113 replies to this topic

[mariner]

The title  says it all, a paper by some UK academics which can be accessed in full here 

 

http://eprints.white.../WRRO_96995.pdf - copyright the authors 

 

It's very hard going and the maths is overall way beyond me but the general idea is by forcing the sum of the team performcnes to zero you extract the trure driver rankings over time.

 

I will let you find out who they reckon is the best ovor 1950 to 2014 but I do agree with them!

 

 

[La Sarthe]

Fascinating as a way of trying to assess performances statistically. Whilst I can't argue with the top few, although of course I would put them in a different order based on personal bias, like all such bits of work, it's as good as its assumptions. This explains the names at No.11 and No.29......  I did pick up that they mark down drivers with poor reliability records, which seems rather unfair and probably explains No.29. It should give this forum something to discuss over the next few days.

[pilota]

Any formula that can rate Christian Fittipaldi above Lewis Hamilton must be seriously flawed. Or am I missing something?

Nathan

[Tim Murray]

What a ludicrous waste of time and money.

[mariner]

I think the key discussion is in 3.3.3 where they reject the usual racing fan argument that one should adjust for car breakdowns to be fair.

 

Their arguments are those of mathematicians but what surprised me , as  Jim Clark fan, is that their  approach still left Clark high up despite his car failing so often when he was in the lead.

 

My own view on it is that you can find oddities  but the top  5 or 6 are what I would expect .Certainly Fangio won 5 titles in 4 different makes which shows his true class.

[Doug Nye]

Oh - good effort at passing the time indulging an interest....so good luck to them.  The result is tosh, of course.

 

Edited to delete "...albeit quite well constructed tosh".  

 

I had only taken a cursory glance.  Upon closer inspection this is a shining example of 'tosh in - tosh out' - and the sad thing appears to be that the study's authors didn't in the first place recognise their input and their chosen comparative parameters - as being such woefully inaccurate or inadequate tosh.  

 

DCN

Edited by Doug Nye, 05 March 2021 - 17:43.

[mariner]

Two things

 

Firstly - it stopped in the 2014 season so much of Hamilton's domination isn't included. That may explain his lower ranking.

 

Secondly, there is fault in the "teams vs driver" analysis for Brabham as he was both!

[ensign14]

We had this before, didn't we?  McClaren is a bit of a blooper and they somehow work out that Riccardo Patrese (1 win) is the rightful 1992 world champion ahead of Mansell (9 wins).  Who incidentally is not in the top 50 drivers of the WC era, unlike Marc Surer.

[opplock]

I seem to have missed the highlights of Christian Fittipaldi's F1 career. Perhaps we should respond by concocting a statistical method to prove that Francois Migault is the true 1972 WC.  

[uechtel]

Formula 1 (F1) is a sport of genuine global appeal. Established in 1950, F1...

Something that starts with a major fault already in the second sentence, what can you expect...

[blackmme]

I love things like this as I’m slightly in this field in a work capacity.

I admire their confidence because if I’d created a model to rank F1 driver performance that has Christian Fittipaldi in 11th and Clay Reggazoni in 45th but doesn’t feature Niki Lauda in the Top 50 my reaction would be to think “We’ve really missed something important here haven’t we?” but they looked at it and thought “Yep, looks about right”.

 

Regards Mike

[Glengavel]

What a ludicrous waste of time and money.

 

This paper, or F1 in general?

[john aston]

Well - today I learned that Marc Surer was a better driver , and by some margin , than Stirling Moss, inter alios . Complex formulae might add  spurious gravitas to the 'research'  but anybody who still believes in their accuracy when they produce such risibly crap results , knows nothing about motor sport. And possibly not very much about doing big sums either . . 

[Charlieman]

I remember reading about a mathematical paper years ago which explained that a bicycle wheel carries its loads primarily by compressing its spokes. Complete nonsense, of course, and easily disproved in a fifteen minute experiment. We expect a few glitches in every model but for this F1 paper it makes sense to go back to the blackboard.

[68targa]

I am sure that for die-hard 'modellers' it has been a good challenge, however personally I think they have spent too much time in lockdown and now need to get out  rather more. Don't know who funded them but they should have spent it on more serious matters.  I have my own list of all time highly talented drivers which will probably be different to everyone else and that's the way it should be.

Edited by 68targa, 05 March 2021 - 15:09.

[john winfield]

I am sure that for die-hard 'modellers' it has been a good challenge, however personally I think they have spent too much time in lockdown and now need to get out a rather more. Don't know who funded them but they should have spent it on more serious matters.  I have my own list of all time highly talented drivers which will probably be different to everyone else and that's the way it should be.

 

We can't even excuse them through Lockdown - the paper dates from 2016!

 

Dr. Andrew Bell contributes here to a 2016 article in University of Sheffield's newsletter, in which he acknowledges the 'surprising results' of Christian Fittipaldi's 11th place ranking, and Niki Lauda's non-appearance in the top 100. But how can intelligent academics bother publishing such a paper, highlighting the 'surprising results' (ie nonsense, for anyone with some knowledge of the sport) without some serious discussion, within the paper, as to where the statistical model fails, at least in part? 
I haven't analysed in detail their graphs and tables, but, presumably, some of the criteria must be shaky, to generate some of the odd results. Has, for example, the matter of car reliability been ignored, benefitting the 'plodder' at the expense of the 'racer'? The paper would be so much more interesting if it identified the likely causes of the spurious findings, particularly if Dr. Bell et al. intend to use the model to help understand issues in wider society.

 

https://www.sheffiel...r-ever-1.567358

[RobertE]

It seems to me that there are some people in life who really need something to do. Thank God this crew are not in charge of modelling epidemics. Oh, wait...

[2F-001]

I haven’t ploughed through all of the maths in great detail - and probably won’t as long as there is wet paint in need of observance. 

 

On one hand, it seems to be based on results (modified to account for various other factors the modellers deem significant) rather than the more nebulous notion of who’s best (which is what they claim to have identified).

On the other hand, I see that three multiple championship winners do not feature in the “Top 50”; I’ve never held a championship title to be the be-all-and-end-all, but you cannot dismiss drivers with that kind of record, without a near total lack of understanding of motor racing.

Somewhere along the line, a grasp of the original “problem”, if it existed, has been lost.

 

I can understanding using a subject of some personal interest as the source data for an exercise in mathematical modelling, but all that seems to be shown here is that underlying model is, at best, flawed, and possibly worthless.

 

John Aston - were we not shown something similar to this a while back on that other forum that we know of?

I think the results of that were even more ridiculous than this one. We concluded that the compliers had never stood and watched any of the drivers in action - and had possibly never actually been to a motor race.

 

(There is a word for this kind of thing, but I don’t want a strike against me for bad language.)

[john aston]

You are  right Tony - more utter tosh ! 

 

My prescription for escaping this madness has just been to watch the late Jim Pace lapping Road America in the Shadow DN4 - wonder how many of the formula boffins have even heard of Shadow ? 

[Tim Murray]

I do hope they can now turn their great minds (and computers) towards solving some of the other great mysteries of our time - the greatest-ever Epsom Derby winner, perhaps, or the best Eurovision Song Contest winner. We need to know!

[john winfield]

I do hope they can now turn their great minds (and computers) towards solving some of the other great mysteries of our time - the greatest-ever Epsom Derby winner, perhaps, or the best Eurovision Song Contest winner. We need to know!

 

This and so much more, Tim!  The paper was peer-reviewed, and published in The Journal of Quantitative Analysis in Sports, 'An Official Journal of the American Statistical Association'. Sadly, I am not a subscriber so cannot delve much further, although some articles and papers are available free of charge. 

 

Where to start? A Hierarchical Bayesian Model of Pitch Framing?  Rao-Blackwellizing field goal percentage?  No thanks, but I am tempted by the latest issue:  A Contextual Analysis of Crossing the Ball in Soccer is top of my list, followed closely by A Variance Gamma Model for Rugby Union Matches.  Essential reading surely?

 

https://www.degruyte...l/key/JQAS/html

[D-Type]

The ultimate test of a sports statistician would be to produce an algorithm that predicts which of next Saturday's 8 Football League matches will end in a draw.

[68targa]

Better still predict the winner of the 4.30 at Haydock Park next week !

[D28]

Motor Sport in their 1999 ranking of greatest ever drivers, mentioned:  " Never have statistics mislead so radically". They were commenting on Gilles Vllleneuve and his 2nd place to, who else, Tazio Nuvolari. In this array he is completely out of the ranking, though they do explain why he and other "rather overrated" drivers  like Mario Andretti are excluded. Fair enough, but they then elevate Louis Rosier to  number 19 place. He had 38 starts with a couple of podiums, but zero wins; as far as I can see he is the only one in that position, though Luigi Fagioli has the one shared win. These two anomalies deserve some explanation, which is not given. Ed (Marc Surer and Christian Fittipaldi also are in this group)

The model might have set a variable of at least 1 win for inclusion.

So it goes, just a normal outcome of the modeling exercise.

Edited by D28, 05 March 2021 - 19:10.

[uechtel]

Fair enough, but they then elevate Louis Rosier to  number 19 place. He had 38 starts with a couple of podiums, but zero wins;

Ahem... Belgian Grand Prix of 1949, as an independent over nobody else than Luigi Villoresi and Alberto Ascari, the works Ferrari team?

 

That doesn´t mean I regard him as one of the greatest drivers ever, but he did have his one moment.

 

In my opinion it is totally useless to make that hard cut in the middle of Grand Prix history and to regard drivers like Rosier, Farina or others as 'novices' just becausemore or less randomly  at some point the started to calculate championship points...

Edited by uechtel, 05 March 2021 - 15:58.

[D28]

Ahem... Belgian Grand Prix of 1949, as an independent over nobody else than Luigi Villoresi and Alberto Ascari, the works Ferrari team?

 

That doesn´t mean I regard him as one of the greatest drivers ever, but he did have his one moment.

 

In my opinion it is totally useless to make that hard cut in the middle of Grand Prix history and to regard drivers like Rosier, Farina or others as 'novices' just becausemore or less randomly  at some point the started to calculate championship points...

OK, but the authors specifically limited the years 1950-2014, and incorrectly label it all as F1, even 1952-53. 

[uechtel]

OK, but the authors specifically limited the years 1950-2014, and incorrectly label it all as F1, even 1952-53. 

Yes, I know. But this only demonstrates the single-mindedness of this kind of approach. Of course you can arbitrarily pick out a certain time period but that will always lead to inaccuracies at the borders, as you simply cut off the careers of some drivers drivers. For example I wonder how sky-high Emmerson Fittipaldi would be ranked if they had chosen for example 1974 as the limit...

 

 

I love things like this as I’m slightly in this field in a work capacity.

I admire their confidence because if I’d created a model to rank F1 driver performance that has Christian Fittipaldi in 11th and Clay Reggazoni in 45th but doesn’t feature Niki Lauda in the Top 50 my reaction would be to think “We’ve really missed something important here haven’t we?” but they looked at it and thought “Yep, looks about right”.

 

Regards Mike

 

I think you have addressed the key problem of the whole effort here: If you have to adjust the algorythms and parameters until you get the result that you would agree to, then you don´t have to do the calculations at all (because you know already the result in advance). But otherwise you have a result that is obviously wrong, so this means also there must be mistakes in the calculation and you never have an "objective" data set for a cross check verification to detect them. So no matter what you do the whole thing always ends in itself without any useable purpose (besides some practice time in mathematics for the authors perhaps).

 

Edited by uechtel, 05 March 2021 - 19:20.

[kayemod]

What a ludicrous waste of time and money.

 

Most previous attempts at assessing F1 driver abilities have been very easily dismissible, but there is one, Formula One - The Real Score, a work by Brian Harvey and published by Veloce in I think 2017, that is in my opinion worthy of serious recommendation. In the Author's own words, this is "A reassessment of driver ability based on an analysis of all F1 and equivalent race results since 1947". I haven't seen this work mentioned much on TNF, but my opinion of its erudition and value was shared by the late and much missed Mike Lawrence, a link to his Pitpass review is posted below, it's still there on their site.

 

https://www.pitpass..../The-Real-Score

 

Like anything of this kind, it is of course just opinion, but it's the result of a massive amount of research, It's well written and expressed, and I found very little to seriously disagree with, and the book is endlessly entertaining. Results obtained racing in categories other than F1 aren't considered, and if they were, Le Mans, Mille Miglia, Targa Florio etc, the ratings of many drivers would be altered very considerably, but as as a small taster, the top ratings in a list of fifty F1 drivers are as follows.

 

1    Juan Manuel Fangio

2    Jim Clark

3    Alberto Ascari

4    Jean-Pierre Wimille

5    Stirling Moss

6    Jackie Stewart

7    Ayrton Senna

8    Michael Schumacher

9    Lewis Hamilton

10  Alain Prost

 

Beyond these first ten, the ratings and positions of many other drivers in the list of fifty surprised me considerably, drivers that I would have rated as near equals to some of the first ten languish well down, though that just shows the near impossibility of any exercise of this kind. There's little surprise though that to name just two, Christian Fittipaldi and Marc Surer are not mentioned at all, computer algorithms didn't help them.

[john winfield]

Mrs. W. works on the practical/vocational side of academia and, quite recently, I studied part-time in the English School at the University of Nottingham. Only now do I realise that most academics are expected to publish a certain number of papers each year, in academically reputable journals. The universities demand it of them.

In the English School, professors and lecturers were regularly publishing articles or chapters on, for example, very specific aspects of Webster, Lawrence, Woolf or Joyce, producing learned work which might seem hopelessly obscure to the layman, but are of great interest to students of those authors. And the Nottingham academics were enthusiasts themselves, and they knew their stuff.

 

I think the problem with the 'Sheffield' paper is that (unlike Brian Harvey's work) the content and findings are so clearly secondary to the modelling process, and The Journal of Quantitative Analysis in Sports seems to be the perfect receptacle for academic statisticians who need to publish. I'm sure most of the journal's published authors are also enthusiasts, of statistical modelling techniques, but presumably not overly interested in the sport. And probably not many of the published papers enter the public domain, catching the eye of readers like ourselves.  I might be wrong, perhaps coaches in sports-orientated American colleges and universities do make use of the papers to improve performance, and maybe 'A Contextual Analysis of Crossing the Ball in Soccer' will prove a hit over here. But I can't see any point to the 'Sheffield' paper except for trying out a modelling process, and earning a tick in the 'articles published' column for the four academics concerned. The danger comes once the paper is read by the likes of pesky TNF members, who might expect the authors' combined brain-power to come up with something slightly more convincing!

[Doug Nye]

"...and The Journal of Quantitative Analysis in Sports seems to be the perfect receptacle for academic statisticians who need to publish."

 

Perhaps in this case a more fitting receptacle could be recommended.

 

DCN

[2F-001]

Indeed John, “Publish or Die” as a friend of mine in the medical science world put it.

 

The main use for this survey would seem to be for post-exercise analysis to understand where it went wrong.

 

The danger with things like this comes if the perpetrators actually imagine they’ve done a great job and go on to apply equally inappropriate or inadequate modelling to something altogether more serious. There seems to be a degree of post-rationalization in there.

 

One comment in there that seemed to undermine credibility concerned the results agreeing with the notion of Mario Andretti being “rather overrated by experts”. Still, what do I know?

Edited by 2F-001, 06 March 2021 - 18:17.

[AJCee]

One comment in there that seemed to undermine credibility concerned the results agreeing with the notion of Mario Andretti being “rather overrated by experts”. Still, what do I know?

Hear, hear.

The older I get the better I think Mario was.

[wheadon1985]

Complete and utter load of rubbish. They have Prost down as the actual 1984 champion on page 41! No wonder Lauda is ranked 142nd 🙄

[D28]

Complete and utter load of rubbish. They have Prost down as the actual 1984 champion on page 41! No wonder Lauda is ranked 142nd

Excellent catch! This shows what regard the authors have for data integrity, or proof reading, and it was peer reviewed. As John Winfield pointed out above, the academics are modelling enthusiasts, not racing enthusiasts. I believe they also take a perverse satisfaction from finding results different from what  "subjective" followers might expect. That could help explain why they use no variable to correct obvious outliers in the results. As mentioned above, a simple 1 win threshold would eliminate most of the outlier drivers mentioned in this forum. Instead they publish a table showing a driver with zero wins in 11th position, 36 places ahead of a 3 time WDC with 14 victories. I don't believe enthusiasm for motorsport is highly correlated among members of this group.

Edited by D28, 07 March 2021 - 02:52.

[Tim Murray]

In addition to the anomalies already mentioned, this one caught my eye. Table A10 (pp 41/42) shows a ‘Comparison between predictions of the champion and the actual champion, for years 1979-2014.’ For 1980 the model gives Reutemann as champion instead of Jones, even though Alan beat Carlos 7-0 in the races they both finished problem-free, and outqualified him 8-6, in the same make of car.

[Tom Glowacki]

I recently read a book; "Calling Bullshit: The Art of Skepticism in a Data-Driven World", by Carl T. Bergstrom and Jevin D. West that was a fascinating read.  It analyzes how otherwise statistically sound analyses can give perverted results, often by the use of faulty assumptions or values.  Looking at pages 23-24 of the paper, that starts with  "Part of the reason for our low positioning of Hunt . . ."  this becomes evident when you see what values and assumptions underlie the methodology:

 

"compared to both Phillips and Eichenberger and Stadelman is his high rate of retirement, and the relatively high penalty that we place on not finishing (compared to Phillips, for example, who does not include non-driver failures in his analysis)"

 

"The high performance (7th place) of Nico Rosberg in Phillips 2014 was, as Phillips suggests, a result of his partnership with an out-of-form world champion (Michael Schumacher), which artificially improved his result."

 

"In our analysis, when Schumacher is separated into two drivers, pre- and post retirement, Rosberg’s performance is less impressive and he is placed 46th."

 

(page 24) "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."

 

I underlined the applied methodology which skews the results, as the writers themselves acknowledge.  Make arbitrary or unfounded assumptions, over-value; or under-value some factor; or make different assumptions, and you get "garbage in, garbage out", even if the math is impeccable.  What we have here is mathematicians who are not racers, placing values on elements of which they have little understanding, and proceeding to crunch the numbers.

Edited by Tom Glowacki, 07 March 2021 - 17:07.

[10kDA]

Methodology? I wouldn't let these guys methodologically pick the numbers for Saturday's lottery drawing if I was buying the ticket with YOUR money.

[opplock]

 

I recently read a book; "Calling Bullshit: The Art of Skepticism in a Data-Driven World", by Carl T. Bergstrom and Jevin D. West that was a fascinating read.  It analyzes how otherwise statistically sound analyses can give perverted results, often by the use of faulty assumptions or values.  

 

This also has real world consequences. The insane notion that UK care homes were a low risk for Covid was driven by modelling. One of the academics responsible for said modelling admitted in a BBC documentary that they assumed care homes were a closed environment.  

 

Anyone brave enough to tell Alan Jones that Reutemann should have been 1980 champion? Count me out!  

Edited by opplock, 07 March 2021 - 19:39.

[Nick Planas]

Take the statement "Some things are unnecessarily analysed" and then simply remove the last four letters.

[arttidesco]

Not sure why university time and presumably other resources are being wasted on this, but I am beginning to detect why there maybe a general undercurrent of mistrust of university modelling of more pressing contemporanious issues. How anyone who failed to score a podium finish in three seasons can appear a couple of spots below his two time championship winning uncle beggars belief. On a scale of 1-10 my headache scores an 11, more medication please nurse 

[10kDA]

I seem to have missed the highlights of Christian Fittipaldi's F1 career. Perhaps we should respond by concocting a statistical method to prove that Francois Migault is the true 1972 WC.  

"Did we say 'Christian'?... WILSON! We meant Wilson!"

[DCapps]

As someone whose former day job was in the models and simulation world, spending a few decades there, plus a bit of time wandering about the hallowed groves of the Academe with the unicorns dwelling there as well, I think that I have probably read, listened to, reviewed, and dealt with more such articles and papers and presentations and whatnot such as this one than most of those here. Needless to point out, the vast, overwhelming majority of the models I dealt with were related to the military, more often than not as they applied to simulations or simulators. I also fully understand the inherent anti-intellectualism (see Hofstadter...) that resides here and elsewhere, not all of it unwarranted, of course. I did download the paper and take a look at it. Full disclosure: I must admit that as much as it bothers me, after 3/4's of a century certain cognitive skills do tend to atrophy and I had to nose about my files and even my lecture notes to get my math & stats skills almost back to where they once were. (Cheat sheets are a blessing...) In the unlikely event that I had been part of the peer review or as a referee for the article, I could see that the math for the model would work and, therefore, be given a thumbs up. Of course, this is the finished product so I have no idea as to what comments or suggestions were made during that process -- take my word for it, nothing escapes the process unscathed (having been on both ends of the process, I can attest to that). From what I bothered to do, it seems that that the "math" -- the model -- does what was intended. That said, as some have mentioned, I did note several places where poor editing along with perhaps poor syntax coupled with the usual circuitous thinking made my Referee instincts flare a bit. I did wonder how they got through the process since I caught them in either the first or second read-through, not a good sign. Plus, I kept getting the sense that some of their assumptions regarding their variables were not quite those that I might agree with or at least as stated or as implied. Given that there appears to be an attempt to standardize the race results across a span of time, 1950 to 2014, during which there were significant differences in the sport, I think that while the model will certainly crank out the results according to the math, that besides Fangio you have Rosier, Moss, Hawthorn, Bira, and Fagioli from that general era is interesting, if not a bit baffling, of course. No Ascari?

 

While finding this an interesting exercise in creating a model for an area that rarely attracts much interest for such things, I did find myself thinking that if I made this change or that change and perhaps did a re-look at the variables... Life is, of course, too short for me to worry about other people's models at this point in my life. I think that as is usually the case when the model spits out results that have most people going WTF?, then perhaps you need to rethink that model. Squishy stuff such as modeling who is the GOAT is, in my personal view, tends to be a stupendous -- but popular -- waste of time and effort. Had I been in for the Q&A if this paper had been presented as part of a panel at a conference, I think that I might have had much to say. Had I been the moderator for the panel, I KNOW that I would have had much to say...

 

So, no reason to get your knives out and be ugly and nasty to these poor members of the Academe. As always, while ad hominum broadsides might make you feel better, they tend not to accomplish much other than demonstrating that Hofstadter was correct. I was just as interested in why they chose to look at this as the model itself. I gave them some credit for daring to wade into dangerous waters if nothing else.

 

And, no, the professor is not taking questions. He is retired and has other things to do at the moment.

 

HDC

[Doug Nye]

A failure, however worthy the effort invested in producing it, still remains a failure...  That's one of the truths that losing small wars teaches us.  The devil is almost always in the detail.  As the UK pandemic modelling experience has demonstrated - at immense monetary and social cost.  

 

​The only saving grace in this case is that their core subject was something as irrelevant and unimportant as motor sport.

 

DCN

[Glengavel]

This also has real world consequences. The insane notion that UK care homes were a low risk for Covid was driven by modelling. One of the academics responsible for said modelling admitted in a BBC documentary that they assumed care homes were a closed environment. 

 

Modelling is also widely used in the financial industry. That could explain a lot.

 

In fact, I believe that there is a lot of 'artificial intelligence' systems actually managing share dealing, with little or no human involvement. That definitely could explain a lot. Not that humans are any more reliable than computers in this regard.

[john winfield]

Don, I am interested in your comments that the math 'would work' or 'does what was intended'. The math and model clearly have some merit as they generate results which are recognisable to followers of the sport. But it seems everybody, including the authors, agree that there are glaring anomalies in the final rankings. And not just minor, subjective preferences, but real clangers. As it seems to be the stated intention of the academics concerned to use the modelling process for other, more important matters, are you not surprised that they, their peers, or The Journal of Quantitative Analysis in Sports, don't show more interest in investigating why the anomalies have occurred? Perhaps I should contact Dr. Bell.


I'm sure not all on this forum are inherently anti-intellectual! Perhaps we could run a statistically sound poll...

Edited by john winfield, 08 March 2021 - 15:43.

[Charles E Taylor]

Hi

 

 

 

 

Perhaps a look at the wisdom of George Box might be useful.

 

 

https://en.wikipedia...odels_are_wrong

 

 

 

Here, the usefulness element might also be questionable.

 

 

 

Have fun and perhaps remember.

“Since all models are wrong the scientist must be alert to what is importantly wrong - It is inappropriate to be concerned about mice when there are tigers abroad”

 

 

 

 

Charlie.

[Charlieman]

John Winfield and Don Capps propose that this paper was an explorative exercise, to see how the maths worked out. John also suggests that push for publication may have motivated the authors.

 

I think a third motivation should be considered: a hobby paper. The Internet Movie Database started as independently created performer and film lists, evolving into the Cardiff Internet Movie Database, then into IMDb. I worked with blokes who typed up data for the first English football databases; I even helped write a few scripts to process their data. Even the most po-faced academics have non-academic interests.

[john winfield]

John Winfield and Don Capps propose that this paper was an explorative exercise, to see how the maths worked out. John also suggests that push for publication may have motivated the authors.

 

I think a third motivation should be considered: a hobby paper. The Internet Movie Database started as independently created performer and film lists, evolving into the Cardiff Internet Movie Database, then into IMDb. I worked with blokes who typed up data for the first English football databases; I even helped write a few scripts to process their data. Even the most po-faced academics have non-academic interests.

 

I take your point Charlie but White Rose Research is funded by the libraries of Sheffield, York and Leeds universities. I think its goals, and mission statement, suggest something more serious, even if some published articles have developed from hobby papers.

 

http://eprints.white...c.uk/about.html

https://whiterose.ac...ject-heading-3/

[Roger Clark]

Several people have commented on the low ranking of Alberto Ascari.  I've been thinking about this and I think it demonstrates a fundamental flaw in the model.

 

The model appears to assess the relative contribution of car and driver by comparing the performance of a driver with that of his team mates. There is little doubt that the Ferrari was the best car in 1952 and most of 1953; 1-2-3 finishes were not uncommon. This seems to count against Ascari in the model. Yet the fact that he consistently and comprehensively beat his team mates, including one past and one future world champion,  should enhance his ranking, not diminish it.

 

Other anomalies may arise.   I am second to none in my admiration for Jim Clark but there is no doubt that for much of his career, Team Lotus were not able to put two cars on the grid with an equal chance of winning.  How does the model deal with private entries and single car teams?  Was Moss in 60 and 61 compared with the works Lotus entries? How are the large numbers of works, semi-works and private entries in the 50s and 60s dealt with?

 

I suspect that the academic peer reviews of the paper were concerned with the mathematics of the model.  I wonder how much effort went into considering its appropriateness in the real world of motor racing.

[Michael Ferner]

Generally, I believe a study shouldn't be rejected because of the results it produces, although it's certainly very tempting to point out the many eyebrow-raising moments in this one! Instead, we should be looking at the methodology, which is a bit difficult because, frankly, the mathematics of this are way beyond me, and I'm generally not a klutz when it comes to numbers. But, apart from that, there are a few things to consider.

 

The stated aim of the paper is to find out "who are the best F1 drivers of all time, conditional on team performance", or in other words, trying to calculate team performance out of the race results. So, what does a "team" perform in motor racing? In simple terms, it provides the car, and prepares it for competition. That's already a pretty complex proposition, as some teams build (almost) all of the components of the cars themselves, others simply buy existing hardware and concentrate on the preparational aspect. I don't see where any of this is conceptionalized in this study, so it doesn't seem to have bothered the authors too much. In fact, the only "definition" of a team can be found on page 29:

 

Teams are defined based on the chassis-engine-constructor combination, unless a constructor changes the chassis or engine used mid-season, in which case the team is judged to continue as that team. Whilst this is problematic where a team changes the car for one driver and not another, this problem only affects a small minority of team-years.

 

So that in fact, it seems we're talking about constructors, not teams. Or, do we? Let's take a look at Figure A3 on page 44 listing the "Top 20 team.level residues":

 

1 Ferrari

2 McLaren

3 Mercedes

4 Red Bull

5 Benetton

6 Lotus (2010s)

7 Williams

8 Daimler-Benz AG

9 Brawn

10 Matra Intl.

11 Renault

12 John Player/Lotus

13 Marlboro-Texaco

14 Cooper

15 Tyrrell

16 Maserati

17 North American

18 BMW Sauber

19 Ligier

20 Porsche

 

That listing alone poses more questions than answers! In short, for an analysis with the declared intention of measuring the impact of "teams" on driver performance, I would have expected a better definition of terms, and a strict adherence to those terms. Of course, one could also argue that the inherent object of motor racing is the competition between cars, and to compute them out of the equation is counter productive in itself, but in an age of one-make racing series I know that I am fighting a losing battle, thank you.    ;)

 

The other thing I found remarkable was that the authors evidently considered Michael Schumacher's ranking in 8th to be too low, and Nico Rosberg's in 13th too high, so they "split" Schumacher into two drivers, a "pre-retirement" one and a "post-retirement" one. They didn't do that with other "un-retired" drivers, like Niki Lauda, Alain Prost or Alan Jones, or other drivers with big gaps in their careers, such as Peter Revson, Mike Hailwood or Jan Lammers, as a matter of fact. But apart from that, it highlights for me again the folly of using career averages in racing statistics: if Schumacher was the third best F1 driver of all time when he first retired in 2006, how can Clark, Senna, Piquet and Stewart have overtaken him by 2012 without having competed in any races at all during that time?