner and paler (less often together) as it moves up and to the left.

This time, that pattern happens but then there's a suddenly dark border along the top and left which consists of Whitley, Bennison and Batchelor, suggesting Saints score when they are on the pitch together. Which makes some sort of sense because Batchelor definitely missed some matches with an injury.

The equivalent network graph is slightly different again.



It's interesting that two ways of presenting the same data give subtly different results.

There's no real changes to the pattern of the "who scores against Saints" diagram so I haven't included it.

The last 10 minutes of the game is still when Saints are most vulnerable.

( Bar chart under the cut )

Blake also leads the "who is present when Saints concede?" chart

( Bar chart under the cut )

The matrix diagram of who is present together when Saints concede is not as pretty as it was after game 25 (https://fulltimesportsfan.wordpress.com/2025/06/11/saints-ahoy-game-25-and-the-2024-season-to-date/). It think it's because the "curls" of more often together in the mid-section are less well defined than they were last time.

( Under the cut )

Unlike the "who is present when Saints score?" matrix and network graphs, the concession network graph mostly matches the matrix diagram.

( Under the cut )

Game 24 of Saint's 2024 season was their Magic Weekend game.

I quite like Wikipedia's description of Magic Weekend - "an annual event organised by the Rugby Football League in which an entire round of Super League matches is played over a weekend at a single stadium to promote the sport of rugby league." (https://en.wikipedia.org/wiki/Magic_Weekend)

I've been lucky enough to go twice, both in Newcastle.

The only downside to Magic Weekend is, that in order to sell tickets, Super League tend to have teams play their local rivals. Which means you can end up playing the same team far too often.

And playing Wigan, again, in a year your team are already not doing well, is far, far too often.

When Saints then lose, 20-0, to Wigan, that's the pits - https://www.saintsrlfc.com/2024/08/17/saints-beaten-on-derby-day-at-magic-wknd/

On the other hand the referees let this sort of thing go:
( Picture under the cut )

Understandably, there can be no diagrams for Saints's point-scoring moments in this game, nor are their any updates to their point scoring moments for the season.

There is no pattern to when Saints conceded, except maybe a slight suggestion that they concede more in the last 10 minutes (but so does everyone else).

( Bar chart under the cut )

The "who is present when Saints concede" diagram is so weird that it made me double check that I'd not made some sort of data entry error.



It makes sense, just, if you consider that he's about the only player who didn't have an extended injury / suspension break.

It does make the "Who is present when Saints concede" matrix look intriguingly different.

( And makes the alt text tricky to write )

The equivalent network graph is shaped like a fox's face.

( Under the cut )

As Royle, Walmsley, Wingfield and Burns are the only players sticking out, it tells me that the other players who had previously been in the little "rarely but when they do they play together" sticking out blobs have now been subsumed into the central blob. This is true, as they are now at the edges of the central blob (Paasi, Vaughan and Stephens on the left hand side and Bennison, Knowles and Batchelor on the right).

That change is most interesting, and suggests Saints have had to start leaning on the full squad of players.

(Yes, I know we're almost at the end of the 2024-2025 season, work with me here)

One major thing, for the purposes of this update post, did happen since the 2022-2023 version of this post. Harry Kane moved to Bayern Munich.

This was probably a very good thing for him (something is definitely rotten in the state of Tottenham Hotspur).

And I've previously included stats from non-Premier League football for Salah, so it's definitely doable method-wise.

It does mean that it will no longer be as direct a comparison, but I think that's okay.

On to the graphs:

First, we have percentage of games played, up to the point of being 30 for all three players. This comes first because the original question L posed was whether Kane's putatively dodgy ankle would allow him to catch Shearer's Premier League goal tally.

Looking at up to 30, Shearer's curve has a serious down curve here, because 30 was when he had his second big injury (carrying that Newcastle team).

Although Salah and Kane's curves have different shapes, they are now converging on the same point in terms of percentage of games they played at 30. Whether this is changes in how coaches use players, resting them more often now to allow more games overall, I don't know.

( Under the cut )

Using the "extrapolated till age 35" data, the curves look like this:

( Under the cut )

It's Salah's curve that shows the greatest drop here (because at 31 he had an injury that kept him out of several games [Hamstring tears are no one's friend), while Shearer and Kane's curves converge to similar points.

Now Goals per game

Using the data up to everyone reaching 30

( Under the cut )

Again, Shearer is the odd one out because of his injury, although to me, the interesting thing is how similar the shape of the curves are for Kane and Salah. The curve also makes it clear that, although Salah is indeed a very awesome striker, Kane is the one who is more of an out and out goal getter. Plus, the entirely ridiculous stat that Kane had a goal per game in 2023-2024.

Using the extrapolated data, Kane stands out:

( Under the cut )

Kane's curve ends much higher than the other two, possible because of that ridiculous 2023-2024 season, with Salah's curve going down because of his injury in 2023-2024.

Finally, we have the goals per possible game, which would normally perk Shearer's stats up because he had fewer opportunities to have games to score goals in, but, as I said, his injury in the year he was 30 meant his curve really does parabola downwards in this graph.

( Below the cut )

The other two curves, again have the same shape, but Kane is higher again.

On the extrapolated till 35 goals per possible game

( Below the cut )

Shearer and Salah have similar trajectories, possibly because they both had bad years around 30/31 with injuries, while Kane's curve has gone shooting off at the top because of his excellent 2023-2024.

For those who don't know, maths is not my subject. I enjoy statistics and it's the only thing that stopped me failing A-Level maths, because it makes sense.

So when so many members of the Tashkent City Women Professional Cycling Team and the Human Powered Health Team withdrew in different stages of the race that the lines in the Kaplan Meier graphs go into the negative, the maths part of my brain is screaming because you can't have negative survival rates, and the statistician in me says either those teams were awful or something appalling happened.



When the green line for Tashkent then goes positive again because the amount of riders they lost in another stage versus the amount that remained, I become reasonably convinced of my "awful or appalling happened" thesis.

It turns out there's something more interesting going on with the Tashkent City Women Professional Cycling Team. They are a Continental Tour team who gamed the system to get a women's Tour de France invite, because that was their aim (https://www.rouleur.cc/blogs/the-rouleur-journal/200-euro-salaries-inexperienced-youngsters-and-gaming-the-system-tashkent-city-justify-their-place-at-the-tour-de-france-femmes).

And that's before we point out that most of them were teenagers, in a sport that acknowledges it's hard to finish the race that young (https://www.cyclingnews.com/news/people-can-think-what-they-want-tashkent-city-women-respond-to-criticism-after-four-riders-abandon-tour-de-france-femmes/)

I'd also like to raise a toast to Yanina Kuskova, the one Tashkent City rider to finish. She's now riding for Euskadi.

This race did seem to have a lot of withdrawals.

( Bar chart of withdrawals in the TDFF 2024 )

It appears that stages 4, 7 and 8 were diabolical.

( Pie chart of withdrawals by stage )

The all-rider Kaplan Meier chart also bears this out.

( Kaplan Meier chart under the cut )

Over the years I've been doing this for both the men's and women's Tour de France, this is is the lowest percentage of finishers.

Stage 4 was just crazy difficult - https://www.cyclingweekly.com/racing/puck-pieterse-pips-demi-vollering-in-photo-finish-sprint-to-win-stage-four-of-the-tour-de-france-femmes

Stage 7 was again general evil, building up to the Alpe d'Huez.

Stage 8 was the Alpe d'Huez itself. Meet the Alpe - https://en.wikipedia.org/wiki/Alpe_d%27Huez#Cycle_racing

Interestingly, despite that, none of the withdrawals were due to racers finishing outside the time limit, they were either mid-stage abandons, or did not start the stage withdrawals.

( Pie chart of withdrawals by type )

Saints's game 23 was an unnecessarily close match that Moses Mbye won for Saints with a drop goal in golden point extra time (https://www.saintsrlfc.com/2024/08/08/saints-down-the-red-devils-in-golden-point/).

The matrix of players playing together when Saints concede for the game isn't particularly informative, other than showing the people interchanging in and out, but it does look pretty.

( Under the cut )

Over the season, there are now 178 point scoring moments for Saints, compared to 104 point-scoring moments conceded.

The chart of when Saints score continues to look like a skyline full of skyscrapers.

( Under the cut )

Percival still has the most point-scoring moment, because he is the kicker. He is now up to 60 point-scoring moments.

( Under the cut )

Waqa Blake is the Saints player present for the most point-scoring moments.

( Under the cut )

The matrix diagram of which players are together when Saints score is a lot more diffuse than it used to be.



The network diagram still has a clear central blob.

( Under the cut )

This Salford team had the third most point-scoring moments against Saints.

( Bar chart under the cut )

There is no pattern to when Saints concede

( Another bar chart under the cut )

Waqa Blake is present for the most point-concessions.

( Bar chart under the cut )

The matrix chart of players present together when Saints concede has two areas of darker red standing representing combinations often together when Saints concede. The first patch, Blake, Welsby, Clark, Sironen, Dodd, Lomax, Bell, Hurrell, Makinson, Percival, Mbye, Lees, Whitley and Delaney. The second is where those players cross over with Mata'utia. I believe this is because, before his injury, Mata'utia was in the main cluster.



The network graph for point-concessions is less blobby.

( And is underneath the cut )