113 replies to this topic

[statman]

Was thinking whether I should post this or not, but why not?

 

Journal of Quantitative Analysis in Sports. These authors have looked at the drivers since 1950 and came up with this analysis.

 

There are some surprises in there. Please make sure to read the abstract at the beginning to understand methodology and objective:

 

This paper uses random-coefficient models and (a) finds rankings of who are the best formula 1 (F1) drivers  of  all  time,  conditional  on  team  performance;  (b)  quantifies  how  much  teams  and  drivers matter;  and  ©  quantifies  how  team  and  driver  effects  vary over  time  and  under  different  racing conditions.  The points  scored by  drivers in  a  race  (standardised  across  seasons  and  Normalised) is used  as  the  response  variable  in  a  cross-classified  multilevel  model  that  partitions  variance  into team,  team-year  and  driver  levels.  These  effects  are  then  allowed  to  vary  by  year,  track  type  and weather  conditions  using  complex  variance  functions.  Juan  Manuel  Fangio  is  found  to  be  the greatest  driver  of  all  time.  Team  effects  are  shown  to  be  more  important  than  driver  effects  (and increasingly so over time), although their importance may be reduced in wet weather and on street tracks. A sensitivity analysis was  undertaken with various forms of the dependent  variable; this did not  lead  to  substantively  different  conclusions.

 

Full analysis here:

 

https://www.research...mance_1950-2014

Edited by statman, 14 April 2016 - 09:16.

[Marklar]

 

 

“Our statistical model allows us to find a ranking and assess the relative importance of team and driver effects, and there are some surprising results. For example the relatively unknown Christian Fittipaldi is in the top 20, whilst three time champion Niki Lauda doesn’t even make the top 100. Had these drivers raced for different teams, their legacies might have been rather different.”

http://www.sheffield...r-ever-1.567358

 

 

I totally respect these researches, but they are going nowhere for me. You can't put drivers in statistical models and judge them then based on that.

[statman]

Graphical explanation here:

 

http://imgur.com/N6HdYeT

[DampMongoose]

Well, I just hope they enjoyed the research.

[Marklar]

And that's the top 50. Well...
 

(additional info: if Schumacher was treated as one driver he would be 9th and Nico Rosberg would be 13th)

Edited by Marklar, 14 April 2016 - 08:41.

[sopa]

You could digest the list down to 10th position, but then I haven't got a clue, how did they get Christian Fittipaldi in there. LOL.

[scheivlak]

Nice  and honest try, but I'm not convinced that Christian Fittipaldi is a better driver than Lewis Hamilton.

Nor that Marc Surer is better than Damon Hill or Ronnie Peterson.

[statman]

You could digest the list down to 10th position, but then I haven't got a clue, how did they get Christian Fittipaldi in there. LOL.

Perhaps  the  biggest  surprise in  our  results is  the  high  ranking  of  Christian  Fittipaldi  at  number 11, despite  only  competing  in  three  seasons  and  never  making  a  podium  finish.  This  ranking  occurs because  C.  Fittipaldi consistently  outperformed  his  team-mates, and  because  he  never  raced  for  a ‘good’ team, the standard required to get a high ranking is lower.

More specifically, C. Fittipaldi’s teammates  had  relatively  high  rates  of  retirement:  he  gains  his  high  ranking  by  being  able  to successfully keep a relatively poor car on the track. Of course, this model cannot say that C. Fittipaldi
would  have  won  championships  had  he  raced  for  a  better  team,  and  his  confidence  intervals  are wider  than  most  of  the  other  highly  ranked  drivers,  but  the  results  suggest  that  in  one  aspect  of good  race  driving  at  least –that  is,  keeping  a  relatively  unreliable  car  on  the road –he  should  be highly regarded.

 

It's down to the team aspect: the team matters 85%, the driver only 15%, with additional team year changes and a constant team effect. The team matters much more, than the driver. i.e. great powerful team -> great results, and great or reasonable results by driver in a piss poor team vs great teams

Edited by statman, 14 April 2016 - 08:53.

[Marklar]

More specifically, C. Fittipaldi’s teammates  had  relatively  high  rates  of  retirement:  he  gains  his  high  ranking  by  being  able  to successfully keep a relatively poor car on the track. 

This is bullocks. It would be more impressive if his team mates didn't retired and he would have beat them on track.

[chunder27]

Hilarious, Alain better than the German!  And Fangio top.

 

As a man who prefers gentleman to banger drivers I would not argue with this!

[johnmhinds]

Their "research" seems just a little flawed, how does having a good team mate or a good car mean your race results mean less? Shouldn't it mean more if you're beating a good driver in the same car?

The top 50 is missing multiple world champions Niki Lauda and Mika Hakkinen for this reason.

Edited by johnmhinds, 14 April 2016 - 09:08.

[ANF]

Michael Schumacher treated as two drivers?

[Balnazzard]

Ye, not really...as pointed out Mika Häkkinen and Niki Lauda are missing the list entirely and Im pretty sure most would place Kimi higher than Barrichello, Brundle or Christian Fittipaldi....again its impossible to rank the drivers properly with this kind of "research".

Edited by Balnazzard, 14 April 2016 - 09:34.

[myattitude]

"Journal of Quantitative Analysis in Sports" - That's where it's gone wrong.

 

Thanks for posting it though.

[Grayson]

I quite like the idea of taking a fully objective and statistical view of the results (trying to strip out the quality of the car by comparing drivers to their teammates) and comparing the results to our own subjective views of the best drivers.

 

My biggest problem with this particular research is the totally subjective decision to treat Michael Schumacher as two drivers while not doing the same to anyone else. It's not that I disagree with this on a subjective level, it's just that it takes away any level of objectivity from the exercise. Why just Schumacher? Why not Raikkonen, for example?

 

This isn't a minor point which only effects one driver - it casts huge ripples into the rest of the rankings. As they mention, it would put Rosberg up from 49th to 13th. This also means that Hamilton and Webber would get more credit for their performances against the 13th best driver on the list, which means that their own teammates would get a tiny bump as well - and the ripples would keep going out.

 

Meanwhile, if we split Raikkonen into two drivers then it would effect every driver who's ever been his teammate as well. Kimi-1 would presumably do quite a bit better in the rankings meaning a boost for Coulthard, Montoya, Heidfeld, etc. and anyone who's ever been a teammate of those drivers. At the same time Vettel, Alonso and Grosjean wouldn't get quite so much credit for their performances against Kimi-2 as they would have done for driving so well against the guy who lit up the early 2000s!

 

The decision to split Schumacher into two drivers seems on the surface to be a very good one which only has a positive impact on the table... But it opens a Pandora's Box of questions around why that sort of subjective judgement wasn't used on other drivers!

[Nonesuch]

And that's the top 50. Well...

 

Schumacher pre-2006? The year he once again took the title to the final race? I'll assume they meant pre-2007.

 

[kraduk]

http://www.sheffield...r-ever-1.567358

 

 

I totally respect these researches, but they are going nowhere for me. You can't put drivers in statistical models and judge them then based on that.

 

Well you can as the article proves 

[Collombin]

I've always tended to rate Moss above Heidfeld and Christian Fittipaldi, but this analysis has really opened my eyes to the error of my ways.

Many thanks.

[Imperial]

Who paid for this crap?

[Oho]

Who paid for this crap?

 

Tax payers?

 

But it really does seem like a case where they should have realized that its not nature that got it wrong its their model....

Edited by Oho, 14 April 2016 - 10:10.

[sopa]

I quite like the idea of taking a fully objective and statistical view of the results (trying to strip out the quality of the car by comparing drivers to their teammates) and comparing the results to our own subjective views of the best drivers.

 

My biggest problem with this particular research is the totally subjective decision to treat Michael Schumacher as two drivers while not doing the same to anyone else. It's not that I disagree with this on a subjective level, it's just that it takes away any level of objectivity from the exercise. Why just Schumacher? Why not Raikkonen, for example?

 

This isn't a minor point which only effects one driver - it casts huge ripples into the rest of the rankings. As they mention, it would put Rosberg up from 49th to 13th. This also means that Hamilton and Webber would get more credit for their performances against the 13th best driver on the list, which means that their own teammates would get a tiny bump as well - and the ripples would keep going out.

 

Meanwhile, if we split Raikkonen into two drivers then it would effect every driver who's ever been his teammate as well. Kimi-1 would presumably do quite a bit better in the rankings meaning a boost for Coulthard, Montoya, Heidfeld, etc. and anyone who's ever been a teammate of those drivers. At the same time Vettel, Alonso and Grosjean wouldn't get quite so much credit for their performances against Kimi-2 as they would have done for driving so well against the guy who lit up the early 2000s!

 

The decision to split Schumacher into two drivers seems on the surface to be a very good one which only has a positive impact on the table... But it opens a Pandora's Box of questions around why that sort of subjective judgement wasn't used on other drivers!

 

They certainly forgot to split Lauda into multiple drivers - that's why he has got such a low rating. Preumably his post-comeback form and 1972 season, when he got battered by Peterson, have hurt him here.

 

But it also shows the mathematical model or whatever they have been using, doesn't have any flexibility in terms of driver performance fluctuations.

Edited by sopa, 14 April 2016 - 10:09.

[kraduk]

I quite like the idea of taking a fully objective and statistical view of the results (trying to strip out the quality of the car by comparing drivers to their teammates) and comparing the results to our own subjective views of the best drivers.

 

My biggest problem with this particular research is the totally subjective decision to treat Michael Schumacher as two drivers while not doing the same to anyone else. It's not that I disagree with this on a subjective level, it's just that it takes away any level of objectivity from the exercise. Why just Schumacher? Why not Raikkonen, for example?

 

This isn't a minor point which only effects one driver - it casts huge ripples into the rest of the rankings. As they mention, it would put Rosberg up from 49th to 13th. This also means that Hamilton and Webber would get more credit for their performances against the 13th best driver on the list, which means that their own teammates would get a tiny bump as well - and the ripples would keep going out.

 

Meanwhile, if we split Raikkonen into two drivers then it would effect every driver who's ever been his teammate as well. Kimi-1 would presumably do quite a bit better in the rankings meaning a boost for Coulthard, Montoya, Heidfeld, etc. and anyone who's ever been a teammate of those drivers. At the same time Vettel, Alonso and Grosjean wouldn't get quite so much credit for their performances against Kimi-2 as they would have done for driving so well against the guy who lit up the early 2000s!

 

The decision to split Schumacher into two drivers seems on the surface to be a very good one which only has a positive impact on the table... But it opens a Pandora's Box of questions around why that sort of subjective judgement wasn't used on other drivers!

 

 

Didnt Prost and Lauda have breaks as well? Then you could go silly with others like Grosean, and Kmag amongst others

[Imperial]

Tax payers?

Maybe Daily Mail readers!

[johnmhinds]

Just looking at the list again, are people like Louis Rosier only on the list because they didn't have a team mate for several years of their careers? 

Edited by johnmhinds, 14 April 2016 - 10:15.

[kraduk]

Tax payers?

 

But it really does seem like a case where they should have realized that its not nature that got it wrong its their model....

 

It looks like some kind of postgrad thesis, so at best only partially I suspect. Even if it was funded and useless, there is always the argument that often research for research sake it valuable. 

[Imperial]

The methodology used in this research smacks of that Bob Dylan/Watson advert.

And Dylan rightfully walks away at the end of that advert...

[Marklar]

The fun thing is that I totally forgot that Christian Fittipaldi was ever driving in F1, simply because he drove in F1 before I started watching and because he wasn't that impressive that it was memorable. I know him more as an sport cars driver, so I was first totally shocked when I saw him on 11th, between Vettel and Hamilton. Crazy!

 

 

Didnt Prost and Lauda have breaks as well? Then you could go silly with others like Grosean, and Kmag amongst others

Not even that: For instance a 20 years old Rosberg is different to judge than a 30 years old Rosberg. Drivers are always in different stages of their career (and on top of that have also good an bad seasons), yet all these models are rating the drivers in every stage of their career equally. That's why these models are ultimately failing IMO.

[Fontainebleau]

Given the data they have used, what I understand is that what this research shows is not "the best F1 drivers" in absolute terms, but "the best F1 drivers in terms of value-added to the results achieved by the car/driver combo". If we assume that the success is to be attributed partly to the car and partly to the driver, this is the rank of which drivers had a bigger % input in those results. Now, that does not necessarily mean that they were the absolute best: a better driver could have had a lower % impact if he had a better car, and still in absolute terms be a better driver.

 

Also, if I am not mistaken the data used ranks the drivers based just on their performance on track; but does not attach any value to their potential contributions to the development of their cars (as that development would go towards a greater % input being assigned to the car). That would be very unfair towards drivers such as, for example, Niki Lauda.

 

But I am running a relatively high fever, and all of the above could be well off target...

[Kristian]

These 'studies' just show the folly of attributing mathematics and models to something that is very human. 

 

The fact that Schumacher's career had to be split into two shows the flaws here - basically going on his second stint he wouldn't have even made the top 100, yet he was one of the most talented people to ever sit in an F1 car. It shows that longevity might be a disadvantage - would Graham Hill have been higher if he hadn't trundled on for years for the fun of it? 

 

Unltimately these exercises feel very empty to me. 

[sportingcp]

According to my study, Markus Winkelhock is the best driver of all time.

[Grayson]

They certainly forgot to split Lauda into multiple drivers - that's why he has got such a low rating. Preumably his post-comeback form and 1972 season, when he got battered by Peterson, have hurt him here.

 

But it also shows the mathematical model or whatever they have been using, doesn't have any flexibility in terms of driver performance fluctuations.

 

I think that it would be pretty much impossible for any mathematical model to look at rise or decline of a driver because of the small sample sizes, especially if those drivers spend quite a long time with the same teammates. They're using the driver comparisons to work as the yardstick of whether those teammates are any good, so they can't then use the comparison against those teammates to work out whether the driver is getting better or worse! It would be like trying to work out whether you're losing weight over the course of 10 years while also trying to work out whether your scales are getting less accurate as the decade goes on.

 

To use the Schumacher example, how can the model work out whether Schumacher was worse after his return or whether Rosberg would have been able to get those results against prime-Schumacher using only the objective results at the end of each race? Unlike us, the mathematical model can't watch trackside to see how Schumacher is taking corners or listen to an interview where Ross Brawn tells us what he thinks. Of course, in his case the researchers decided to intervene to stop the model from making what they thought would be a mistake...

 

All of this comes back to the basic point that everyone in this thread is making. The attempts to rank the drivers objectively can be extremely interesting but most of us doubt that they're as accurate as our subjective opinions.

[statman]

Given the data they have used, what I understand is that what this research shows is not "the best F1 drivers" in absolute terms, but "the best F1 drivers in terms of value-added to the results achieved by the car/driver combo". If we assume that the success is to be attributed partly to the car and partly to the driver, this is the rank of which drivers had a bigger % input in those results. Now, that does not necessarily mean that they were the absolute best: a better driver could have had a lower % impact if he had a better car, and still in absolute terms be a better driver.

 

Also, if I am not mistaken the data used ranks the drivers based just on their performance on track; but does not attach any value to their potential contributions to the development of their cars (as that development would go towards a greater % input being assigned to the car). That would be very unfair towards drivers such as, for example, Niki Lauda.

 

But I am running a relatively high fever, and all of the above could be well off target...

Despite the fever, a good analysis

[Amphicar]

If only this insight had been available in the early '90s - then one of the top F1 teams could have secured Christian Fittipaldi's services. Instead, they had to turn to the clearly inferior talents of the likes of Nigel Mansell, Mikka Hakkinen, Damon Hill, Gerhard Berger, Riccardo Patrese, Rubens Barrichello, David Coulthard, Heinz-Harald Frentzen, Eddie Irvine, Jean Alesi and Olivier Panis.

 

I am reminded of Peter Self's description of the Roskill Commision's cost-benefit analysis into the location of the third London airport - "nonsense on stilts".

[Oho]

Quote from the report:

 

"Lauda only performed notably well when racing for Ferrari (1974-1977) and his results dropped when racing for other, lower achieving teams."

 

The very same Lauda who actually won title while being paired with number two on their list Alain Prost. Instead of considering this and other rather similar results as a worrying signs of weakness in their model, they simply shrug it of as a surprise and comment how experts got it wrong. Oh incidentally half a minute with google and Wiki gives a pretty good idea why Lauda's results took a nose dive, its damn hard to get results when you dont finish races.

Edited by Oho, 14 April 2016 - 11:31.

[Lotus53B]

Lies, damn lies, and statistics.

 

I once went to a lecture on statistics at the University, and watched as the lecturer took one set of data, and then used it "imaginatively" to prove 2 mutually opposing hypotheses.  Then, when reading the Statistica manual, I found a test called "Tukey's honestly significant difference test", which kind of proves my original quote.

 

The report makes many subjective choices, and as the example of splitting Michael into two drivers shows clearly, this massively changes the outcomes, and, in my very humble, but to me highly significant, opinion, makes it bollocks.

[superden]

Not this old chestnut again.

[F1matt]

The fun thing is that I totally forgot that Christian Fittipaldi was ever driving in F1, simply because he drove in F1 before I started watching and because he wasn't that impressive that it was memorable. I know him more as an sport cars driver, so I was first totally shocked when I saw him on 11th, between Vettel and Hamilton. Crazy!

 

 

Not even that: For instance a 20 years old Rosberg is different to judge than a 30 years old Rosberg. Drivers are always in different stages of their career (and on top of that have also good an bad seasons), yet all these models are rating the drivers in every stage of their career equally. That's why these models are ultimately failing IMO.

 

 

I do remember Fittipaldi in F1 and he was never that impressive, I remember him in F3000 when he had a huge budget and beat Zanardi who was driving for the Ill baron Rampante team, possibly the best name for a team ever, the highlight of the season if my memory serves me correctly was Naspetti and the Forti team winning everything thanks to some special jungle juice fuel which burnt your eyes every time he came past.....

[sabjit]

How many times have we seen statistical models applied to make such lists? As a Maths student with a focus on statistics (I know student doesn't make me an expert, and in fact modelling is something I wont be onto until next year although I have read up on it) I know well enough a lot of the time you can fudge a result by applying different tests or models. If you really wanted a statistical model which, say, put Vettel as the best driver in history, I'm sure there could be a way of forming one and legitimately validating it.

 

Must admit I once did something similar in a formative paper I did where a hypothesis test wasn't giving me the result I would find most convenient and faced having to back track a fair bit. So I halved the significance level and came up with some airy fairy reason as to why I chose it, and voilà I got the result I wanted.

Edited by sabjit, 14 April 2016 - 12:02.

[HP]

Read the list til

 

Marc Surer 17th.

 

And Heidfeld before Kubica? (They were for a while in the same car)

 

Do I need to know more about this research? I certainly don't mind trying, but if results obviously don't match, then it's better to think about the used math formulas again.

 

It's probably not a coincidence that great drivers were not math geniuses and vice versa.

[Nemo1965]

Mmm

 

I like stats like these because obviously, they did not stop and say: 'Wait. Christian Fittipaldi and John Watson ahead of Niki Lauda? Let us change the numbers somewhat.'

 

However, I think that extrapolating these stats with the actual results would probably bring up a good list. I am not willing to believe that Christian Fittipaldi was a better driver than Lauda but I am willing to believe he was much better than he was given credit for.

[Vitesse2]

One word - with a big question mark.

 

Ascari.

[Oho]

One word - with a big question mark.

 

Ascari.

 

"Other surprises are the low ranking of champion drivers such as Niki Lauda (142nd) and Alberto Ascari
(76th)."

 

Only surprises? Not really reasons to pause and actually think about it?

Edited by Oho, 14 April 2016 - 12:13.

[Jimisgod]

Michael Schumacher treated as two drivers?

 

Michael and Ralf?

[sabjit]

I've got a better list based off "Statistics"

 

1. Michael Schumacher

2. Alain Prost

3. Lewis Hamilton

4. Sebastian Vettel

5. Ayrton Senna

6. Fernando Alonso

7. Nigel Mansell

8. Jackie Stewart

9. Jim Clark

10. Niki Lauda

 

As you may have noticed, that's the all time win list. But we are already at a list more agreeable than this one.

[secessionman]

Given the data they have used, what I understand is that what this research shows is not "the best F1 drivers" in absolute terms, but "the best F1 drivers in terms of value-added to the results achieved by the car/driver combo". If we assume that the success is to be attributed partly to the car and partly to the driver, this is the rank of which drivers had a bigger % input in those results. Now, that does not necessarily mean that they were the absolute best: a better driver could have had a lower % impact if he had a better car, and still in absolute terms be a better driver.

 

Also, if I am not mistaken the data used ranks the drivers based just on their performance on track; but does not attach any value to their potential contributions to the development of their cars (as that development would go towards a greater % input being assigned to the car). That would be very unfair towards drivers such as, for example, Niki Lauda.

 

But I am running a relatively high fever, and all of the above could be well off target...

 

Considering this, Alain Prost has done remarkably well to rank 2nd considering the relative strength of his team-mates over the years (Watson, Arnoux (2), Cheever, Lauda (2), Rosberg, Johansson, Senna (2), Mansell, Alesi, Hill).

 

Indeed, in 13 seasons only outscored twice, by a point to Watson in his debut 1980 season, and by 0.5 points to Lauda in 1984.

[Sheldon835]

Considering this, Alain Prost has done remarkably well to rank 2nd considering the relative strength of his team-mates over the years (Watson, Arnoux (2), Cheever, Lauda (2), Rosberg, Johansson, Senna (2), Mansell, Alesi, Hill).

 

Indeed, in 13 seasons only outscored twice, by a point to Watson in his debut 1980 season, and by 0.5 points to Lauda in 1984.

Prost was underrated during his career and continues to be even today. As you say, his achievements against the very strongest of team-mates tell us a whole lot.

[Oho]

Prost was underrated during his career and continues to be even today. As you say, his achievements against the very strongest of team-mates tell us a whole lot.

 

Underrated? If I am not mistaken he features in the top five on pretty much every list of top drivers, with the other usual suspects being Senna, Schumacher, Fangio and Clark... On this list he is 2nd despite being marginally beaten by Lauda at 142nd to championship.

[Collombin]

And Prost did avenge those 2 season defeats to Watson and Lauda, beating them in total over their full careers as teammates.

This is what sways me towards agreeing that Prost probably should be several places ahead of Marc Surer.

[HoldenRT]

These lists.. they can never win, not just with F1 but with most sports, or even other things like arts, music etc.. but it's brave of them to try.

[Archer]

Statistics, that science that says that if I'm unemployed and other person has 1 million bucks, we both have half million. Statistics can say whatever you wish to, you only have to change some details to do it.

 

Lauda and Hakkinen out of the 50 best ever, even having in mind that Lauda won with the same car a world championship ahead of Prost who is number 2 in the list, and and Christian Fitipaldi 11th ahead of Lewis Hamilton, Vettel out of the list with 4 world titles, but Raikkonen is inside the list with only 1 and being beated hard last season by Vettel. Ricciardo inside of the list and Hakkinen out of it.

 

This list is a joke, they are trolling everyone.