24 replies to this topic

[F1Frog]

This is a slightly weird and complicated concept, but please hear me out


I have attempted to work out the expected points for the Formula 1 drivers in 2023. This works as following.


Firstly, in each race, I have attempted to rank the ten cars in terms of competitiveness. So there is an element of subjectivity there. As we are talking about points, it doesn't matter so much about the lesser teams in most races (as I have forgotten whether the Williams or the Haas appeared more competitive in Monaco), but in terms of whether the Ferrari or Mercedes was better in Jeddah, I have had to just choose one and sorry if you disagree with me on some of them.


In Bahrain, for example, I have ranked them in the order: Red Bull, Ferrari, Aston Martin, Mercedes, Alpine, Alfa Romeo, Williams, Alpha Tauri, Haas, McLaren.


So that means that Verstappen and Perez are both expected to come 2nd (I did it like this so there is more of an element of gaining points as well as just losing points). Leclerc and Sainz are expected to come fourth, Alonso and Stroll sixth etc. But in Bahrain, Leclerc retired through no fault of his own. So this shifts so that Leclerc is expected to get zero points, but Sainz is expected 15, Alonso and Stroll 10 etc.


However, in Jeddah, due to Verstappen's bad luck in qualifying, I have shifted it to be Perez expected 25, Verstappen expected 18, then Alonso 15, Hamilton and Russell 10, Sainz 8 and Leclerc 6 due to Leclerc's engine penalty etc. These shifts to decide which teammate should finish ahead of the other are not extended between teams as that is a step too far from objectivity (e.g. in Silverstone I have Hamilton expected ahead of Russell due to the safety car, but not Hamilton expected ahead of Piastri as it is assumed these little quirks can only affect which teammate should be ahead. Some races like Zandvoort and Las Vegas are difficult in this regard).


In Australia, it is more complicated. Perez made a mistake in qualifying that meant it was easy for Verstappen, but as it was his own mistake, both Red Bulls are still expected to get 18 points and anything Perez does worse than that due to his mistake is points dropped. Hamilton is expected 15 and Russell 0 due to the engine failure. Alonso and Stroll each expected 10. Leclerc and Sainz are both expected 8 as each of the other made a mistake that should have given the other an easy sixth, but as they both made a mistake they both dropped points. As Sainz finished, I count that as a position and move down to seventh as the next one. Alpine was the next best car so Ocon and Gasly should have finished seventh and eighth. But each had a problem so I would move them both up to seventh as their expected position. As the crash is Gasly's fault, he is expected 6 points, but Ocon wasn't to blame for his retirement so is expected 0 points. Next best car was Williams. Albon and Sargeant are both expected seventh now that the Alpines are not there. But they both crashed out as well through their own mistakes, so both are expected six points. Then it is Haas. Magnussen and Hulkenberg are expected 6 points each but Magnussen crashed out through his own mistake, however Hulkenberg finished. So next is McLaren, who should be eighth and ninth so are both expected 2 points. And next comes Alpha Tauri so Tsunoda is expected 1 point as De Vries was taken out through no fault of his own.


One difficulty is when a driver who gets a bad result through their own fault should move the people behind up, and whether they should be said to have outperformed the car by beating that driver. I have decided to do it based on if the driver finished the race, the drivers in slower cars who beat him are still expected to finish behind (e.g. Sainz in Australia), but if they retire, the drivers behind are now expected to do one better.


Does this make sense? Or do I need to explain it more?


It will say how many points each driver should have scored, in isolation, so can be compared to the main table. And it is subjective because it depends on my opinion about how good the cars are, so is not meant to be definitive and does not prove anything. It is just meant to be something interesting to evaluate. I hope people find it interesting.


I haven’t evaluated the sprint races or fastest lap points at all. I wish they had never happened. So at the end I have just removed these points from the actual championship table for a more direct comparison.

So here are the results (in brackets is the change of points required to get to their actual total when sprints and fastest laps are ignored):

Max Verstappen 392 (+129)
This means that he scored 129 points more than expected in 2023. This is because he is generally expected to finish second, but always finishes first.

Sergio Perez 392 (-134)
So Perez dropped 134 points in 2023 through underperformance. I think this is reflective of just how terrible a season he had.
Note: He has the same expected score as Verstappen as he is expected to win Baku and Jeddah due to his teammate's bad luck, but his retirements in Mexico and Japan mean Verstappen is expected to win those. So Perez had better luck than Verstappen in 2023.

Carlos Sainz 189 (-11)
Carlos Sainz should have finished third in the championship this year. The Ferrari was generally the second-best car and he had slightly better luck than Charles Leclerc.

Charles Leclerc 174 (+11)
The Ferrari may have been the second-best car but Leclerc had slightly more bad luck than Sainz. Either way, he should have beaten Hamilton and Alonso in the championship.

Lewis Hamilton 166 (+47)
This suggests that, by finishing third in the championship, Hamilton outperformed his car in 2023. He scored 47 points more than he should have, and did very well to allow Mercedes to beat Ferrari in the championship with a weaker car.

George Russell 159 (-3)
George Russell had slightly worse luck than Hamilton, but not really enough to explain the difference in points between them. However, Hamilton's destruction of Russell was more down to him having a great season than Russell having a bad one.

Lando Norris 141 (+42)
Lando Norris was another driver to outperform his car in 2023, scoring 42 points more than he should have, and therefore did very well to finish on par with the Ferraris in the points despite having a lot fewer expected points than them.

Fernando Alonso 124 (+73)
But of all the drivers in the fight for fourth in the championship, Alonso had the least right to be there, consistently outperforming his car to finish fifth in the standings.

Oscar Piastri 107 (-26)
Piastri may have had worse luck than Norris, but he still dropped 26 points over the course of the season, and wasn't in the same league as his teammate.

Lance Stroll 104 (-36)
Stroll had slightly worse luck than Alonso but not nearly enough to justify the points gap between them, and dropped 36 points over the season.

Pierre Gasly 30 (+24)
When Gasly has a bad day, he doesn't drop any points because the Alpine usually isn't quite expected to. But on a good day, he can gain points.

Esteban Ocon 12 (+44)
Ocon has a much lower expected score than Gasly due to his retirements in Bahrain, Australia, Singapore and COTA, all races when Alpine were expected to score decent points. So while Ocon scored slightly fewer points overall than Gasly, I think he had a lot more bad luck and was the better Alpine driver.

Yuki Tsunoda 12 (+1)
Tsunoda generally did just as well as he should have over the season, dropping some points but gaining them back in other places.

Alexander Albon 11 (+14)
The Williams was capable of points but Albon got more out of it than expected in a very impressive season.

Logan Sargeant 11 (-10)
Logan Sargeant, on the other hand, had exactly the same opportunities to score as his teammate, but failed to do so.

Daniel Ricciardo 8 (-2)
While he gained points in Mexico, Ricciardo should have scored in Spa and COTA.

Nico Hulkenberg 7 (-1)
He got the expected seventh place in Australia, but should have nicked another point in Singapore.

Kevin Magnussen 5 (-2)
The Haas was capable of that eighth and seventh place Hulkenberg achieved in Australia, and a point in Singapore, but Magnussen's other points were outperforming the car.

Nyck de Vries 2 (-2)
The Alpha Tauri was capable of a point in Baku and in Monaco but De Vries couldn't get them.

Valtteri Bottas 2 (+8)
Most of the time, the Alfa Romeo wasn't good enough for points and while Bottas should have scored in Hungary, he outperformed the car in Qatar for those points.

Zhou Guanyu 2 (+4)
Similarly, Zhou should have scored in Hungary and Bahrain, but instead managed it in Spain and Australia.

Liam Lawson 0 (+2)
The Singapore points were a welcome bonus and unexpected.

So that is my 'expected points' system. I will try to go through it in more detail at some point when I have more time. I hope you find it interesting and don't despite it too much

Edited by F1Frog, 12 December 2023 - 17:19.

[pacificquay]

You “expect” Lando should have scored 99 points and Piastri 133.

 

There is no world in which anyone would reasonably expect that, therefore this system is flawed.

[Ferrim]

No. He expects Lando to score 141, he actually scored 183 (ignoring sprints and fastest laps). For Piastri, it is the other way around.

Edited by Ferrim, 12 December 2023 - 21:57.

[eab]

I haven’t evaluated the sprint races or fastest lap points at all. I wish they had never happened. So at the end I have just removed these points from the actual championship table for a more direct comparison.

So here are the results (in brackets is the change of points required to get to their actual total when sprints and fastest laps are ignored):

Fernando Alonso 124 (+73)
But of all the drivers in the fight for fourth in the championship, Alonso had the least right to be there, consistently outperforming his car to finish fifth in the standings.

Alonso was even better, he finished fourth 

[eab]

Also, going by this score, Ocon standout performer. DOTY material.

[danmills]

Applaud the effort gone into this but its absolutely bonkers.

 

It's like saying I should have eaten a salad but I ate McDonald's three times last week. I weigh 103kg but I should be 92kg. Therefore +11kg I'm doing worse than why I hypothetically should have eaten if I extracted my maximum eating potential.

 

Should, would and could are not fixed values. This also basically puts the value on Saturday far too much. I can't fathom how Trulli and Toyota would have fared in this system.

 

Should have been WDC probably. 

 

Fun thread though. 

Edited by danmills, 12 December 2023 - 23:40.

[noikeee]

I like xG in football but I really don't think it works in F1. At least not with this methodology, sorry. I just don't think we have enough data to figure out which car was the 5th or 6th best each GP. Even if we did there are still a hell of a lot of holes in the method that don't make this particularly useful, IMO. But I admit it probably made for a fun experiment on your side.

[Myrvold]

Frog. Do you go under the nickname AT13579 on Reddit? 

 

That user presented an overview of JVilleneuves vs his teammates throughout his career, and managed to make Frentzen outperform JVilleneuve that season. Because, e.g Australia was a win for Frentzen, due to JV's bad start, Frentzen was ahead of JV when he was taken out in T1. Thus, Frentzen was the better driver that race. That also meant that when Frentzen was taken out in Germany that year on the first lap, he was given the "win" in that head-to-head battle, as he was ahead of JV when he was taken out.

 

I get the same vibes here. There are so much subjectivity here, that it's impossible to get anything expected from it.

[ARTGP]

I wonder if it's necessary to do the "2nd place" thing. Let's just assume the driver should have beaten his teammate, then the +- ranking might also capture who maximized the car, and who wasted it. 

 

This stood out to me: 

 

 

 

This means that he scored 129 points more than expected in 2023. This is because he is generally expected to finish second, but always finishes first.

 

 I can't tell but for example, did you calculate that Lec/Sainz should each have finished 4th in Las Vegas? I would have done it slightly differently (expectation would be 3rd). 

Edited by ARTGP, 13 December 2023 - 03:13.

[NewMrMe]

I am not sure there is a valid starting point for calculating this.

 

Expected goals in football works as that is based on similar shots from the same area of the pitch and what proportion of them ended up being goals. I don't think there are any identifiable incidents or stats during a race that you can use to determine where a car should finish. As others have pointed out basing it on ranking the teams has too much subjectivity.

 

Motorsport's equivalent to expected goals would be more related to overtakes. For example, you could look at the gap between cars leading onto a straight and see how often that gap has resulted in an overtake into the next corner.

[southernstars]

Oscar Piastri 107 (-26)
Piastri may have had worse luck than Norris, but he still dropped 26 points over the course of the season, and wasn't in the same league as his teammate.
 

 

This is striking me as unfairly harsh, considering that Oscar had multiple times over the course of the season (notably Austria, Silverstone, and Singapore) where he was one or two upgrades behind Norris (particularly Austria, where Norris had the major upgrade, and Singapore, where Norris had eleven new parts and Oscar one).

 

It should also be considered that he got beaten up by a lot of more experienced drivers at a far greater rate than Norris - Gasly, Hamilton (twice!), Ocon, Tsunoda, Sainz (three times!), Norris, Magnussen...after Qatar, he was hit and damaged in every race until Abu Dhabi.

[PayasYouRace]

I find the premise extremely flawed. If Red Bull are the fastest in a race, then Max is expected to finish first and Checo is expected to finish second. Thats the way of it for most teammate combinations.

[danmills]

The results might look more interesting if it was done that way. It would make Checos expectation more realistic.

 

This system seems to make number 2 drivers look way worse.

[messy]

Is this like the F1 equivalent of XG in football?

It's interesting but I'm not sure it works.

Edited by messy, 13 December 2023 - 09:41.

[1player]

That is a very interesting model, but it is completely based off subjective data (i.e. what you call accounting for bad luck and other factors, also the order of teams) which means you can basically have it show whatever result you want.

A good statistical model needs to only have the model itself as a variable you can tweak, but the input data needs to be as objective and indisputable as possible, so that if we agree on the model, we agree on its output.

So at first sight I quite like the result it's produced, but perhaps it is because your bias and mine are similar.

Still, it's more work than I have done, so well done for the effort, and it's a step towards something that passes the smell test

[PlatenGlass]

Btw if e.g. you're saying Red Bull is the best car at a given race and the drivers should finish 1-2, it seems to make more sense to give them each an expected position of 1.5 rather than 2.

[PayasYouRace]

How many teammates are expected to dead heat in a race?

[Collombin]

I wonder if a variant of the rugby world rankings system could ever be used, in essence everyone has a score that goes up or down based on results, with a higher scored driver getting less credit for a good result because it's become expected of them. Nah, sounds crap actually.

[F1Frog]

I think Perez is actually one of the few drivers that this is an accurate representation for, as I am predicting that he finishes second to Verstappen most races, and so anything less than that is dropped points. It is more unrealistic for Verstappen because he didn't really outperform his car at all this season, because there was no way of outperforming it. You can't do better than first.

 

Esteban Ocon does very well partly because he beat a lot of faster cars in both Monaco and Las Vegas, in my opinion. The latter was lucky, the former not at all. Also, in Jeddah for example, the Alpines are expected to finish eighth and ninth as they do, but as my system predicts them both to finish ninth, Ocon gains two points there. He didn't really outperform the car, he just outperformed Gasly. But also, if Ocon has a terrible race like in Monza, for example, the Alpine isn't expected to score points so he loses nothing. If Sainz has a terrible race like Abu Dhabi, he loses a lot of points here.

 

I could do it based on fastest lap of the weekend for the order of the teams to make it objective rather than subjective but that would create different problems, such as the Haas being great in qualifying but not in races, or rain in Canada meaning that practice 2 would set the order of the teams.

 

Overall, I think this system is mostly just useful for things like showing that Leclerc had worse luck than Sainz and Ocon than Gasly, to explain points differences. In intra-team instances, someone like Alonso gains points due to having a weaker teammate that he can always beat, while the Ferraris take points off each other. For comparing Hamilton and Leclerc, for example, it is useful for me but not for anyone who doesn't trust my evaluation of the pace of the cars. I think I would need to post a proper analysis of each driver's season according to this model for it to have some meaning for other people, which I will try to do at some point but I am quite busy right now.

 

I could try PlatenGlass's suggestion.

 

Overall, it is never going to be a perfect model. Although neither is expected goals in football because I don't believe it takes into account where defenders and goalkeepers are? And even if it does, it doesn't know who the player shooting. A shot from the D on the edge of the box with a packed penalty area probably doesn't have a high score for expected goals, but if Conor Chaplin is taking the shot I feel confident that it's going in.

[PlatenGlass]

How many teammates are expected to dead heat in a race?

 

Well it's like how the "expected" score on a dice roll is 3.5. It's the mean rather than what someone would actually expect. This is kind of analogous to that.

Edited by PlatenGlass, 13 December 2023 - 17:40.

[PayasYouRace]

Well it's like how the "expected" score on a dice roll is 3.5. It's the mean rather than what someone would actually expect. This is kind of analogous to that.

But weren’t not talking about a random outcome. We’re talking about a prediction of where one teammate would finish relative to another.

[Clatter]

I may be missing something, but how does rejigging the results from 2022 equate to the expected results in 2023?

[PlatenGlass]

But weren’t not talking about a random outcome. We’re talking about a prediction of where one teammate would finish relative to another.

But I think it's based on car performance on the basis that, subject to reliability, each driver in the team has an equal chance to finish ahead.

[PayasYouRace]

But I think it's based on car performance on the basis that, subject to reliability, each driver in the team has an equal chance to finish ahead.

Well that’s definitely a fundamental flaw in “expected results”.

[PlatenGlass]

Well that’s definitely a fundamental flaw in “expected results”.

I think you're focusing too much on the word used rather than the idea behind it.