Who was the greatest F1 driver?

Arguments over whether Schumacher was greater than Fangio are not new.

I’m going to present something that is new: the results of a mathematical model designed to answer who was the greatest Formula 1 driver of all time.

You may disagree with any attempt to answer this question on principle, and that’s okay by me. I hope you will nevertheless find this to be an interesting approach.

The first attempt to rank drivers using a model was in 2009 by Eichenberger and Stadelmann (paper here and pdf here). Without giving too much away, my model agrees with theirs on several of the top ranked drivers. However, many of the previous model’s results don’t pass a sanity check. Ukyo Katayama is ranked ahead of Nigel Mansell, Erik Comas is ranked ahead of Ayrton Senna, Arturo Merzario is ranked ahead of Jack Brabham, Mark Blundell is ranked ahead of Niki Lauda, and Mike Hawthorn is ranked 5th best of all time. Understandably, the model didn’t get a warm reception in online forums.

Nobody seems to have returned to the question using a model since then, so I decided to be that fool who builds a new model to (hopefully) give more sensible results. Without getting too deep into details, there are a few important differences between my model and the previous model. First, my model is more inclusive — I analyzed 345 drivers across the history of the sport (only requiring that a driver had at least 3 races without mechanical DNFs in a single season during their career), whereas Eichenberger and Stadelmann restricted their analysis to the 124 drivers with at least 40 starts in their career (which excludes some big names, including Ascari and Farina). Also, my model’s performance metric is based on points scored, whereas their metric was based on finishing position. These differences turn out to be important.

You may not agree with every ranking that my model produces — I don’t either — but each ranking can be logically explained in terms of the underlying data. The model is built to predict how drivers would perform relative to one another if differences between teams were taken out of the equation. In many cases, what looks like a very surprising ranking turns out to be eminently sensible once you dig into the data. In these cases, the model helps to highlight inherent biases of subjective rankings. Some drivers are consistently overrated by experts and fans, while others are consistently underrated.

How can you possibly compare drivers between different eras?

Before we get to the model and its results, there is a reasonable objection to be raised. Formula 1 cars have evolved almost beyond recognition from 1950 to the present day. The cars of the early 1950s had manual gearboxes, no real aero devices, no seatbelts, up to 4.5-L naturally-aspirated engines, and no strict regulations on fuel. Precision powersliding was the name of the game and the cars raced on tracks with few safety barriers.

The demands today are completely different. Racing is safer but also faster paced, requiring incredible reflexes. Cornering technique is different, as are the tracks. David Coulthard remarked after driving the Lotus 25,

It was fascinating for me to see how cutting edge technology was at that time; with the 200 horsepower, dampers, but no down-force, and, with the changes in technology, it reminded me of a very powerful Formula Ford car which relies more on driver grit and using yourself to balance the car, very different to the aggressive, hostile environment of modern, twitchy, high RPM F1 cars with very powerful brakes and overall downforce.

It is therefore impossible to know how Fangio would perform in a modern Formula 1 car, were he to be magically transported to the present day. Similarly, it is impossible to know how Schumacher would perform, were he to be magically transported back to Fangio’s heyday. In that sense, there can be no direct comparisons made between different eras. We can’t hope to sensibly answer the question: Who would win out of Schumacher and Fangio racing the same car in the same era?

What we can do is make direct comparisons between drivers in the same era by comparing them to their teammates. Since drivers typically have many different teammates across their careers, we can obtain an estimate of how drivers rank relative to one another. We can then compare relative performances between drivers in different eras, where no direct comparison is possible. An overall ranking list can then be constructed.

The model (in brief)

It is difficult to directly compare Formula 1 drivers because they drive for different teams. Differences in performance between the best and worst teams are typically larger than the differences in performance between the best and worst drivers. If we want to compare drivers between teams, we need to somehow correct for these differences. The goal is to estimate how each year of the championship might have looked if everyone had been racing in an equally good car.

The premise of my model is to assume that there are two main contributors to a driver’s performance in the championship: their driving performance and the performance of their team(s). It is sensible to start with the simplest possible model, so I assume that these two factors do not interact, even though they may interact in the real world (e.g., some cars may better suit the driving styles of some drivers than others). For each driver in each team in each season, I define:

Performance = Driver Performance + Team Performance.

Interestingly, once the model is fitted to the data, team performances account for 61% of the variance in overall performances, while driver performances account for 39%. This is consistent with the common wisdom that team performance is a more important factor than driver performance in Formula 1.

As my measure for performance, I use a driver’s average scoring rate in each season. To calculate this, I first exclude any races where a driver failed to finish due to a mechanical problem or a problem outside of their control (e.g., a disqualification on technical grounds). The remaining races — those in which a driver finished or failed to finish due to a ‘driver failure’ (using the same definition as I used for driver-related DNFs in my previous post) — are called counting races.

The scoring system in Formula 1 has changed slightly across the years. To make fair comparisons between eras, it is necessary to use the same scoring system in all years. I therefore retroactively applied the 10-6-4-3-2-1 scoring system to all years, since this is a good compromise to the systems used throughout the history of the sport. To allow the model to differentiate between drivers who scored no points, I scored fractional points for lower positions, going down by a factor of 10 for every 5 places, so 6th scores 1 point, 11th scores 0.1 points, 16th scores 0.01 points, etc. This makes a negligible difference for front-running drivers, but separates the performances of backmarkers.

The scoring rate can potentially range from a minimum of 0 points per race to a maximum of 10 points per race. I therefore used a scoring function S, which ranges from 0 to 10, to convert performances to scoring rates:

Scoring Rate = S(Performance – Competition Effect).

A competition effect was included for each season that depends on the strength of the drivers/teams in that season. There are only so many points available per race, so it is harder to score points against a more competitive field and/or a larger field. The dependence of the competition effect on the drivers/teams in each year requires a complicated mathematical derivation, so I won’t describe it in detail here.

In brief, the model looks at the scoring rates of each driver and their teammates in each year. Using this, the model makes its best estimate of team performances and driver performances in each year from 1950 to the present. Although quite simple, the model does a good job of fitting the data — it accounts for 82% of the variance in the data. For a full mathematical description of the model, you can see the peer-reviewed paper here, in the Journal of Quantitative Analysis in Sports. For this blog post, I analyzed all data up to and including the 2014 British Grand Prix. Note that the rankings in the published paper go up to the end of 2013, so some are slightly different.

Using my model, I ranked all drivers on a single scale: equivalent performance in the place of Sebastian Vettel in the 2013 Red Bull. For reference, Vettel scored 8.33 points per race in 2013 (using the 10-6-4-3-2-1 system). A scoring rate >8.33 points per race is considered more impressive than Vettel’s 2013 performance by the model, while a scoring rate not as a prediction of how a driver from another era would actually perform in the 2013 Red Bull, since that is impossible to predict.

All models should be as simple as possible and all models have their limitations, which must be kept in mind when interpreting their results. This model is about the simplest you could propose to reasonably address the question at hand. There are some important limitations.

  • The model does not explicitly account for changes in driver performance across the career arc, so some drivers may benefit by comparison to teammates who are rookies or well past their peak form.
  • The model excludes mechanical DNFs, so it ignores the fact that some drivers (especially in the past) contributed to their own car failures through rough driving. This choice was made because it is difficult to distinguish the effects of rough driving from bad luck.
  • The model does not account for team orders, team politics, or preferable treatment of one driver, since it is impossible to know to what extent these affected race results.
  • The model does not attempt to attribute blame for crashes, since this would be highly subjective (and impossible for many crashes in older races).

The top 60

Below, I list the top 60 drivers. I chose 60 because it allowed me to discuss the sport’s most notable drivers, including all but three of the world champions (leaving out Giuseppe Farina in 85th, Mike Hawthorn in 86th, and Phil Hill in 136th). The ranking metric used is the peak performance achieved by each driver, averaged over a period of 3 consecutive years, after correcting for team and competition effects. This provides an estimate of a driver’s peak form. For each driver, the peak 3-year interval selected by the model is listed, as well as the average points per race (ppr”) after adjusting for team and competition effects.

The results are not particularly sensitive to the choice of 3 consecutive years; the top 5 drivers in the list appear in the exact same order based on peak performances over 4 or 5 consecutive years. Some drivers were not active for 3 consecutive years and therefore have no ranking. Notable absences are Mark Donohue (ranked 30th by 1-year peak), Jose Froilan Gonzalez (ranked 43rd of all time by 1-year peak), Peter Revson (ranked 53rd by 1-year peak), and Stefan Bellof (ranked 83rd by 1-year peak).

To help understand the rankings, I have included statistical comparisons to teammates. These include qualifying, race, and points statistics across rounds they raced together. For race statistics (e.g., driver X beat driver Y 3-2 in races), I refer only to races in which neither driver suffered a mechanical or other non-driver DNF. In cases where mechanical DNFs played an important role in the match up, I have mentioned it.

60. Jacques Laffite (1975-1977, 5.92 ppr)

Starting the list is Jacques Laffite, whose name is synonymous with the Ligier team, where he spent 9 of his 13 years in the sport. During his career, he edged out talented teammates including Patrick Depailler, Didier Pironi (ranked 122nd), and Rene Arnoux (ranked 112th). However, he was stretched beyond his limits by both Keke Rosberg and Eddie Cheever.

59. Romain Grosjean (2012-2014, 6.02 ppr)

Romain Grosjean arrived in Formula 1 in 2009 as a brief replacement for Nelson Piquet Jr. at Renault. He was quickly whack-a-mole-d by teammate Fernando Alonso. Without a seat for 2010, he returned to prove himself in junior series. He got a second chance with Lotus in 2012 and was fortunate to be retained by the team after a season filled with careless accidents. Across 2012-2013, he was soundly beaten by Kimi Raikkonen, but showed improved judgement and speed towards the end of 2013, matching Raikkonen’s race performances in the latter half of the season. This year, he has showed that he may deserve a shot in a better car, beating teammate Pastor Maldonado (ranked 81st) 8-1 in qualifying, 2-1 in races, and 8-0 in points.

58. Timo Glock (2008-2010, 6.03 ppr)

Glock had an unusual Formula 1 career. After a very successful career in junior series (130 points in 4 seasons, using this junior driver performance metric I previously posted), he was signed as test driver for Jordan in 2004. When regular driver Giorgio Pantano failed to impress, Glock made a four-race appearance, with 7th on debut ahead of teammate Nick Heidfeld in 8th.

Without a regular seat for 2005, Glock spent the year in Champ Car. In 2006 and 2007, he competed in GP2, winning the title on the latter attempt. In 2008, he was back in Formula 1, alongside the prodigiously fast Jarno Trulli at Toyota. Trulli dominated qualifying 14-4, but had only a small edge on Glock in races, finishing ahead 9-7 in races and 31-25 in points. In 2009, Trulli again outqualified Glock, this time 11-4, but Glock held the advantage 8-5 in races and 24-22.5 in points.

Glock then crashed heavily in qualifying for the 2009 Japanese Grand Prix, ending his season prematurely. With an uncertain future in Formula 1, Glock signed for Virgin in 2010. His career did not recover from this move — from 2010-2012 he raced for backmarkers Virgin and Marussia. Although he did not achieve meaningful results, he had the edge on each of his teammates. In 2010, he beat Lucas di Grassi 17-2 in qualifying and 4-3 in races. In 2011, he beat Jerome d’Ambrosio 14-5 in qualifying and 7-6 in races. In 2012, he beat Charles Pic 14-5 in qualifying and 7-5 in races.

57. Jack Brabham (1959-1961, 6.08 ppr)

The first big surprise is the model’s low ranking of Jack Brabham, who has been ranked between 12th and 18th by experts in Autosport and F1 Racing. Two factors account for the discrepancy.

First, Brabham’s legendary status is partly due to achievements as a racer-mechanic, including winning a championship in his own car. In terms of legacy, there is perhaps no greater driver in the sport’s history. However, my model corrects for team performances, thus discounting any contribution Brabham had to his own cars’ performances. His results are adjusted for having more competitive cars and fewer mechanical DNFs than others in his era.

Second, Brabham was outperformed by several teammates, including Roy Salvadori (no 3-year peak) in 1958, fellow Lotus-Climax driver Jim Clark in 1962, Dan Gurney from 1963 to 1965, Jochen Rindt in 1968, and Jacky Ickx in 1969.

56. Mika Hakkinen (1998-2000, 6.09 ppr)

Many fans will be unhappy to see Hakkinen ranked 56th, especially since he has been ranked as high as 11th by experts. My model’s lower ranking is corroborated by Eichenberger and Stadelmann’s analysis, which placed Hakkinen 39th.

Copyright: Bridgestone

Hakkinen spent most of his career at McLaren, partnered with David Coulthard (ranked 66th), meaning Hakkinen’s rating is largely based on their relative performances. During their 6 years together, Coulthard twice outscored Hakkinen (in 1997 and 2001). Overall, Hakkinen was quicker in qualifying, beating his teammate 68-31. However, in races the gap was small, despite McLaren’s preference for the Finn and occasional team orders in his favor. Hakkinen finished ahead 35-26 in races and 360-296 in points. This means Coulthard finished ahead 43% of the time and scored 82% of Hakkinen’s points. By comparison, Coulthard finished ahead of Kimi Raikkonen 27% of the time and scored 73% of Raikkonen’s points.

Hakkinen’s first regular teammate was Johnny Herbert (ranked 88th) from 1991-1992. Herbert beat Hakkinen 14-10 in qualifying, but Hakkinen was ahead 7-2 in races and 11-2 in points. Hakkinen then made a brief cameo for McLaren in 1993. Ayrton Senna beat him 2-1 in qualifying, 1-0 in races, and 20-4 in points. In 1994, Hakkinen comfortably beat Martin Brundle (ranked 73rd) 15-0 in qualifying, 5-1 in races, and 26-13 in points. He then beat Mark Blundell (ranked 99th) 12-1 in qualifying, 5-1 in races, and 15-10 in points.

1998 is rated Hakkinen’s best year by the model, but the model largely attributes his results to the dominant Newey-designed McLaren he drove. The model awards him 2nd in the championship that year, behind Michael Schumacher. The model’s team performance ratings (excluding mechanical DNFs) are shown below, with thick lines indicating time Newey worked for each team. Hakkinen’s error-prone 1999 season, in which he almost lost the title to Eddie Irvine (ranked 61st) is instead awarded to Heinz-Harald Frentzen.

55. Nelson Piquet (1982-1984, 6.10 ppr)

Nelson Piquet and Jack Brabham are rated considerably lower than any other triple world champions. The earlier analysis of Eichenberger and Stadelmann came to a similar conclusion, ranking Brabham 47th and Piquet 48th. The models contend that Brabham and Piquet are overrated by virtue of having three titles to their name. The human mind naturally wants to put the most successful drivers near the top of the list and is susceptible to selective perception of data that challenge this belief, as well as bandwagoning (most expert lists look similar). A mathematical model is free of these cognitive biases.

Piquet is one of several drivers from the 1980s ranked lower by the model than by experts. This is because the model finds the level of competition was relatively low during this era. The level of competition has generally increased from year to year. In 1950, the year with the lowest level, several of the entrants were previous Grand Prix racers from the 1920s-1940s.  Most of the top drivers were aged in their 30s, 40s, or 50s.

Competition reached a peak around the mid-1970s, then dropped slightly across the late 1970s and early 1980s, before rising again in the 1990s, reaching new heights after 2000. The most competitive year to date is found to be 2011. The drivers of the 1980s are thus penalized for racing in what the model judges a slightly less competitive environment. This is in large part due to the woeful reliability in the 1980s; when a driver did finish, they had fewer competitors to beat, making it easier to score more points per counting race. The model’s competition strength function accounts for this.

Piquet spent most of his career outperforming less competitive teammates. These included Hector Rebaque (ranked 160th out of 162 drivers with a 3-year peak), Teo Fabi (no 3-year peak), Marc Surer (ranked 114th), Satoru Najakima (ranked 139th), Alessandro Nannini (ranked 102nd), and Roberto Moreno (no 3-year peak). The model also accounts for the dominant cars Piquet drove during his championship years.

From 1986-1987, Piquet performed similarly to teammate Nigel Mansell. Overall, Piquet was behind 17-13 in qualifying and 11-8 in races, but ahead 145-124 in points. A further indirect comparison with Mansell is possible via their shared teammate Riccardo Patrese (ranked 76th), which works in Mansell’s favor. Overall, Piquet beat Patrese 19-11 in qualifying, 5-3 in races, and 79-34 in points, while Mansell beat Patrese 34-12 in qualifying, 20-5 in races, and 192-117 in points.

54. Olivier Panis (1995-1997, 6.18 ppr)

Panis looked set for great things before breaking both legs at Montreal 1997. He had bettered every teammate, including Eric Bernard (no 3-year peak), Martin Brundle (ranked 73rd), Aguri Suzuki (ranked 144th), Pedro Diniz (ranked 95th), and Shinji Nakano (no 3-year peak). He had also taken an unexpected victory in the 1996 Monaco Grand Prix.

On return, Panis struggled to achieve the same results, perhaps due to lingering injury, or perhaps because he now had stronger teammates in less competitive cars. He had no trouble with one-lap pace, losing the qualifying war just 17-15 to Jarno Trulli, but fell behind in races. From 2000-2002, he was outperformed by Jacques Villeneuve. He then joined Cristiano da Matta (no 3-year peak) at Toyota for 2003-2004. Panis outqualifyied da Matta 18-8, but was beaten 9-7 in races.

53. Mika Salo (1995-1997, 6.25 points per race)

Mika Salo achieved little success in Formula 1, spending his career in midfield teams. He was closely matched with Mika Hakkinen in British Formula 3, with the two winning 15 of the 17 races in 1990. Salo’s career was then stalled by a drink-driving charge that cost him a superlicense in 1990.

Salo’s talent was recognized early on — he was ranked 74th by a panel of experts in F1 Racing magazine in 1997, with Hakkinen ranked 58th. Hakkinen then took two titles while Salo battled in midfield. With hindsight bias, Salo has been largely forgotten. My model suggests that at their respective bests, Hakkinen and Salo would have been evenly matched in the same car.Salo’s record relative to his teammates indicates that he deserved a shot with a better team. From 1995-1996, he dominated Ukyo Katayama (ranked 121st) at Tyrrell, beating him 23-9 in qualifying, 13-2 in races, and 10-0 in points. In 1997, Jos Verstappen (ranked 135th) joined Tyrrell — Salo beat him 10-7 in qualifying, 6-0 in races, and 2-0 in points.

Salo partnered Pedro Diniz (ranked 95th) in 1998 and 2000. Overall, he beat Diniz 22-11 in qualifying, 13-3 in races, and 9-6 in points. Salo had no full-time drive for 1999 and was called to race for Ferrari after Michael Schumacher‘s injury. He was beaten by Eddie Irvine (ranked 61st) 4-2 in qualifying, 4-1 in races, and 28-10 in points, although on one occasion Salo surrendered a win to Irvine. Salo spent 2001 as a development driver for Toyota, before returning for a final season to partner Allan McNish (no 3-year peak). He beat McNish 15-2 in qualifying, 5-3 in races, and 2-0 in points.

52. Stefano Modena (1990-1992, 6.28 ppr)

Stefano Modena is viewed by many as a wasted talent. In 1987, he was Formula 3000 champion and was invited to race for Brabham at the Australian Grand Prix. He made an ignominious debut, retiring due to physical exhaustion. His career thereafter consisted of drives for unreliable midfield teams. He never made it to a competitive team — his case not helped by his temperamental nature. Formula 1 observers recognized his incredible talent, and many viewed him as a future champion, even comparing him to Senna. In 1990, Joe Saward wrote:

Stefano Modena will replace Jean Alesi at Tyrrell next season and there are many in Formula 1 who believe Stefano has just as much — if not more — talent than F1’s current rising star.[article]

After a year spent struggling to qualify the hopeless EuroBrun car, Modena was paired with the experienced Martin Brundle (ranked 73rd) at Brabham in 1989. The two drivers were very closely matched. Modena outqualified Brundle 9-7, while the two were equal 3-3 in races and 4-4 in points, Modena’s points courtesy of a brilliant 3rd place at Monaco.

The 1990 Brabham was no better, with Modena taking the team’s only points. In qualifying, David Brabham (no 3-year peak) was an average of 1.33 seconds behind Modena. In 1991, Modena’s Tyrrell was occasionally competitive, and he dragged it to 8th in the championship, comfortably outperforming Satoru Nakajima (ranked 139th). He scored an astonishing 2nd place at the Canadian Grand Prix when Mansell embarrassingly stalled waving to the crowd on the final lap.

In 1992, Modena raced alongside Mauricio Gugelmin (ranked 64th) for Jordan. Gugelmin had a small edge in qualifying, beating Modena 9-7, but it was 2-2 in races, and Modena dragged the Jordan to 6th in the final round, scoring the team’s only point that year. Modena was unable to find a drive for 1993, having burned too many bridges.

51. Gilles Villeneuve (1979-1981, 6.31 ppr)

It is difficult to distinguish Gilles Villeneuve the driver from Gilles Villeneuve the icon. Villeneuve is fondly remembered for his derring-do and legendary feats behind the wheel, including that famous 1979 wet qualifying session at Watkins Glen where he was 11 seconds quicker than anyone. If we take anecdotes over achievements, Villeneuve is surely one of the greatest of all time, and that’s where experts have typically rated him: between 10th and 13th.

With a mathematical model, we can critically examine Villeneuve’s results, taking sentimentality out of the equation. Villeneuve was a spectacular driver with uncanny car control and wet-weather flair, but he also made frequent driver errors. In 1978, he was beaten by the more experienced Carlos Reutemann 12-4 in qualifying, 6-6 in races, and 48-17 in points. He then fractionally outperformed Jody Scheckter in 1979, based on points per counting race, and significantly outperformed him in 1980 (see Scheckter’s entry for a full discussion). From 1981 to his fatal accident in 1982, he raced alongside Didier Pironi (ranked 118th) at Ferrari. Overall, Villeneuve was ahead 14-6 in qualifying, Pironi was ahead 4-3 in races, and Villeneuve was ahead 31-19 in points.

50. Rubens Barrichello (1996-1998, 6.31 ppr)

Rubens Barrichello’s early career was filled with promise. From 1993-1998, he outperformed teammates Thierry Boutsen (ranked 84th), Eddie Irvine (ranked 61st), Martin Brundle (ranked 73rd), Jan Magnussen (no 3-year peak), and Jos Verstappen (ranked 135th). In 1999, he was closely matched with Johnny Herbert (ranked 88th).

From 2000-2005, Barrichello went head-to-head with Michael Schumacher at Ferrari. While Barrichello undoubtedly suffered from team orders, he was comprehensively outperformed by Schumacher, scoring only 61% of his points.

Barrichello spent 2006-2009 alongside Jenson Button, racing for the disappointing Honda team that eventually flourished as Brawn in 2009. Barrichello stayed in close contention in qualifying, but was consistently outperformed by Button in races. In his final two years at Williams, Barrichello slightly outperformed the rookies Nico Hulkenberg (ranked 74th) and Pastor Maldonado (ranked 81st).

49. Mario Andretti (1977-1979, 6.35 ppr)

Andretti made an impressive debut appearance for Lotus at the 1968 US Grand Prix, taking pole position. He continued to make occasional appearances over the following years, including a victory for Ferrari in only his 11th start at the 1971 South African Grand Prix. However, it wasn’t until 1975 that Andretti committed to a full season.

From 1976-1977, Andretti was paired with Gunnar Nilsson (no 3-year peak) at Lotus, whom he outscored 69-31. For 1978, Lotus paired Andretti with Ronnie Peterson. Peterson was considered the quicker driver by many, yet Andretti came out ahead 11-3 in qualifying, 8-2 in races, and 64-51 in points, although Peterson twice appeared to obey team orders by not challenging Andretti for the lead. The Lotus car was extremely dominant in 1978. Once the model corrects for this, the championship is awarded to Niki Lauda, with Andretti demoted to 5th.

Lotus were less competitive in 1979 and Andretti was outscored by Carlos Reutemann 20-14, although he was fractionally ahead on points per counting race, due to worse luck with mechanical DNFs. In 1980, Andretti was joined by Elio de Angelis, who turned heads by outscoring his experienced teammate 13-1. After a difficult season with Alfa-Romeo in 1981, Andretti made only brief appearances for Ferrari and Williams in 1982 before closing the curtain on his Formula 1 career.

48. Ralf Schumacher (1999-2001, 6.35 ppr)

Ralf Schumacher lacked the consistency of his brother, but was unbeatably quick on his day. Across his career, he was matched with several strong drivers and performed well by comparison. In his rookie year, he was only slightly outperformed by Giancarlo Fisichella (in his second year). In 1998, he partnered ex-champion Damon Hill and was again only slightly outperformed; in fact, he could have outscored Hill if allowed to pass for the lead at Spa. In 1999, he dominated Alex Zanardi (no 3-year peak), putting a rapid end to Zanardi’s comeback.

In 2000, Ralf was paired with the precocious rookie Jenson Button and again came out on top. From 2001-2004, Williams paired Schumacher with Juan Pablo Montoya. The two drivers proved to be an excellent match, with Montoya having a small edge overall (see Montoya’s entry for more details). In 2005, Ralf moved to Toyota, where he spent 3 years alongside Jarno Trulli. Again, Schumacher was a very close match for his teammate, outscoring Trulli 70-66.

47. Jean Alesi (1991-1993, 6.39 ppr)

The fact that Alesi achieved only 1 win in his 201 starts is viewed as a grave injustice by many. In 1989, Alesi finished 4th on debut for Tyrrell and dominated teammate Jonathan Palmer (ranked 116th) across the season. His star seemed on the rise in 1990 when he twice finished 2nd — including his memorable duel with Senna for the lead at Phoenix — and dominated Satoru Nakajima (ranked 139th).

Tyrrell, Williams, and Ferrari all tried to sign Alesi for 1991 leading to a contract dispute. Ultimately, Alesi replaced Nigel Mansell at Ferrari, while Mansell moved to Williams. Mansell’s move was fortuitous, as Williams delivered a top car, while Ferrari’s performance dipped. Alesi was beaten by Alain Prost 13-2 in qualifying, 4-2 in races, and 34-21 in points. Ferrari’s performance slipped further in 1992, but Alesi impressed with consistent points, outscoring Ivan Capelli (ranked 72nd) 18-3. The model considers Alesi’s performance the best of any driver in 1992, underlining the fact that Alesi may have been champion had he gone to Williams.

From 1993-1997, Alesi partnered Gerhard Berger (ranked 67th), first at Ferrari, then at Benetton. The two formed one of the closest partnerships in Formula 1. Across their 77 races together, Alesi’s record vs. Berger was 42-35 in qualifying, 19-22 in races, and 151-126 in points.

The final years of Alesi’s career were spent racing for Sauber, Prost, and Jordan. Alesi outperformed Johnny Herbert (ranked 88th), Gaston Mazzacane (no 3-year peak), and Luciano Burti (no 3-year peak), but was closely matched with Pedro Diniz (ranked 95th) and talented rookie Nick Heidfeld.

46. Nigel Mansell (1990-1992, 6.42 ppr)

Mansell is much admired for his dogged racing and has been rated as high as the top 10 by experts, especially in polls with a pro-British bias. My model ranks him considerably lower, due to being significantly outperformed by two teammates — Alain Prost and Elio de Angelis — and driving extremely dominant cars in his most successful years.

Mansell began his career at Lotus in 1980, racing there alongside de Angelis to 1984. The only year in which Mansell outscored de Angelis was 1983, when de Angelis suffered an astonishing 12 mechanical DNFs in 15 rounds to Mansell’s 4. Overall, de Angelis outscored Mansell 76-38.

In 1985, Mansell moved to Williams to partner Keke Rosberg. The two drivers were well matched, with Rosberg ahead 9-7 in qualifying, Mansell ahead 5-2 in races, and Rosberg ahead 40-31 in points. From 1986-1987, Mansell raced with Nelson Piquet at Williams. Again, the drivers were closely matched, with Mansell ahead 17-13 in qualifying and 11-8 in races, but behind 145-124 in points.

Mansell was joined by Riccardo Patrese (ranked 76th) at Williams in 1988 and from 1991-1992. Overall, Mansell dominated Patrese, beating him 34-12 in qualifying, 20-5 in races, and 192-117 in points. 1989 and 1990 were spent at Ferrari, first alongside Gerhard Berger (ranked 67th), then alongside Alain Prost. Versus Berger it was 7-7 in qualifying, with Mansell ahead 2-1 in races and 38-15 in points. Versus Prost, it was 8-8 in qualifying, with Mansell behind 4-3 in races and 73-37 in points.

45. Jacky Ickx (1968-1970, 6.43 ppr)

Ickx was remarkable for simultaneously competing at the highest levels of single-seater and sportscar racing. During his Formula 1 career, he outperformed several of the highest rated drivers of the 1960s and 1970s, twice finishing runner-up for the world championship.

Ickx’s first full season was at Ferrari in 1968 alongside the more experienced Chris Amon. Ickx was behind 7-2 in qualifying, but ahead 4-1 in races and 27-10 in points, finishing an impressive 4th in the championship. In 1969, Ickx raced for the Brabham team alongside team founder Jack Brabham. Ickx equaled Brabham 4-4 in qualifying, and beat him 3-0 in races and 18-14 in points.

Ickx spent 1970-1973 racing for Ferrari with Mario Andretti, Clay Regazzoni (ranked 90th), and Arturo Merzario (ranked 107th). Versus Andretti, Ickx was ahead 8-3 in qualifying, behind 3-2 in races, and ahead 31-16 in points. Versus Regazzoni, Ickx was ahead 16-13 in qualifying, 14-3 in races, and 86-61 in points. Versus Merzario, Ickx was ahead 5-4 in qualifying, 4-2 in races, and 17-7 in points.

Cahier Archive: http://www.f1-photo.com/

From 1974-1975, Ickx raced alongside the brilliant Ronnie Peterson at Lotus. For the first time in his career, Ickx was comprehensively beaten by his teammate, losing 21-3 in qualifying, 7-3 in races, and 38-15 in points. In the following years, Ickx made only sporadic appearances for Williams, Ensign, and Ligier before bowing out of Formula 1.

44. John Surtees (1964-1966, 6.45 ppr)

Surtees is the only driver to win world championships in both Formula 1 and motorcycles. While still on his way to a fourth 500cc title in 1960, he started four races for Lotus, astonishing the Formula 1 world by finishing 2nd in only his second race.

In 1961, Surtees made a full transition to Formula 1, racing for Cooper-Climax alongside Roy Salvadori (no 3-year peak). Other Cooper-Climax drivers that year included Bruce McLaren and Jack Brabham. Surtees performed similary to Salvadori and Brabham, but worse than McLaren.

Copyright: Getty Images Sport

Surtees moved to Ferrari in 1963, racing alongside Lorenzo Bandini (ranked 152nd) until 1966. Surtees beat Bandini 21-3 in qualifying, 7-1 in races, and 66-47 in points. The model penalizes both drivers in 1964 for driving a dominant car — Surtees’s title that year is awarded to Jim Clark.

Interestingly, the model picks 1966 as Surtees’s stand-out year, awarding him the championship slightly ahead of Clark. Surtees began 1966 in strong fashion with Ferrari, but moved to Cooper-Maserati alongside Jochen Rindt when Ferrari strangely omitted him from their Le Mans team. Surtees beat Rindt 6-1 in qualifying, 1-1 in races, and 19-18 in points, despite 4 mechanical DNFs to Rindt’s 1.

Surtees was sole driver for Honda from 1967-1968, then outperformed teammate Jackie Oliver (ranked 147th) at BRM in 1969. Surtees transitioned from driver to team manager across 1970-1971, racing his last full season in 1971, outperforming Rolf Stommelen (ranked 140th).

43. Dan Gurney (1963-1965, 6.45 ppr)

Gurney never quite made a serious challenge for the WDC, but was deeply respected by his peers. His career began at Ferrari in 1959 alongside Tony Brooks (ranked 109th) and Phil Hill (ranked 136th). The points tally was Brooks 18, Hill 16, Gurney 13. In 1960, Gurney was slightly outperformed by Graham Hill and Jo Bonnier (ranked 104th) at BRM.

1961 saw an improvement in form, as Gurney dominated Bonnier at Porsche. He carried this momentum into 1962, again outperforming Bonnier. Across 1960-1962, he beat Bonnier 13-9 in qualifying, 6-2 in races, and 36-10 in points.

From 1963-1965, Gurney beat Jack Brabham‘s 19-7 in qualifying, 5-5 in races, and 54-34 in points. Gurney was also joined by the rookie Denny Hulme in 1965, beating him 5-0 in qualifying, 1-0 in races, and 13-5 in points.

Gurney spent 1966-1967 as sole driver at Eagle, with a crippling 14 mechanical DNFs in 19 starts. In the remaining 5 races, he finished 4 in the points, including a win at the 1967 Belgian Grand Prix. Thereafter, Gurney made only occasional race appearances.

42. Denny Hulme (1970-1972, 6.46 ppr)

Hulme was never especially quick over a single lap, but was a master of smooth and conservative racing, making him a quiet and extremely effective performer, in much the same mold as Jenson Button. This is reflected in his statistics: he took 8 wins and 9 fastest laps, but only 1 pole position. He has the fewest pole positions per win of all drivers to achieve 5 or more wins, as shown in the graph below (some notable drivers and outliers are labeled, with Hulme at the extreme left).

In his rookie year at Brabham in 1965, Hulme was outperformed by teammates Dan Gurney and Jack Brabham. He was again outperformed by Brabham in 1966, but then turned the tables on Brabham in 1967. Overall, Brabham beat Hulme 20-5 in qualifying, 9-3 in races, and 95-69 in points. The model accounts for both Brabham and Hulme driving a fast and reliable car in 1967, relegating Hulme to 5th in the championship and awarding the title to Jim Clark instead.

From 1968-1970, Hulme raced for the McLaren team alongside countryman Bruce McLaren. There was healthy competition, with Hulme ahead 20-5 in qualifying and 6-5 in races, but behind 54-51 in points. Following McLaren’s death, Hulme raced alongside Peter Gethin (ranked 156rth) from 1970-1971. Although Gethin took a win in 1971, Hulme comfortably outperformed him overall.

From 1972-1973, Hulme was paired with Peter Revson (no 3-year peak), who had a short but extremely promising career, before he died in a 1974 testing crash. Hulme and Revson were a close match, with Hulme ahead 14-9 in qualifying, but behind 9-8 in races and 61-53 in points. Hulme’s last full season was spent with Emerson Fittipaldi at McLaren in 1974. He was not able to keep pace with his teammate and retired at the end of the season.

41. Juan Pablo Montoya (2002-2004, 6.49 ppr)

Montoya enjoyed a relatively short but spectacular career in Formula 1, with a reputation for aggressive racing and a penchant for oversteer. In 2001, he was a surprise replacement for Jenson Button at Williams, partnering Ralf Schumacher from 2001-2004.

Montoya proved to be a good match for Ralf, outscoring him 194-173. In 2003, Montoya came close to the title, eventually finishing 3rd, with Williams-BMW having the best powerplant that year. After correcting for team effects, Montoya is rated 5th that year, slightly behind Fernando Alonso, Michael Schumacher, Heinz-Harald Frentzen, and Kimi Raikkonen.

In 2005, Montoya moved to McLaren. His two seasons there were trying, as he struggled to keep pace with Raikkonen, while failing to fit into the culture of the McLaren team. He was dropped midway through 2006 after crashing into the back of Raikkonen at the start of the US Grand Prix.

40. Patrick Depailler (1977-1979, 6.49 ppr)

Depailler was a no-nonsense racer who enjoyed taking risks and driving at (or slightly over) the limit. You can see him manhandling his Tyrrell around the streets of Long Beach in 1978 in this video.

Depailler spent most of his career with the declining Tyrrell team, achieving only 2 wins and a highest championship finish of 5th. This was a poor reflection of his talent. My model ranks him similarly to a 1997 expert poll by F1 Racing magazine that placed him 43rd.

From 1974-1976, Depailler raced alongside Jody Scheckter. Overall, Scheckter had the edge on Depailler, outscoring him 114-65, but Depailler showed improvement from year to year. In 1977, Depailler was joined by Ronnie Peterson at Tyrrell. He demonstrated his phenomenal pace by beating Peterson 9-8 in qualifying and 20-7 in points, although he finished behind 3-2 in races.

Depailler’s last full season was at Tyrrell in 1978. He dominated Didier Pironi (ranked 122nd), beating him 14-2 in qualifying, 6-2 in races, and 34-7 in points. In 1979, Depailler raced only the first 7 rounds for Ligier before breaking both legs in a hang-gliding accident. Over those 7 rounds, he was beaten by Jacques Laffite 4-3 in qualifying, 3-0 in races, and 24-22 in points. Shortly after his comeback with Alfa-Romeo in 1980, Depailler died in a testing crash.

39. Jacques Villeneuve (2000-2002, 6.50 ppr)

Before Lewis Hamilton, Jacques Villeneuve was arguably Formula 1’s most successful rookie. In his first year at Williams, he surprised Damon Hill, taking the championship fight to the last round. In 1997, Williams was still the best car, but not by the same margin as it had been in 1996. Meanwhile, Hill had been replaced by Heinz-Harald Frentzen, who was touted as the next Schumacher. Villeneuve turned opinion on its head by dominating Frentzen, outscoring him 81-42. Villeneuve won the championship, while Michael Schumacher with 78 points was excluded for Jerez. After correcting for team effects, my model rates Schumacher’s performance the best of 1997 by a significant margin, with Villeneuve 2nd. In 1998, Frentzen and Villeneuve were closely matched.

In 1999, Villeneuve moved to BAR, where he would spend 5 uncompetitive years. Villeneuve’s performance was thought by many to have dropped off during this fallow period. However, his performances alongside teammates indicate that he was still driving at a very high level. Careful observers of the time recognized how hard Villeneuve was still pushing:

Villeneuve never stinted in his efforts on the track. No matter how bad the car, he drove it with the manic determination and huge confidence and bravery that had been in evidence from his first days in F1, and you could not but admire him for that. — Andrew Benson.

From 1999-2000, Villeneuve was paired with Ricardo Zonta (no 3-year peak), who arrived in Formula 1 with a strong junior record, including championship titles in Brazilian Formula 3, South American Formula 3, Formula 3000, and the FIA GT Championship. Villeneuve beat Zonta 26-3 in qualifying, 10-2 in races, and 17-3 in points. From 2001-2002, Villeneuve was paired with Olivier Panis. This was a closer match, but Villeneuve again came out on top, 21-13 in qualifying, 8-5 in races, and 15-8 in points. The model rates 1997 as Villeneuve’s best single-year performance, but rates 2000-2002 his 3-year peak. After correcting for team effects, he is ranked 5th in the championship in both 2000 and 2002.

In 2003, Villeneuve was joined by Jenson Button at BAR. At first he clearly underrated his young teammate, comparing him to a boy-band member. Button beat Villeneuve 8-7 in qualifying, 3-2 in races, and 12-6 in points. After a 3-race cameo for Renault in 2004, Villeneuve returned for a full season with Sauber in 2005. He was beaten by Felipe Massa 13-6 in qualifying, 10-3 in races, and 11-9 in points. In 2006, Villeneuve was joined by Nick Heidfeld. Villeneuve held a 7-5 edge in qualifying, but was behind 6-2 in races and 13-7 in points when he was replaced by Robert Kubica.

38. Bruce McLaren (1963-1965, 6.61 ppr)

The eponymous founder of the McLaren team was teammate to 4 world champions during his career: Jack Brabham, Phil Hill, Jochen Rindt, and Denny Hulme. McLaren’s performances relative to each of them demonstrate his quality.

Champ #1: McLaren started alongside Jack Brabham at Cooper from 1958-1961. Overall, he was slightly outperformed by his more experienced teammate, beaten 22-3 in qualifying, 10-5 in races, and 75-66.5 in points.

Champ #2: In 1964, McLaren raced with Phil Hill at Cooper. Hill is the model’s lowest ranked champion: 136th of all time. McLaren beat Hill 8-1 in qualifying, 2-0 in races, and 7-1 in points.

Champ #3: In McLaren’s last year with Cooper in 1965, he raced alongside the young Jochen Rindt, who was in his first full season. McLaren beat Rindt 8-2 in qualifying, 2-0 in races, and 10-4 in points.

Champ #4: After forming the McLaren team, McLaren raced alongside Denny Hulme from 1968-1970. McLaren was beaten 20-5 in qualifying and 6-5 in races, but outscored Hulme 54-51.

37. Eddie Cheever (1981-1983, 6.62 ppr)

Although he never won a race, Eddie Cheever is ranked the highest American driver by the model, slightly ahead of Dan Gurney and Mario Andretti. Besides 1983, Cheever spent his career in midfield and backmarker teams. After a year driving alone for backmarkers Osella, Cheever established himself by outperforming Michele Alboreto (ranked 79th) in 1981 and Jacques Laffite in 1982.

In 1983, he faced Alain Prost at Renault and was beaten 13-2 in qualifying, 4-1 in races, and 57-22 in points. The points difference was exacerbated by Cheever suffering 8 mechanical DNFs to Prost’s 2. Cheever slumped at Alfa-Romeo from 1984-1985, being outperformed by Riccardo Patrese (ranked 76th) 20-12 in qualifying, 4-1 in races, and 8-3 in points. With no full-time drive for 1986, Cheever returned with Arrows alongside Derek Warwick (ranked 68th) from 1987-1989. Overall, the two drivers were closely matched, with Warwick ahead 34-12 in qualifying, Cheever ahead 10-9 in races, and Warwick ahead 27-20 in points.

36. Graham Hill (1964-1966, 6.68 ppr)

Graham Hill has been rated between 17th and 21st by experts, whereas my model rates him lower, accounting for the fact that he drove very dominant cars in his championship years.

Hill consistently outperformed his teammate Richie Ginther (ranked 138th) at BRM from 1962-1964. However, he was outperformed by several teammates during his career, including Innes Ireland (ranked 62nd) in 1959, Jo Bonnier (ranked 104th) in 1960, Tony Brooks (ranked 109th) in 1961, Jim Clark in 1967, Jochen Rindt in 1969, Tim Schenken (no 3-year peak) in 1971, and Carlos Reutemann in 1972. Additionally, Hill was outperformed by two drivers using the same chassis-engine combination in1970: Jochen Rindt and Emerson Fittipaldi.

35. Damon Hill (1994-1996, 6.69 ppr)

Ranked just ahead of his father is Damon Hill, who took an unlikely path to Formula 1, switching from bikes to cars in his early twenties, reaching Formula 1 at 31, and winning the title at 36. Hill spent most of his career at Williams, driving strong cars. Once team effects are removed, the 1996 championship is awarded to Michael Schumacher by a large margin, with Hill 2nd in both 1994 and 1996.

Hill stacked up well against his teammates. In his first full season, alongside Alain Prost, he was outscored 99-69, while suffering 4 mechanical DNFs to Prost’s 1. Leading the Williams team in the wake of Ayrton Senna‘s death, he partnered David Coulthard (ranked 66th) from 1994-1995. Hill beat Coulthard 17-8 in qualifying, 10-4 in races, and 131-63 in points.

In 1996, Hill beat Jacques Villeneuve 13-3 in qualifying, 6-6 in races, and 97-78 in points. Despite winning the championship, Hill was ousted by Williams, joining Arrows for 1997. In this car, which was 5 seconds off the pace at the first round, he showed his true mettle. At Hungary, he finished a spectacular 2nd, very nearly winning the race but for a hydraulic pump failure. He beat Pedro Diniz (ranked 95th) 14-3 in qualifying, 3-2 in races, and 7-2 in points. For 1997, Hill had offers from Ferrari, McLaren, and Benetton. Based on his performance for Arrows, the model predicts that he would have scored 42 points for Ferrari, 44 points for McLaren, and 46 points for Benetton.

In 1998, Hill moved to Jordan, partnering Ralf Schumacher. Hill was outqualified 10-6, but beat Ralf 4-3 in races and 20-14 in points. By 1999, Hill looked demotivated and at one point announced that he would quit mid-season. In his final race at Japan, he withdrew despite no mechanical problems. Across the season, he was utterly dominated by Heinz-Harald Frentzen, who demonstrated that the car was capable of 3rd in the championship. Hill was outscored 54-7, thus creating one of the most interesting driver triangles: Hill beat Villeneuve, Villeneuve beat Frentzen, Frentzen beat Hill (demonstrating the dangers of comparing any driver to just a single teammate).

34. Carlos Pace (1973-1975, 6.72 ppr)

Carlos Pace is often forgotten today, overshadowed by the successes of his Brazilian compatriots Emerson Fittipaldi, Nelson Piquet, and Ayrton Senna. However, he was highly rated by his contemporaries, and ranked 51st of all time by a poll of experts in F1 Racing magazine in 1997.

Copyright: LAT Photographic

In his first two seasons, Pace dominated teammates Henri Pescarolo (ranked 128th) and Mike Hailwood (ranked 94th). He started 1974 with the Surtees team, outperforming Jochen Mass (ranked 82nd), before moving to the Brabham team. He raced alongside Carlos Reutemann from 1974-1976, and was slightly outperformed overall by his teammate. In 1975, he took his only career win at Interlagos — the track now named in his honor.

1977 got off to a good start with 2nd at the first round. Unfortunately, Pace was killed after the third round in a light aircraft accident.

33. Jean-Pierre Beltoise (1972-1974, 6.79 ppr)

Based on his performances relative to teammates, Beltoise is identified by my model as a driver that has been significantly underrated by experts. Like Olivier Panis, Beltoise is best remembered for a single brilliant victory at Monaco in torrential rain.

Beginning with Matra, Beltoise was the team’s sole driver for most of 1968. He impressed with 2nd place in the wet Dutch Grand Prix in only his 8th start. In 1969, he was joined at Matra by Jackie Stewart and was dominated by his outstanding teammate 10-1 in qualifying, 7-1 in races, and 63-21 in points. The next strongest teammate that Beltoise faced was Chris Amon in 1971. Again he was beaten, this time 6-1 in qualifying, 3-2 in races, and 6-1 in points.

In all other seasons, Beltoise outperformed his teammates. He outscored Henri Pescarolo (ranked 128th) 16-8 in 1970; scored 9 points (with that famous Monaco win) in 1972 to the 4 of Howard Ganley (ranked 141st), 1 of Peter Gethin (ranked 156th), and 0 of Reine Wisell (no 3-year peak); outscored Niki Lauda 7-2 and Clay Regazzoni (ranked 90th) 6-2 in 1973; and scored 10 points in 1974 to the zero of Pescarolo and Francois Migault (no 3-year peak).

32. Giancarlo Fisichella (1999-2001, 6.79 ppr)

Fisichella was tipped for greatness early in his career, after matching his more experienced teammate Pedro Lamy (ranked 119th) on debut at Minardi, then outperforming Ralf Schumacher at Jordan in 1997, including a brilliant 2nd place in the very wet Belgian Grand Prix.

Over the next 3 years at Benetton, Fisichella raced alongside Alex Wurz (ranked 105th), beating him 36-13 in qualifying, 18-11 in races, and 47-22 in points. Fisichella then slightly outperformed rookie teammate Jenson Button in 2000, outscoring him 18-12.

Fisichella dominated his rookie teammates Takuma Sato (ranked 127th) in 2002 and Ralph Firman (no 3-year peak) in 2003, and also took his first race victory in the unbelievable 2003 Brazilian Grand Prix. Then in 2004 he outscored Felipe Massa 22-12.

Fisichella’s great chance came in 2005, when he moved to Renault, who produced the year’s best car. He was dominated by Fernando Alonso, and was ultimately outscored 133-58. Alonso underlined his dominance in 2006, outscoring Fisichella 134-72. Fisichella continued to struggle at Renault in 2008, being outscored 30-21 by Heikki Kovalainen (ranked 78th).

In 2008 and 2009, Fisichella partnered Adrian Sutil (ranked 103rd) at Force India. Fisichella had a slight edge, beating Sutil 16-14 in qualifying, 12-8 in races, and 8-0 in points. Following Massa’s accident, Fisichella had a brief but disappointing appearance with Ferrari to end his career.

31. Keke Rosberg (1983-1985, 6.82 ppr)

Keke Rosberg won the championship under perhaps the strangest circumstances in Formula 1 history, winning only 1 race in the season. Across his career, he took 5 wins in total, meaning only the two lowest ranked champions have fewer wins: Mike Hawthorn (ranked 86th) and Phil Hill (ranked 136th). These statistics do little justice to Rosberg’s abilities.

Rosberg showed his potential with the Fittipaldi team in 1980, beating Emerson Fittipaldi 10-4 in qualifying, 3-2 in races, and 6-5 in points. In 1981, the Fittipaldi car was an abomination, with both drivers struggling to qualify. Rosberg outqualified rookie teammate Chico Serra 14-0.

1982 could hardly have delivered a greater change in fortune, as Rosberg signed for Williams. It proved to be an extraordinarily volatile season. Ferrari drivers Gilles Villeneuve and Didier Pironi (ranked 122nd) were cut down by death and injury, while Alain Prost‘s Renault was fast but unreliable. Keke’s consistent results, allied with reliability, took him to an unlikely title. Across the year, he beat Derek Daly (ranked 132nd) 12-1 in qualifying, 9-1 in races, and 42-8 in points.

Copyright: Williams Grand Prix Engineering Ltd.

Rosberg continued at Williams with Jacques Laffite from 1983-1984, beating him 27-3 in qualifying, 9-2 in races, and 47.5-16 in points. He then partnered Nigel Mansell in 1985 and the two performed similarly. Rosberg was ahead 9-7 in qualifying, Mansell was ahead 5-2 in races, and Rosberg was ahead 40-31 in points, despite 7 mechanical DNFs to Mansell’s 5.

In his final year, Rosberg raced alongside Alain Prost at McLaren. The more calculating Prost dominated Rosberg 12-4 in qualifying, 5-1 in races, and 74-22 in total points. As Prost put it, Keke “didn’t have quite the necessary finesse to get the very best out of the turbocharged fuel consumption era of racing we were in at the time”.

30. Jarno Trulli (2003-2005, 6.87 ppr)

Trulli is renowned for his one-lap pace, beating every teammate in qualifying until his uninspired final stint with Lotus alongside Heikki Kovalainen (ranked 78th). Of all the drivers to have won a race in Formula 1, Trulli has the greatest number of poles per win, with 4 poles and 1 win.

In his first three years, Trulli outperformed Ukyo Katayama (ranked 121st), Shinji Nakano (no 3-year peak), and Olivier Panis. From 2000-2001, he raced alongside Heinz-Harald Frentzen. He came out ahead 18-9 in qualifying, but was beaten 7-4 in races and 17-15 in points. In 2002 at Renault, he beat Jenson Button 12-5 in qualifying and 4-2 in races, and was outscored by Button 14-9, although Trulli suffered 8 mechanical DNFs to Button’s 5.

Trulli stayed at Renault alongside new recruit Fernando Alonso from 2003-2004. In 2003, he was beaten 8-8 in qualifying, 7-2 in races, and 55-33 in points. In 2004, Trulli put the pressure back on Alonso, beating him 8-7 in qualifying and 6-4 in races, and finishing behind 50-46 in points. Bizarrely, Trulli was sacked by Briatore before the end of the season.

Trulli next moved to Toyota alongside Ralf Schumacher. Trulli’s qualifying pace was too much for Ralf, beating him 39-17. In race trim, the two drivers were closely matched, with Trulli ahead 22-16 in races and behind 70-66 in points. From 2008-2009, Trulli partnered Timo Glock. Again, Trulli was ahead in qualifying, this time 25-8, and again he was closely matched with his teammate in race results, behind Glock 15-14 in races and ahead 53.5-49 in points.

29. Felipe Massa (2008-2010, 6.87 ppr)

Massa began his career fast but crash-prone, before developing into one of the sport’s best drivers under the guidance of Michael Schumacher. In his debut year, Massa crashed out 5 times and was significantly outperformed by Nick Heidfeld. After a year testing, he returned alongside Giancarlo Fisichella. Massa showed signs of improvement, but was again outperformed by his teammate. In 2005, he raced alongside Jacques Villeneuve and outperformed his teammate for the first time.

In 2006, Massa moved to Ferrari, racing alongside Michael Schumacher. Massa was beaten 13-4 in qualifying, 12-3 in races, and 134-80 in points. Nevertheless, he showed significant improvement towards the end of the year and stayed at Ferrari to partner Kimi Raikkonen from 2007-2009. The advantage swung between the two drivers, with a very small advantage to Massa overall. He finished ahead 25-19 in qualifying, 18-16 in races, and 213-195 in points.

Massa’s career nearly ended at Hungary 2009 when he was struck by an errant spring, but he returned to race at Ferrari alongside Fernando Alonso from 2010-2013. Massa was dominated 57-19 in qualifying, 62-12 in races, and 1029-496 in points. It is unclear how much Massa’s performance suffered as a result of his injury; some drivers have fully recovered from severe head injuries (e.g., Hakkinen), while others have not (e.g., Wendlinger). Below are Massa’s results relative to Alonso in each half-season from 2010-2013. Massa’s relative performance on his immediate return was actually one of his strongest performances relative to Alonso. This is suggestive of Massa’s performances tailing off due to the competitive and political climate that developed, especially after the 2010 German Grand Prix.

Massa’s results relative to teammate Alonso’s in qualifying, races, and points, across each half of each season from 2010-2013.

28. Jody Scheckter (1974-1976, 6.97 ppr)

Scheckter started in Formula 1 with a bad reputation. There was no doubting his raw speed, but his driving was impetuous. On debut, he was running 3rd until he spun. He then looked set for victory in his third race, but crashed into Emerson Fittipaldi, prompting the Brazilian to call Scheckter a “menace to himself and everybody else”. At the next race, Scheckter caused a monumental crash at Silverstone, leading to the McLaren team resting him for 4 rounds. On return, he crashed again.

Witnessing the death of Francois Cevert first-hand had a profound effect on Scheckter: “From then on, all I was trying to do in Formula One was save my life.” In the next three years, he crashed only once.

At Tyrrell, Scheckter drove alongside Patrick Depailler, beating him 27-18 in qualifying, 19-10 in races, and 114-65 in points. From 1977-1978, Scheckter was sole entrant for the Wolf team, taking 2nd in the championship in 1977 with an impressive Postlethwaite-designed chassis.

Cahier Archive: http://www.f1-photo.com/

In 1979, Scheckter made a canny move to Ferrari to partner Gilles Villeneuve, replacing Carlos Reutemann. Ferrari delivered the best car and Scheckter duly took the championship, outscoring Villeneuve 60-53 (and 51-47 in points counting towards the championship), but points don’t tell the full story. Although Scheckter was ahead 8-7 in qualifying, Villeneuve was ahead 6-4 in races and suffered 4 mechanical DNFs to Scheckter’s 1. Villeneuve’s bad luck dropped him behind in the points tally, leading to team orders in Scheckter’s favor at the Italian Grand Prix. After correcting for team effects, the model awards the 1979 championship to neither Ferrari driver — instead selecting Alan Jones — dropping Villeneuve to 7th and Scheckter to 11th.

The 1980 season was an unmitigated disaster, as Ferrari dropped from world-beaters to bottom of midfield. Scheckter seemed completely uninspired; he was beaten by Villeneuve 13-1 in qualifying, 5-1 in races, and 6-2 in points. He retired at the end of the year.

27. Elio de Angelis (1980-1982, 6.99 ppr)

Elio de Angelis was much loved within the Formula 1 community, but never had the car to challenge for a title. Interestingly, my model awards him the 1982 championship and rates him above contemporaries Nigel Mansell, Nelson Piquet, Jody Scheckter, Mario Andretti, and Keke Rosberg.

de Angelis began at Shadow in 1979, racing alongside Jan Lammers (no 3-year peak). de Angelis scored all of the team’s 3 points and outqualified Lammers 9-5, but was behind 3-2 in races. In 1980, de Angelis joined Lotus to race alongside Mario Andretti. Against expectations, de Angelis outperformed his teammate, outqualifying him 8-6 and outscoring him 13-1, although Andretti was ahead 3-2 in races.

Late in 1980, Nigel Mansell debuted at Lotus, and he raced alongside de Angelis until 1984. de Angelis beat him 46-15 in qualifying, 12-7 in races, and 76-38 in points.

In 1985, de Angelis was paired with young phenom Ayrton Senna. Senna demolished de Angelis in qualifying 13-3, but held only a 5-3 edge in races and a 38-33 edge in points.

de Angelis died in a testing crash in 1986. He was the last driver to die in a Formula 1 car until Roland Ratzenberger.

26. Chris Amon (1970-1972, 7.05 ppr)

Chris Amon and the next driver in the list are ranked the two greatest drivers to never win a race. Amon’s career was hobbled by unreliable cars and poor luck, with mechanical DNFs in 40% of his race starts. Nonetheless, his talent showed through. In 1967, he dragged an uncompetitive Ferrari into 5th in the championship. In 1968, he showed great qualifying pace, taking 3 pole positions, but was outscored 27-10 by teammate Jacky Ickx. Reliability problems disrupted Amon’s 1969 campaign, with 5 failures in 6 starts; he nonetheless achieved Ferrari’s best race finish of 3rd that year.

Copyright: The Klemantaski Collection

Copyright: The Klemantaski Collection

In 1970, Amon drove a much more reliable March-Ford and outscored fellow March-Ford drivers Francois Cevert (ranked 65th), Jo Siffert (ranked 100th), and Ronnie Peterson. The only March-Ford driver to outscore him that year was Jackie Stewart, by just 2 points. Amon was equally impressive in 1971, outscoring Jean-Pierre Beltoise 6-1. His appearances subsequently dwindled until retirement in 1976, mostly appearing as a sole entrant for Matra, Tecno, Ensign, and the shortly-lived Amon team.

25. Nick Heidfeld (2005-2007, 7.07 ppr)

Nick Heidfeld is arguably the most underrated driver of the past decade. He was limited by the cars at his disposal, but showed his capabilities alongside 7 drivers from this top 60.

After a superlative junior career (221 points by my junior driver scoring metric), Heidfeld raced for Prost in 2000 alongside Jean Alesi. The car was uncompetitive and scored no points, but Heidfeld matched his more experienced teammate, finishing behind 9-7 in qualifying and equal 2-2 in races. In 2001, Heidfeld beat highly-rated rookie Kimi Raikkonen 10-7 in qualifying, 6-5 in races, and 12-9 in points, but Raikkonen suffered 5 mechanical DNFs to Heidfeld’s 1.

Setting the tone for the rest of his career, Heidfeld was overlooked by top teams in 2002, with McLaren taking Raikkonen. Heidfeld spent two more years at Sauber, first alongside rookie Felipe Massa, whom he beat 11-5 in qualifying, 8-4 in races, and 7-4 in points. He then raced alongside Heinz-Harald Frentzen. In qualifying, Heidfeld was ahead 10-7. In races, Heidfeld finished ahead 6-2, but Frentzen achieved the team’s best finishes, including a podium, outscoring Heidfeld 13-6.

After a year in the doldrums with Jordan in 2004, where he dominated rookie Giorgio Pantano (no 3-year peak), Heidfeld moved to Williams alongside Mark Webber (ranked 63rd). Webber was ahead 9-5 in qualifying but Heidfeld beat Webber 6-4 in races and 28-24 in points, despite suffering 3 mechanical DNFs to Webber’s 1.

In 2006, Heidfeld outperformed Jacques Villeneuve at BMW Sauber before being joined by Robert Kubica. The two remained together for 3 years and were extremely closely matched. Heidfeld outscored Kubica 150-137 points overall, but also had slightly better luck, with 4 mechanical DNFs to 6.

With BMW withdrawing from the sport, Heidfeld was left without a seat for 2010, narrowly losing a McLaren seat to Jenson Button. He regained a seat at Renault in 2011 under unfortunate circumstances, when he replaced the injured Kubica. Heidfeld was surprisingly outqualified by Vitaly Petrov (ranked 92nd) 7-3, but beat Petrov 5-4 in races and 34-32 in points. He was dropped by the team before the end of the season, ending his Formula 1 career.

24. Carlos Reutemann (1973-1975, 7.18 ppr)

Reutemann took his place in history with pole on debut — a feat only matched by Mario Andretti and Jacques Villeneuve. However, Reutemann is also remembered as a perennial bridesmaid, finishing the championship in the top 3 four times, but never claiming a title. After correcting for team effects, my model awards him the 1974 championship, just ahead of Emerson Fittipaldi.

During his early career with Brabham, Reutemann matched Graham Hill and outperformed Wilson Fittipaldi (no 3-year peak). From 1974-1976, he raced alongside Carlos Pace. Pace beat Reutemann 17-15 in qualifying, but Reutemann beat Pace 10-6 in races and 63-39 in points.

In 1977, Reutemann joined Niki Lauda at Ferrari. In qualifying, Lauda was ahead 8-7, but Reutemann was unable to handle Lauda’s race pace, being beaten 11-1 in races and 72-36 in points. For the last two races of 1977 and all of 1978, Reutemann was joined by the rookie Gilles Villeneuve. Reutemann saw off the challenge, beating Villeneuve 15-3 in qualifying, 6-6 in races, and 54-17 in points.

Reutemann made a poor decision in 1979, moving to the now declining Lotus, while Ferrari delivered the year’s best car. Reutemann was closely matched with Mario Andretti, finishing behind 8-7 in qualifying, behind 3-2 in races, and ahead 20-14 in points. Andretti had 10 mechanical DNFs to Reutemann’s 4, meaning Andretti scored fractionally more points per counting race.

Copyright: Williams/LAT

Reutemann made a better move in 1980, joining Williams just as they became a serious force. From 1980-1981, he was closely matched with Alan Jones (see Jones’s entry for a full description). He came painfully close to a championship in 1981, missing out by a single point after leading going into the final round. After just two races with Williams in 1982, he quit following a dispute with Frank Williams.

23. Emerson Fittipaldi (1974-1976, 7.19 ppr)

Like Damon Hill at Williams in the aftermath of Ayrton Senna‘s death in 1994, Fittipaldi galvanized the Lotus team around him following the tragic death of Jochen Rindt in 1970.

In 1971, Fittipaldi beat Reine Wisell (no 3-year peak) 5-2 in qualifying, 2-1 in races, and 12-9 in points. He then partnered David Walker (no 3-year peak) in what proved to be the most mismatched driver pairing in history — Fittipaldi took the championship in dominant fashion, winning 5 of 12 races, while Walker scored zero points, never finishing higher than 9th. In qualifying, Walker was between 1.3 seconds (Jarama) and 19.6 seconds (Nurburgring) slower than Fittipaldi, with a mean time difference of 5.0 seconds and a median time difference of 2.7 seconds!

The next year, Fittipaldi was joined by an equal talent: Ronnie Peterson. Fittipaldi finished ahead 55-52 in points, with 4 mechanical DNFs versus 6 for Peterson. Seeking more success, Fittipaldi moved to McLaren. He dominated his two teammates in 1974, outscoring Denny Hulme 55-20 and Mike Hailwood (ranked 94th) 37-12. The next year, he outscored Jochen Mass (ranked 82nd) 45-20.

Much like Jacques Villeneuve, Fittipaldi had a career in two acts, with great success in the first and disastrous failure establishing a new team in the second. Fittipaldi raced for his own Fittipaldi Automotive team from 1976-1980. He scored just 37 points across those 5 years — fewer than he scored in each year from 1972-1975. He was the team’s sole driver for most races. After a disappointing year in 1980 where he was outpaced by young teammate Keke Rosberg, Fittipaldi retired from Formula 1.

22. Alan Jones (1978-1980, 7.29 ppr)

In naming the best drivers of the 1980s, Alan Jones is frequently forgotten. My model rates him better than Nelson Piquet, Gilles Villeneuve, and Nigel Mansell. It also rates him the best Australian, ahead of Jack Brabham.

Jones’s early years were spent with the Hesketh, Hill, and Surtees teams, consistently dragging uncompetitive cars into unlikely points-scoring places. In 1977, he took his maiden win at Austria for Shadow in a storming drive from 14th on the grid. He trounced teammate Riccardo Patrese (ranked 76th) 6-3 in qualifying, 5-0 in races, and 13-1 in points.

Jones was sole driver for Williams in 1978, then raced with Clay Regazzoni (ranked 90th) in 1979. Jones won 4 races and would have challenged for the title but for 7 mechanical DNFs. The model awards Jones the 1979 and 1980 titles. Regazzoni suffered 2 mechanical DNFs yet was outscored 43-32.

Williams delivered a reliable car in 1980 and a strong line-up: Jones and Carlos Reutemann. Jones took the title comfortably. In 1981, however, the teammates battled fiercely, costing the title. Overall, Reutemann was ahead 16-13 in qualifying, and Jones was ahead 12-9 in races and 117-98 in total points.

Jones retired temporarily, but made a full return in 1986. He struggled for pace alongside Patrick Tambay (ranked 83rd), being outqualified 11-4, but beat Tambay 3-2 in races and 4-2 in points.

21. Robert Kubica (2008-2010, 7.29 ppr)

Kubica’s Formula 1 career was short but spectacular, ending prematurely with a rallying crash that left him severely injured. In late 2006, he was called up to replace fading star Jacques Villeneuve at BMW Sauber, alongside Nick Heidfeld. Although less experienced, Kubica proved to be an extremely close match for Heidfeld. Across their time together from 2006-2010, the results were 29-28 to Kubica in qualifying, 25-24 to Heidfeld in races, and 150-137 to Heidfeld in points.

In 2010, Kubica moved to the Renault team, alongside rookie Vitaly Petrov (ranked 92nd). Kubica far outperformed his car, taking an impressive 8th in the championship and outscoring Petrov 136-27. The model rates Kubica’s performance the 3rd best that year. Unfortunately, Kubica’s career ended before he could show his potential in a top team. My model’s ranking of Kubica suggests that he certainly had the ability to be champion, given the right car.

20. Jenson Button (2005-2007, 7.32 ppr)

In an era of sharp, dynamic driving styles, Button stayed true to smooth, classical lines, achieving great success. Of the five world champions currently in Formula 1, Button is often maligned as the worst of the group. My model rates him behind the four others, but not far behind. Statistics suggest that he is also the best wet-weather driver of his generation.

In 2000, Button was a surprise pick for Williams, after just two years in single-seaters. His first two years in Formula 1 were difficult. He was beaten by his more experienced teammate Ralf Schumacher 11-6 in qualifying, 6-3 in races, and 24-12 in points. Then he moved to Benetton and was beaten by Giancarlo Fisichella 13-4 in qualifying, 7-0 in races, and 18-12 in points. Button admits today that a year in Formula Ford and a year in Formula 3 was insufficient preparation.

For me when I arrived in 2000, I was nowhere near ready for Formula 1 … if I had the option to race for two more years and know I would get into a Formula 1 car after that, I would have taken that option.

By 2002, Button had matured. He outscored Jarno Trulli 14-9, although Trulli had 8 mechanical DNFs to Button’s 5. Then in 2003 he outscored Jacques Villeneuve 12-6. From the last race of 2003 to the end of 2005, Button raced alongside Takuma Sato (ranked 127th) at BAR. Sato was a fan-favorite, but he was utterly crushed by Button, 28-8 in qualifying, 20-1 in races, and 127-38 in points.

https://i2.wp.com/upload.wikimedia.org/wikipedia/commons/thumb/a/ab/Jenson_Button_2005_Canada_2.jpg/640px-Jenson_Button_2005_Canada_2.jpg

From 2006-2009, Button raced for Honda then Brawn with Rubens Barrichello. Barrichello was ahead 36-34 in qualifying, and Button was ahead 34-21 in races and 160-118 in points. Although 2009 was his title year, the model rates 2006, 2005, and 2011 his best single-year performances.

Despite winning the 2009 championship, Button was not expected by most to keep up with Lewis Hamilton at McLaren from 2010-2012. Over one lap, Hamilton indeed had a huge advantage, beating Button 44-14. However, Button’s race pace, consistency, and wet-weather skills saw him closely match Hamilton, finishing behind 27-20 in races and ahead 672-657 in points.

Button continues to perform to a very high level, comfortably outscoring Sergio Perez (ranked 71st) in 2013 and Kevin Magnussen (no 3-year peak) this year.

19. Ayrton Senna (1985-1987, 7.33 ppr)

Ayrton Senna was arguably the best driver over one lap in the history of the sport. He was outqualified by his teammate only 11% of the time. By comparison, Juan Manuel Fangio was outqualified 12% of the time, Alberto Ascari 27%, Jim Clark 26%, and Michael Schumacher 14% before 2010 (25% in total). Head-to-head, Senna outqualified every teammate: Cecotto 8-2, Johansson 2-0, de Angelis 13-3, Dumfries 16-0, Nakajima 16-0, Prost 28-4,  Berger 40-8, Andretti 13-0, Hakkinen 2-1, and Hill 3-0. Senna was also one of the greatest wet-weather drivers.

In many expert and fan polls, Senna has been ranked number 1. Like Gilles Villeneuve before him, Senna was killed at his peak and has since become a racing icon. My model places Senna in the top 20 of all time, but does not support ranking him above all others.

Senna’s ranking is lowered slightly by the fact that competition was relatively weak in his era, as explained above. His ranking is also lowered slightly by including all races in calculating scoring rates. In 1988, Senna outscored Alain Prost in the best 11 finishes that counted towards the championship, whereas Prost outscored Senna over all races. I explored other model specifications (e.g., different scoring functions) before selecting this one as best (based on parsimony and goodness of fit), and the rankings were robust; Senna consistently appeared between 10th and 25th.

Senna dominated four weak teammates: Johnny Cecotto (no 3-year peak), Johnny Dumfries (no 3-year peak), Satoru Nakajima (ranked 139th), and Michael Andretti (no 3-year peak). Considering his stronger teammates, he beat Gerhard Berger (ranked 67th) 19-7 in races and 224-135 in points, and beat Elio de Angelis 5-3 in races and 38-33 in points. Versus Alain Prost, he was ahead 14-9 in races and behind 186-154 in points (behind 163-150 in points that counted towards the championship), with 5 mechanical DNFs to Prost’s 3. As a result, Senna is ranked well above Berger, slightly above de Angelis, and close to Prost. The model considers Senna and Prost very evenly matched across their careers.

18. Ronnie Peterson (1971-1973, 7.38 ppr)

Rated just place above Senna is Ronnie Peterson. Both were one-lap specialists with phenomenal car control and both died at 34. You can watch Peterson in action in rare onboard footage here.

Peterson entered Formula 1 in 1970 as sole driver in a privateer March-Ford entry. In his second year, he was promoted to the March works team and astonished everyone by finishing 2nd in the championship, scoring 33 points to his teammates’ collective zero.

In 1972, Peterson dominated the rookie Niki Lauda 10-2 in qualifying, 6-1 in races, and 12-0 in points. Peterson was then closely matched with Emerson Fittipaldi at Lotus in 1973, beating him 11-4 in qualifying and equaling him 3-3 in races. Fittipaldi was ahead 55-52 in points, but Peterson suffered 6 mechanical DNFs to Fittipaldi’s 4.

From 1974-1975, Peterson dominated Jacky Ickx 21-3 in qualifying, 7-3 in races, and 38-15 in points. 1976 was spent in a March-Ford, beating teammates Hans-Joachim Stuck (ranked 75th) 13-2 in qualifying, 2-2 in races, and 10-5 in points, and Vittorio Brambilla (ranked 91st) 11-5 in qualifying, 5-1 in races, and 10-1 in points.

1977 was a difficult year for Peterson, racing the uncompetitive Tyrrell P34B with Patrick Depailler. Peterson finished behind 9-8 in qualifying, ahead 3-2 in races, and behind 20-7 in points. As a driver with great natural feel, Peterson seemed less comfortable in the higher-downforce cars of the late 1970s. In 1978, he returned to Lotus with Mario Andretti and was beaten 11-3 in qualifying, 8-2 in races, and 64-51 in points. The title was sealed for Andretti when Peterson died from injuries sustained in a crash at the 1978 Italian Grand Prix.

17. Heinz-Harald Frentzen (1998-2000, 7.39 ppr)

Frentzen’s appearance at 17th is surprising, but arises from the fact that Frentzen was outperformed only once. Unfortunately for Frentzen, this occurred in what should have been his breakthrough year. Failing to capitalize on a dominant car scuppered his chances of racing for a front-running team again, and he subsequently spent his career in midfield.

In his early years at Sauber, Frentzen upstaged Karl Wendlinger (ranked 108th), Andrea de Cesaris (ranked 87th), Jean-Christophe Boullion (no 3-year peak), and Johnny Herbert (ranked 88th). He seemed set for a title challenge in 1997 when he moved to Williams alongside Jacques Villeneuve. Instead, he wilted. Villeneuve beat him 13-4 in qualifying, 7-5 in races, and 81-42 in points. 1998 saw a partial revival for Frentzen, finishing the season behind 10-6 in qualifying, 7-7 in races, and 21-17 in points. Dropped by Williams, Frentzen moved to Jordan, where he obliterated the uninspired Damon Hill 14-2 in qualifying, 10-1 in races, and 54-7 in points. Hill was at this point just shy of 40 years old, so Frentzen’s ranking likely benefits from comparison to Hill near the end of his career arc. This can be explored by completely removing Hill’s 1999 results from the analysis, which drops Frentzen’s ranking to 57th and raises Hill’s ranking to 26th. This treatment is obviously unfair on Frentzen, given it ignores probably his strongest ever year, but it also suggests he may be overrated by the model based on this particular teammate comparison.

From 2000-2001, Frentzen raced at Jordan with Jarno Trulli. Frentzen was beaten 18-9 in qualifying, but finished ahead 7-4 in races and 17-15 in points. In 2002, he moved to Arrows, beating Enrique Bernoldi (no 3-year peak) 11-1 in qualifying, 2-1 in races, and 2-0 in points. In his final season, Frentzen returned to Sauber alongside Nick Heidfeld. Heidfeld was ahead 10-7 in qualifying, but Frentzen was ahead 6-2 in races, and 13-6 in points.


Comparing Frentzen with Mika Hakkinen gives a lesson on the importance of timing for success in Formula 1. Most fans and pundits rate Hakkinen well above Frentzen, whereas my model does the opposite. It is Frentzen’s failure to perform in his one year in a top car that captures attention. As a hypothetical, imagine that Hakkinen had performed to the exact same level in each of his years of racing, but also imagine that he spent only 1997 in a top car, like Frentzen — a year in which he was outperformed by Coulthard.


16. Alain Prost (1984-1986, 7.53 ppr)

Ranked at number 16, just ahead of his great rival Ayrton Senna, is Alain Prost. After correcting for team effects, he is awarded 6 titles: 1983-1986, 1988, and 1991.

Prost was not significantly outperformed by any teammates in his career, despite racing against some of the greatest drivers of all time. He was matched by John Watson in 1980, then dominated Rene Arnoux (ranked 112th) from 1981-1982 and Eddie Cheever in 1983.

In 1984, he was beaten to the title by Niki Lauda by just half a point. The following year, he massively outscored Lauda, but this was due to Lauda’s incredible reliability problems, with only 4 starts in which he did not have a mechanical DNF. Overall, Prost beat Lauda 29-2 in qualifying, 7-5 in races, and 137.5-86 in points. The two got along famously and enjoyed the fact that they could use almost identical set-ups. Prost learned from Lauda the value of settling for consistent points: “In ’84, if I had this philosophy, I would have been champion easy.”

In 1986, Prost dominated Keke Rosberg 12-4 in qualifying, 5-1 in races, and 74-22 in total points. He then beat Stefan Johannson (ranked 113th) 16-0 in qualifying, 7-2 in races, and 46-30 in points.

On Prost’s own advice, McLaren signed Ayrton Senna for 1988-1989. More has been written about their fractious relationship than any other in Formula 1. Suffice to say, the two drivers were very closely matched based on their race results.

Prost joined Ferrari in 1990 and was paired with Nigel Mansell. Not for the first time, Prost’s deceptive pace in race trim proved too much for his teammate, outscoring Mansell 73-37. It was a similar story in 1991, when he outscored Jean Alesi 34-21.

Prost was given the boot from Ferrari before the last round of 1991 for making critical remarks about the Ferrari chassis. After a sabbatical year in 1992, he returned with the dominant Williams team in 1993. There, he slightly outperformed Damon Hill before retiring.

15. John Watson (1979-1981, 7.54 ppr)

John Watson makes a rather outlandish appearance at 15th. In 1997, a panel of experts assembled by F1 Racing magazine rated Watson 61st. There are good reasons to think that the model has overrated Watson and that experts have underrated Watson — the truth likely lies somewhere in the middle.

Watson had mixed results in his early career, then blossomed at McLaren, matching the performances of two outstanding drivers: Alain Prost and Niki Lauda. However, Watson benefited from racing Prost as a rookie and racing Lauda when he was on return from a two-year break. These factors are not considered by the model, meaning Watson’s rating is probably inflated.

In 1977, Watson joined the works Brabham team alongside Hans-Joachim Stuck (ranked 75th), following the death of Carlos Pace. Watson was ahead 11-3 in qualifying and 3-2 in races, and behind 12-8 in points. In 1978, Watson partnered Niki Lauda and was dominated 10-6 in qualifying, 6-1 in races, and 44-25 in points, despite Lauda suffering 7 mechanical DNFs to Watson’s 3.

Copyright: McLaren Racing Ltd.

Copyright: McLaren Racing Ltd.

1979 saw Watson move to McLaren to begin the most successful phase of career. He first beat Patrick Tambay (ranked 83rd) 13-2 in qualifying, 4-1 in races, and 15-0 in points. He then raced the rookie Alain Prost, and was beaten 8-5 in qualifying, 1-0 in races, and 5-3 in points across their races together. Watson suffered 6 mechanical DNFs to Prost’s 3, and scored an additional 3 points in a race Prost missed with a broken wrist. Prost was replaced by the inexperienced Andrea de Cesaris (ranked 87th), whom Watson thrashed 14-1 in qualifying, 7-2 in races, and 27-1 in points.

Watson’s last two seasons with McLaren were spent alongside Niki Lauda, who had returned from his two-year break. It is difficult to know how much Lauda had declined or how much Watson had improved since 1978, but in 1982 the two drivers were closely matched. Lauda was ahead 10-5 in qualifying and 5-2 in races, but Watson was ahead 39-30 in points. In 1983, the car was less competitive and Lauda’s season was disrupted by 7 mechanical DNFs to Watson’s 3. Lauda was ahead 12-3 in qualifying, but Watson beat Lauda 4-2 in races and 22-12 in points.

14. Kimi Raikkonen (2003-2005, 7.56 ppr)

Raikkonen is an example of talent spotting done right. As discussed in a previous post, he spent just two years in Formula Renault 2.0 before Formula 1. On debut, Raikkonen was paired with the more experienced Nick Heidfeld. There was a real risk of Raikkonen’s career being rapidly extinguished, but he rose to the challenge, driving with incredible composure and equaling his teammate for points per counting race. McLaren were impressed and signed him to fill the seat left by Mika Hakkinen.

From 2002-2004, Raikkonen partnered David Coulthard (ranked 66th). He outperformed his teammate 32-19 in qualifying, 19-7 in races, and 160-116 in points. For 2005-2006, Coulthard was replaced by Juan Pablo Montoya, with Raikkonen again having the advantage, beating Montoya 19-8 in qualifying, 10-7 in races, and 145-86 in points.

Raikkonen then moved to Ferrari alongside Felipe Massa. In 2007, Raikkonen slightly outperformed his teammate, but Massa took the upper hand in 2008. In 2009, Massa held a slight edge prior to his near-fatal accident at the Hungarian Grand Prix, but Raikkonen had a superb second half to the season. Across their time together, the two drivers were very closely matched, with a marginal edge to Massa.

Following a temporary retirement, Raikkonen returned with Lotus for 2012-2013. He seemed to lose none of his speed, beating Romain Grosjean 20-16 in qualifying, 23-7 in races, and 380-210 in points. This year, he has rejoined Ferrari, alongside Fernando Alonso, providing a direct comparison between two of the era’s best drivers.

13. Stirling Moss (1959-1961, 7.65 ppr)

Copyright: Getty Images

Moss is often cited as the greatest driver to never win a championship. The model rates him the highest performing driver in each year from 1958-1961.

Despite a lack of early results in Formula 1, Moss was highly regarded due to victories in the 1952 Rally of Monaco, the 1954 12 hours of Sebring, and the 1955 Mille Miglia (chronicled here by co-driver Denis Jenkinson). He followed Juan Manuel Fangio to Mercedes in 1955 and proved a fast but compliant number 2 driver, beaten 6-2 in qualifying, 5-1 in races, and 58-28 in total points.

In 1956, Moss emerged from Fangio’s shadow to lead Maserati, beating Jean Behra (ranked 123rd) 4-3 in qualifying, 3-1 in races, and 28-22 in total points. Moss then partnered Tony Brooks (ranked 109th) at Vanwall from 1957-1958. He beat Brooks 10-4 in qualifying, 3-1 in races, and 57-35 in total points.

Moss drove a Cooper-Climax for most of 1959, winning the only 2 races in which his car did not fail. He then raced a Lotus-Climax from 1960-1961, outscoring other Lotus-Climax drivers Innes Ireland (ranked 62nd), John Surtees, and Jim ClarkMoss suffered a serious head injury in a crash trying to pass Graham Hill at Goodwood before the 1962 season, ending his career.

12. Lewis Hamilton (2007-2009, 7.67 ppr)

At the end of 2006, Fernando Alonso was top dog, having finally ended Michael Schumacher‘s reign. It therefore came as an almighty shock when Hamilton matched Alonso’s performance in his rookie year. Between them, it was 9-8 to Hamilton in qualifying, 10-7 to Alonso in races, and 109-109 in points. Had Hamilton’s career been cut short at the end of 2007, we would have been left to speculate on whether he was the greatest driver of all time. His results since then have demonstrated that he belongs among the greats, but has trouble maintaining top form.

Following Alonso’s acrimonious split with McLaren, Heikki Kovalainen (ranked 78th) joined Hamilton and was beaten 26-9 in qualifying, 18-10 in races, and 147-75 in points. Hamilton then partnered Jenson Button from 2010-2012. Hamilton demolished Button 44-14 in qualifying. However, he failed to convert this into a points advantage, partly through bad luck and partly through errors of judgment. Overall, the pair were closely matched, with Hamilton ahead 27-20 in races and Button ahead 672-657 in points. Hamilton retired 7 times due to crashes, whereas Button retired 3 times due to crashes. Hamilton’s mechanical DNFs (of which there were 6 to Button’s 5) tended to occur from higher positions, costing Hamilton 90 potential points to Button’s 40 potential points.

Hamilton joined Nico Rosberg at Mercedes in 2013. To date, the drivers have been extremely closely matched, with Hamilton ahead 15-13 in qualifying, 12-12 in races, and 350-336 in points.

11. Jochen Rindt (1968-1970, 7.90 ppr)

Rindt is most often remembered for his posthumous victory of the 1970 world championship. During his relatively short career, Rindt established himself as one of the greatest drivers of his era. In his rookie year, he was outperformed by teammate Bruce McLaren, but then began to match John Surtees in 1966. In the following years, he outperformed Pedro Rodriguez (ranked 89th) in 1967, Jack Brabham in 1968, and Graham Hill in 1969.

In 1970, Rindt performed on a whole new level. In fact, the model rates it the greatest single-year performance of all time, with an adjusted scoring rate of 9.64 points per race. Of the drivers running a Lotus-Ford that year, the points tally was: Rindt — 45 (10 rounds), Fittipaldi — 12 (6 rounds), Hill — 7 (12 rounds), Wisell — 4 (2 rounds), Soler-Roig — 0 (3 rounds).

Looking at the graph below of adjusted scoring rates, one wonders what could have been if Rindt had survived and decided to race on (although he had promised his wife Nina that he would quit after 1970).

10. Niki Lauda (1976-1978, 7.92 ppr)

Starting the top 10 is Niki Lauda. It is no surprise to find Lauda this far up this list. Across his mid-career, he dominated Clay Regazzoni (ranked 90th) from 1975-1976, Carlos Reutemann in 1977, and John Watson in 1978.

However, the performances at either end of his career were less impressive. In his rookie year, he was handily outperformed by Ronnie Peterson. He also only matched Regazzoni’s performances from 1973-1974, and faced problems against Jean-Pierre Beltoise in 1973. On his return in 1982, he was slightly outperformed by John Watson. He was then well matched with Alain Prost in 1984 and outperformed by Prost in 1985.

The model considers 1978 Lauda’s most impressive year — a year in which he took the championship fight to Lotus, despite his clear mechanical disadvantage. He is awarded both the 1977 and 1978 championships. Interestingly, once team effects are corrected for, the 1975 and 1976 championships are both awarded to James Hunt, who is discussed below.

9. Alberto Ascari (1951-1953, 7.94 ppr)

Alberto Ascari stands alonside Juan Manuel Fangio as one of the two greats of the early 1950s. His domination of the sport with Ferrari from 1952-1953 is unparalleled. He won 6 of the 7 races he started in 1952, with a mechanical failure in the Indy 500. His record of 9 consecutive victories stands to this day, having just been equaled 60 years later by Sebastian Vettel.

Even after adjusting for the very dominant Ferraris he drove, the model rates Ascari’s 1952 performance as the 8th best single-year performance of all time. In that year, where he was free from Fangio, he completely dominated several strong Ferrari teammates, including Giuseppe Farina (ranked 85th), whom he beat 53.5-27 in total points. Based on adjusted scoring rates, 1952 has the all-time largest margin between 1st and 2nd in the championship: 8.90 points per race for Ascari vs. 4.36 points per race for Hawthorn.

8. Sebastian Vettel (2011-2013, 7.96 ppr)

With Vettel now a four-time champion, it is difficult to dispute that he belongs among the all-time greats, but where he stands relative to others is still in flux, made all the more uncertain by his failure to beat Daniel Ricciardo (ranked 80th) this year, who is himself an unknown quantity. Based on results to date, the model ranks Vettel 8th of all-time and 3rd among currently active drivers.

Vettel’s ranking is primarily based on how he measured up against long-term teammate Mark Webber (ranked 63rd). Overall, Vettel beat Webber 73-23 in qualifying, 63-20 in races, and 1410-947.5 in points, i.e., Webber scored 67% of Vettel’s points. Webber is ranked among the likes of Eddie Irvine (62nd) and David Coulthard (66th). Vettel’s ranking is therefore consistent with the fact that his domination of Webber was less than Michael Schumacher‘s domination of Irvine (Irvine scored 48% of Schumacher’s points); and slightly greater than Kimi Raikkonen‘s domination of Coulthard (Coulthard scored 73% of Raikkonen’s points).

Vettel is not yet awarded any championships by the model. His scores are reduced once the model accounts for his dominant Red Bull cars, leaving him a close runner up to Alonso in 2011 and 2013. He receives a boost to his scoring rate in 2008 for his less competitive Toro Rosso car, raising him to 2nd in the championship. In 2009 and 2010, he is ranked 6th. In 2012, he is ranked 4th.

7. Nico Rosberg (2010-2012, 8.02 ppr)

Nico Rosberg at number 7 is one of the biggest surprises from this analysis, but follows computationally from him beating the higher ranked Michael Schumacher and matching the similarly ranked Lewis Hamilton. As a rookie, he was fractionally outperformed by Mark Webber (ranked 63rd). He then fractionally outperformed Alex Wurz (ranked 105th) in 2007 and crushed the relatively inexperienced Kazuki Nakajima (no 3-year peak) from 2008-2009.

At Mercedes, Rosberg dominated Schumacher in 2010, beating him 15-4 in qualifying, 13-3 in races, and 142-72 in points. The pair were then closer matched across 2011-2012, with Rosberg ahead 26-13 in qualifying, Schumacher ahead 17-13 in races, and Rosberg ahead 182-125 in points (mechanical DNFs cost Schumacher a potential 28 points, whereas Rosberg lost only 2 points this way).

It is natural to ask whether Schumacher had declined from 2006 to 2010, especially since his performance improved in 2011. If so, Rosberg would obviously benefit from this comparison with one of the sport’s greatest drivers, explaining his seemingly inflated rating. To test this, I tried excluding Schumacher’s results from 2010 when fitting the model. Rosberg’s ranking fell to 15th, with no change in Schumacher’s ranking. This highlights an important area for future improvement of the model: accounting for changes in ability across the career arc.

As a top driver near peak form, Hamilton provides a more reliable benchmark. The two drivers have been extremely closely matched, with Hamilton ahead 15-13 in qualifying, 12-12 in races, and 350-336 in points. This suggests Rosberg is indeed among the best current drivers, although his exact position among that group should of course still be considered in flux.

6. James Hunt (1973-1975, 8.23 ppr)

James Hunt at number 6? This is probably the most surprising result of all. Hunt was a great driver, but it’s unusual to hear him mentioned in the same breath as Jackie Stewart, Emerson Fittipaldi, and Niki Lauda, except in relation to his rivalry with Lauda. It is also jarring to see him ranked above four-time champions Alain Prost and Sebastian Vettel, given the tendency for subjective lists to assign driver rankings in close correlation with number of titles. Expert rankings have previously placed Hunt between 19th and 24th, usually behind Ronnie Peterson, Nelson Piquet, and Mario Andretti. My model places Hunt above all drivers of his era, except Stewart.

So what is going on here? Surprisingly, this is not a spurious result. It seems to instead be a case of a driver being significantly underrated by experts relative to his objective results, perhaps due to the fact that he secured only a single championship. Indeed, Eichenberger and Stadelmann support a high ranking for Hunt, putting him 14th (ahead of Peterson and Lauda, and just behind Fittipaldi and Rindt), despite their model penalizing DNFs more heavily than my own (Hunt DNFed often). With the exception of his two-race appearance in a March-Ford in 1974, Hunt outperformed every teammate. This is particularly impressive given how often he ended races by his own hands, which significantly harmed his scoring rate. The closest any teammate came was Patrick Tambay (ranked 83rd) in 1978 — both drivers scored 8 points, but Hunt had 3 mechanical DNFs to Tambay’s zero, as well as what must be one of the unluckiest seasons in history.

In 1973, Hunt’s rookie year, he dragged the uncompetitive March-Ford into 8th in the championship, despite only starting 7 of the 15 races. He finished twice on the podium, scoring 14 points to the 0 points of fellow March-Ford drivers Beuttler (ranked 143rd), Jarier (ranked 118th), Purley (no 3-year peak), Williamson (no 3-year peak), Pescarolo (ranked 128th), and Wisell (no 3-year peak), who collectively made 32 starts that year! The model considers this the best rookie driver performance of all time, just ahead of Lewis Hamilton‘s performance in 2007.

Hunt spent most of 1974 and 1975 as a sole-entrant for Hesketh, allowing no meaningful comparison with teammates. In 1976 and 1977, he was paired with Jochen Mass (ranked 82nd). This provides an excellent benchmark, since Mass raced with Fittipaldi near the height of the Brazilian’s powers in 1975. Fittipaldi beat Mass 11-3 in qualifying, 7-2 in races, and 45-20 in points. Hunt beat Mass 22-2 in qualifying, 14-5 in races, and 109-44 in points.

5. Juan Manuel Fangio (1955-1957, 8.43 ppr)

Although most fans never had a chance to see Fangio race (this is the closest most have come), there are still many who consider him the greatest Formula 1 driver, due to the extremely dangerous era in which he excelled. Putting the technical and safety challenges of different eras aside, the model ranks him the 5th best driver of all time.

Fangio was peerless among his teammates. Based on scoring rates in counting races, Fangio outperformed many of the greatest talents of his era, including Giuseppe Farina (ranked 85th) from 1950-1951, Stirling Moss in 1955, Peter Collins (ranked 98th) in 1956, Luigi Musso (ranked 120th) in 1956, and Jean Behra (ranked 123rd) in 1957.

The model awards Fangio the championship in 1950, 1951, and from 1954-1957, for a total of 6 world championships, one more than he historically scored and equal with Alain Prost. Only two drivers are awarded more championships by the model, and they appear next in the list.

4. Michael Schumacher (1994-1996, 8.49 ppr)

As the driver with the greatest cumulative statistics in the history of the sport, Schumacher belongs near the top of any greatest drivers list.

In his rookie year, Schumacher was closely matched with Nelson Piquet. Then from 1992-2006, Schumacher outperformed every teammate he had — most by incredible margins. His qualifying and racing record versus teammates is shown below.

Schumacher_record

The model awards Schumacher the championship each year from 1993-1998, 2000-2002, and 2004. His best single-year performance is 1994, a year in which he won 8 races, despite being disqualified or suspended from 4 of the 16 rounds. This is rated the 13th best single-year performance of all time. The gulf between Schumacher and his main championship competitors was significant, as seen below.

schumacher_graph

Schumacher was less competitive on his return to Formula 1, perhaps due to age, or perhaps due to him struggling with lower durability tyres and limited testing. In 2010 he particularly struggled. However, excluding 2010 has no effect on his overall ranking. Rosberg’s impressive performances alongside Lewis Hamilton so far suggest that Schumacher may have still been performing to a very high level.

3. Fernando Alonso (2012-2014, 8.87 ppr)

In creating greatest driver lists, it is always difficult to know where to rank present drivers. There is the issue of present bias, as well as the tendency to view some past drivers through rose-tinted glasses. A mathematical model circumvents these issues, although there is still greater uncertainty for drivers who have been racing for only a short time or against only one or two teammates.

Despite Sebastian Vettel‘s recent domination of the sport, most F1 pundits have no hesitation ranking Alonso as the best driver of the past decade. Alonso’s ability to drag less competitive cars to the front, coupled with his incredibly adaptable and much-copied driving style, has built him an impressive reputation. Bob McMurray, who had the pleasure of working with both Ayrton Senna and Alain Prost at McLaren, recently gave a supreme compliment:

If anybody ever combined the very best of both Ayrton Senna and Alain Prost then I think it can only be Alonso. [article]

Alonso has been an outstandingly consistent performer. By the model’s reckoning, if all cars were equal, Alonso would have won the championship in 2003, 2005, 2006, and each year from 2008-2013. With these 9 championships, he is only one short of Schumacher’s hypothetical haul of 10 championships. His six best years — 2014, 2009, 2006, 2005, 2012, and 2013– are all ranked within the top 20 best single-year performances, making him the driver with the most top 20 single-year performances. Remarkably, for a driver in his 13th year of Formula 1, Alonso is currently in his 3-year career peak.

To understand Alonso’s ranking, it is helpful to examine his racing record. There are only three drivers who have come close to matching Alonso for points per counting race in the same car: Lewis Hamilton, who equaled Alonso in 2007; Jarno Trulli, who came out fractionally behind in 2004; and Tarso Marques, who had slightly higher best and average finishing positions in Alonso’s rookie year in 2001 (but was dominated by Alonso in every other racing and qualifying metric).

Every other season, Alonso has completely crushed his teammate, including some very highly ranked drivers: Jarno Trulli in 2003; Giancarlo Fisichella from 2005-2006; Nelson Piquet Jr. (no 3-year peak) from 2008-2009; Romain Grosjean in 2009; and Felipe Massa from 2010-2013. His domination of Massa was particularly outstanding, with 1029-496 points, meaning Massa scored only 48% of Alonso’s points. By comparison, Barrichello scored 61% of Schumacher’s points (678-412) during their 6-year stint as teammates.

Prior to 2014, I used the model to generate estimated probability distributions for driver performances of Alonso, Kimi Raikkonen, and Massa, based on all of their performances up to 2013.

Based on these estimates, the model predicted that Raikkonen would be expected to outperform Massa in the same car with 84% likelihood, that Alonso would be expected to outperform Massa in the same car with 99.7% likelihood, and that Alonso would be expected to outperform Raikkonen in the same car with 94% likelihood.

2. Jackie Stewart (1969-1971, 9.16 ppr)

Today, Stewart is as much remembered for his contributions to driver safety as he is for his three world championships. In his time, he was a legend, setting a record for most race wins that would stand until 1987. This is particularly impressive when one remembers that the number of races per season increased from around 11 in his era to around 16 by the late 1970s. In 1973, he made the pragmatic decision to retire, while still near the height of his powers.

stewart

Expert polls universally rate Stewart among the top 6 drivers of all time. Looking at his career record it is clear that he deserves his position near the top. He was only once outperformed by a teammate based on points per counting race, and that was by the experienced Graham Hill in Stewart’s rookie year in 1965. The difference between them was marginal and Stewart turned the tables in 1966. In later years, Stewart dismantled several highly rated teammates, including Jean-Pierre Beltoise in 1969, Francois Cevert (rated 65th) from 1970-1973, and fellow March-Ford drivers Chris Amon, Mario Andretti, and Ronnie Peterson in 1970. His longest-term teammate, Cevert, was beaten an astonishing 37-5 in qualifying, 24-6 in races, and 190-83 in points.

The model considers Stewart the most deserving winner of the championship in 1968, 1969, 1971, and 1972, with close 2nd places to Jochen Rindt in 1970 and James Hunt in 1973.

1. Jim Clark (1962-1964, 9.24 ppr)

The driver picked as the best of all time by this mathematical model is Jim Clark. Jackie Stewart, the next highest ranked driver and a contemporary of Clark, described him as follows

Jimmy was the epitome of excellence. He was smooth and clean and hardly ever had an accident. He drove with precision. He was the best driver I ever drove against.[article]

Despite starting only 72 races (with 26 mechanical DNFs in those races), he still holds the record for most grand slams (pole, fastest lap, win, and every lap led) with 8 in total. Alberto Ascari and Michael Schumacher are next with only 5.

Once Clark hit his stride in 1962, no teammate got near him. The numbers he achieved during his purple patch from 1963 to 1965 are still difficult to believe. Of the 29 races he started over this period, he won 16, finishing an average of 58 seconds ahead of 2nd place! In each of the other 13 races, his car had mechanical problems. He amassed a total of 159 points to his teammates’ collective 28 points.

After correcting for team and competition effects, the model rates 1962-1964 as Clark’s peak 3-year form, but it rates his top three individual years as 1962, 1965, and 1963. These three years are respectively rated the 2nd, 3rd, and 7th best single-year performances of all time, making Clark the only driver with 3 years in the top 10. Even after correcting for the speed of the Lotus cars he drove, Clark stands head and shoulders above his competitors. The model rates John Surtees‘s performance in 1966 fractionally ahead of Clark’s, but then Clark is back on top in 1967, for a total of 5 championships.


Conclusion

This is the first attempt to objectively rank Formula 1 drivers that has yielded reasonable results. In many cases, there is close agreement between the model and expert opinion, but there are also plenty of surprises, e.g., Ayrton Senna down at 19th and James Hunt in 6th. Naturally, the model used is not perfect. It doesn’t account for changes in ability across a driver’s career, which can be important in some cases. For example, the high rankings of Nico Rosberg and John Watson, due to comparisons with highly rated teammates who may have been far from their peak form.

A nice thing about these rankings is that they are internally self-consistent — they must be, since they are all generated by a single model. By contrast, subjective lists often contain some peculiar inconsistencies. For instance, the 2009 Autosport list had Ayrton Senna 1st, Alain Prost 4th, Nigel Mansell 11th, Nelson Piquet 13th, Keke Rosberg 25th, Gerhard Berger 38th, Riccardo Patrese 39th, and Elio de Angelis outside the top 40. These rankings are difficult to reconcile with the following facts, without assuming extraordinary changes in performance across Mansell’s career:

  • Mansell was significantly outperformed by de Angelis from 1980-1984.
  • de Angelis was just slightly outperformed by Senna in 1985.
  • Mansell was outperformed by Prost in 1990.
  • Senna and Prost were closely matched when they raced alongside each other from 1988-1989.
  • Mansell was closely matched with Rosberg in 1985 and Piquet from 1986-1987.
  • Berger was significantly outperformed by Senna (1990-1992) and Mansell (1989).
  • Patrese was significantly outperformed by Piquet (1982-1983) and Mansell (1988, 1991-1992).

Taken together, these results suggest the following self-consistent ordering:

Prost/Senna > de Angelis > Rosberg/Mansell/Piquet > Berger/Patrese.

This is precisely the ordering found by the model.

As races are completed in the future, the model rankings can be easily updated. Applying the model to the 2014 season so far generates the following driver rankings.

# Driver Adjusted Scoring Rate (ppr)
1 F Alonso 9.20
2 L Hamilton 8.17
3 V Bottas 7.04
4 N Rosberg 6.80
5 J Button 6.66
6 D Ricciardo 6.61

In future posts I will explore other uses of this model, including determining the best teams of all time, and running historical hypothetical (what if so-and-so had driven for another team instead?). Thanks for reading, I hope you enjoyed!

Advertisements

126 comments

  1. hi again! I find a few inconsistences between the data in this post, the 2014 post and the published paper. The paper ranks Vettel 3rd in 2012 but this post says he was 4th. And the graph on the adjusted scoring rate show data not matching those on the paper, e.g. Alonso gets > 9 in 2005, 2006 and 2009 but all this scores are < 9 in the paper.

    Are the results for previous seasons changing as new data are introduced? (just imagine the mess if if happened in the real championship LOL). Was it simply a typo? How do I get the updated scores? pretty please explain!!! and a huge thank you ❤

    1. The paper used only data up to the end of 2013. As mentioned in this article, more recent data were used for the blog post rankings. Including new data naturally causes slight changes in the model rankings, as the model is able to base its rankings on additional evidence.

  2. Hi, just wanted to say I really enjoyed reading the blog and found the results very interesting, some of the results are a bit hard to swallow (Frentzen being so far ahead of Hill/Villeneuve for example, Rosberg > Hamilton etc) but it’s very interesting to see how the analysis plays out. I also don’t think it’s a coincidence that the top 5 drivers were undisputed number one drivers for long stretches of their careers; as I understand it, this would over-represent them in the model, because it assumes that all team-mates have equal equipment (and obviously equal chances to use that equipment).

    A request for a future blog post – list the top 5/10 drivers for each year? Or make this data available?

    Thanks again!

  3. […] kind people at F1Metrics have developed a mathematical model that, while it comes with certain limitations, attempts to disentangle car performance from driver […]

  4. A couple ideas I’ve came up with after reading this for maybe the 20th time. Have you considered discounting races for Force Majeure incidents? Such as the track nearly becoming undriveable in 2007, retiring 5 drivers. The pile up in Austrailia in 2002, where the track got blocked after Ralf and Rubens collided and mayhem behind them. A few laps before the race is abandoned and awards half-points(which I believe you count as full points) such as Malasia 2009 when 3 or 4 cars spun out of the race. Interestingly, no one spun out for 8 laps before the infamous 1984 Monaco GP was called. Only Mansell, Lauda, and Winklehock had spun out of the race after lap 2, which begs the question if the race needed to be abandoned.

    The other idea is to consider what long time #2 driver’s are rated discounting the years they had dominant team-mates.

    1. Thank you for the suggestions.

      Regarding point 1, it’s possible to think of many cases where a driver has something happen to them beyond their control that adversely affects their race. The types of accidents you describe are one case (although others might argue that staying on a nearly undriveable track is not to be discounted). Others might include mechanical faults on cars (e.g., loss of gears), issues in qualifying, or being crashed into by another driver. Unfortunately, there are two essential problems with trying to include such factors in a model:

      1) Data are not cataloged as accurately in the early days of the sport and race reports are less detailed. Determining the mechanical status of a car or the conditions of the track or who was at fault in an accident is often lost to the mists of time. There’s no video footage to review either in many cases. At best, it becomes a prohibitively onerous task to attempt this for the whole sport’s history.

      2) There is inherent gray area — and therefore subjectivity — in deciding which types of incidents qualify and which do not. We might ultimately improve the quality of the rankings by including such factors, but at the cost of losing complete objectivity from the framework. I would much rather keep the model 100% objective, since any deficiencies are then at least completely transparent.

      Regarding point 2, I certainly agree that drivers who spent a long time in either a #1 or #2 role were probably advantaged or disadvantaged, respectively, in the ranking scheme. However, I think we run into a definitional issue of what is a “long-time #2 driver” and what is a “dominant teammate”. Unfortunately, these roles aren’t defined in a clear binary fashion — sometimes drivers only become #2 drivers when the championship is mathematically decided, other times they are given clearly inferior machinery from the get-go. It also becomes very difficult to make these assessments back into the 1950s where information is scant.

  5. As statistician George E.P. Box said, “All models are false – some are useful”. This one is very useful!

  6. […] which few would argue was a close-fought battle. (For more details on the methodology check the mathematical model and this very brief […]

  7. Reblogged this on Racing Club Manager and commented:
    This is a very interesting read if you’re an F1 fan or a stats/charts person.

  8. Hi,

    Congratulations for your work, I rate it highly.

    However, I have a few points in disagreement.

    I cant really understand how the 80’s are judge of lesser competitive decade on the drivers level because of car reliability issues. Yes, maybe fewer drivers would finish the race, but that’s not something that should affect driver performance rating and then competitiveness based on driver performance alone. It’s a confounding factor which the model doesn’t solve well – as I see it. There is a reason that many people are led to conclude the opposite about the 80’s, and that is the number of potentially world champions until late races on a given year. That , it seems to me, is a sound measure of competitiveness on the field, and could only be discarded if it could be shown that such close championships were due to a lucky distribution of drivers over car makers, leading to a evened result when considering the binomial of driver/machine. It just doesn’t seem to be the case (for most of the decade). Anyway, even if it was, that is not the reason that is claimed. Fewer drivers finished the races but those drivers that used to finish were enough to make those championships a close call. The thing is: it really doesn’t matter what happened to the non competitive portion of the field if those in position to really content for the title, are more then it happends in other times.

    Also, I didnt quite understand how DNF is handled. Pilots should be compared to team mates only on races both finished. Is that what you do?

    1. Thank you.

      I think you may be misunderstanding how the model works. The performance of a team/driver in a given year is computed based on their points per counting race. When there are more opponents finishing races, either due to more cars in total or lower failure rates, it becomes harder to score points, simply because there are only so many points available. Similarly, if the opponents are stronger in a given year, it becomes harder to score points. The competition strength function simply takes these two factors into account as a means of meeting the requirement that the total points scored per race add up to the total points on offer per race.

      Closeness of racing should not be conflated with competition strength. Nor should competition strength be conflated with the ability of an era to produce high ranking drivers. Clark, Fangio, Moss, and Ascari all come from eras with low values of the competition strength function. Driver rankings are still largely a function of teammate comparisons. Indeed, in versions of the model that contained no competition strength function at all, Senna, Prost, and co. were still all outside the top 10.

      The metric for performance is points per counting race, averaged on a per-season basis.

  9. João Coutinho · · Reply

    I think I can understand your answer, but I still disagree. Specifically here:

    “When there are more opponents finishing races, either due to more cars in total or lower failure rates, it becomes harder to score points, simply because there are only so many points available.”

    Although that would be 100% true if all drivers had the same probability of wining the race, since they have very different probabilities this parameter must be diluted to far less then 100%. What i think is that it should be much closer to zero then to 100%

    This would also be a lot more meaningful if there weren’t something such as qualifying, which sorts low probability contenders out even before the race starts. I dont think its plausible to assume that putting more drivers on the grid that low scoring probability raises competitivness.

    1. The effect of the competition strength function on scoring rate isn’t uniform, because there’s a sigmoid function applied to the performance variables. If a driver is already far ahead or far behind the rest of the field, the addition of a few extra drivers to the middle of the pack is of course going to have less effect on their scoring rate. That’s incorporated in the model.

  10. There are more realistic reasons to say that Schumacher is a better driver than anybody else, because he is simply leading most relevant statistics, more Championships than any other, drastic more wins than any other and in many other fields… This here is not of relevance. It is like who was the fastest Driver on 1.6.78 between curve 2 and 3. He must be the best of all times. The reason for ignoring Schumahcers outstanding success or to diminuish it is because he is disliked, especially by the Brits. Any british llist, will Show this.

    1. Schumacher should not be considered the best ever for two reasons: he cheated to at least oome of his titles, if not 2 or 3, and the fact that on his return he was beaten by Rosberg

  11. […] statistics are not a metric for assessing greatness. At least not raw statistics without context (for that see the brilliant F1Metrics blog). The top 100 represents what journalists thought at the time – it is not clouded by statistical […]

  12. Excellent work. I wish Alonso had half the championship wins that your model gives. It was very upsetting to watch his struggle in the Vettel era.

  13. Although your analysis is well researched, for me statistics can never capture the romantic essence of what fans and pundits believe to be the true attributes of great racing drivers. Ayrton and Gilles threw away a lot of results away in promising positions, but they often drove beyond the limit in pursuit of achieving the impossible.
    James Hunt’s ranking of 6th is incredibly bizarre. He only drove in F1 for six years (92 starts) and had no team-mate in his first two seasons (1973-74/Ian Scheckter DNQ) and 1979. Therefore he drove exclusively in what I call “The Kit-car era”, where Ford Cosworth DFVs were used by almost every team and budgets were fairly limited in proportion to its decade (rate of financial inflation) when compared to subsequent decades. James Hunt (and John Watson to a certain extent) appear to be a beneficiaries of having driven in an era where driver comparison is more linear and less convoluted- closer time gaps over qualifying, heroic giant-killing acts more frequent (1970’s Austrian Grands Prix being notable), but until the appearance of ground effects and turbos, very little variety of technical differences between cars. F1 in the 1980’s could not be more different, with rapidly increasing budgets- therefore much larger time gaps between the fastest and slowest cars, where turbos and “atmos” were permitted, far, far more mechanical failures due to increased experimentation and more rigid tech regulations.
    I can understand that this “greatest driver” formula makes sense if you believe true greatness was based on a driver maximising his machinery, making as few errors as possible and performing as far above his team-mate(s) as possible. Modern drivers such as M Schumacher & Alonso had particularly weak team-mates (Jos, Lehto & Herbert/ Fisichella, Piquet Jr. & Massa), whilst others such as Hamilton have had strong team-mates (Kovalainen being a weak exception).
    Vettel, however, had a team-mate in Webber who compared to him fairly well in their first two seasons (2009/10)- but due to the new Pirelli tyres and the re-introduction of the KERS (2011-13), which caused huge weight balance issues for Webber due to his hefty 75kg body weight- caused such a dramatic growth in performance gap between both of them. Some may say due to Vettel’s increasing maturity, such an improvement was inevitable, but others feel that the tech changes plus the Red Bull management’s treatment of Webber contributed to such developments.
    As for your criteria on basing a driver on his “best three consecutive year period”, it helps to reflect better upon some great drivers than others. Your top five (Clark, Stewart, Alonso, Schu & Fangio) were almost untouchable in their pomp and were very much the undisputed no 1, but had team-mates miles behind them for various reasons (poor car handling characteristics, lesser team assistance etc.). Hamilton, Senna & Prost tended to have team-mates closer to their level (Alonso for Ham, de Angelis for Senna & Lauda/Keke for Prost), therefore enjoying a smaller advantage.
    What your particular brand of statistical analysis succeeds at, it is removing emotion normally reserved for fan-based lists. For fans, whilst Senna & Gilles had the demeanour of a gladiator, but for managers they were EXTREMELY infuriating and often disobedient (not so much team orders, but more in terms of bringing the car home and not being so wrapped up in terms of personal glory). It also does a good job of highlighting the careers of drivers (i.e. Salo & Beltoise) who deserved better machinery, but also how underrated Watson and Frentzen were.
    Ultimately, no model can ever disguise the extreme and often bizarre performance discrepancies shown through F1’s statistical history. Perversely, whilst outsiders decry the lack of mechanical equality, the deep-lying fascination fans have is the sight of a truly magnifying giant-killing act that no other sport can replicate.
    It’s a good thing that you did not include any complicating factors (team orders, blame for crashes, age) because I actually think you have provided the best formula/criteria for determining this pesky best driver debate. After all, there is no end to the whole notion of what a real great driver should adhere to be.
    As for Nico’s ranking of king of 7th, it is also weird but he has actually proven to be underrated. Some of his performances pre-2014 indicated he is better than a so-called no. 2 (i.e. Barrichello, Coulthard, Fisichella), better than Webber and possibly better than Button. For me, most of why Rosberg has done worse this year is his indecisive mentality (and unnecessary calls for help via radio) inhibiting his talent. Your ranking for Nico suggests he should be doing much, much better and if that is true, then to put it plainly, he’s choking big time.

  14. Fantastic read. Enjoyed it throughout. Thanks!!!!!

  15. I understood the idea, but I don’t know if I agree…
    I think to choose the best driver in the history is a very hard task, and there are many ways to do that…
    The purists will try to idetify qualities and characteristics in the drivers they love. The more mathematicals will try prove with numbers e records. I believe all methods are important, and should be considered…
    I like this method but, as the method by Eichenberger and Stadelmann, bring a questionable result. It’s normal!!!!…
    I am a purist. I watch formule 1 for a long time and I like to identify characterists in the drivers that make them better than others…
    Senna was the fastest driver in the history… Nobody was better than Senna in qualifying laps and flying laps. He was the truth Driver, with a big D…
    He got in the car to win and to be the fastest.
    Prost was the genius… He was not faster than Senna, but was very smart and cunning driver. He knew exactly what he had to do to win…
    Schumacher and Vettel knew enjoy their moments to be champion. Both never had strong opponents in the same car (In this point I agree, when you compare the competitors with their teammates).
    So, they ran alone, without any threat…
    Piquet was amazing with his knowledge about mecanic.
    Fangio was a great driver. He showed to the world what was drive professionally. Amazing to that era…
    Each driver has your characterists and particularity. It’s very hard to compare based in numbers, because of the variables of each era.
    If somebory ask me who I choose as the best driver in the history, I say…
    Speed by Senna, brain by Prost, brightness Vettel and Schumi stars, the knowdlege by Piquet and the story by Fangio. THIS IS THE PERFECT DRIVERr!!!!!

  16. JohnS · · Reply

    “What we can do is make direct comparisons between drivers in the same era by comparing them to their teammates. Since drivers typically have many different teammates across their careers, we can obtain an estimate of how drivers rank relative to one another.”

    That does not actually tell us anything though, since it is common in F1 for certain teams to have a “Number One” driver who is favored over his teammate. So your methodology has a built in bias in favor of drivers who have had that team status – Alonso prominent among them. Your technique would only be useful if all teams at all times treated their drivers with absolute equality, something which Ferrari for instance is well known not to have done.

    Also, the notion that Trulli, Fisichella, and Nelson Piquet Jr. were “very highly ranked drivers” calls your impartiality into question.

    1. Regarding #1 drivers, you are of course correct. This is already discussed in the article; nothing new is added here.

      Regarding Fisichella, Trulli, and Piquet Jr., this is not my opinion. These drivers were each very highly ranked by journalists and pundits, especially at the time of their respective arrivals in F1.

  17. Although I have to admit it’s the best try at an objective ranking list I would like to make some comments regarding both the idea and the model itself because the discussion in some forums gets heated and it might have deterred you from commenting .

    Regarding the idea:

    1. As you say you weren’t able to account for changes in a career because of age.

    2. It is implied, in my opinion, that these rankings are based on skill, which indeed they are heavily based upon. But one thing you can’t measure is motivation and say that regardless their age Schumacher was equally motivated when he came back, Piquet was equally motivated from ‘88,’89 onwards (when by his admission he was racing for the money after the accident), Damon Hill in ’99 or his father for that matter at the latter stages in his career after his accident while trying to build his own team. But at the end of the day performance in Formula 1 is dependent in driving skill as well as motivation, technical amplitude to convey what you need from the car to your team, ability to work with every team etc. So in that respect I agree that Driver Performance and Team Performance should be independent of each other (excluding No2 drivers who get hampered by their own teams).

    3. Accurate comparison between two drivers can’t happen through a common teammate, especially due to the limited number of data. You can’t possibly claim that for example driver A having scored 50% of driver B in one car and driver B scored 50% of driver C in another car, that driver A would score 25% of driver C in a third car. Or that Raikkonen would score the same percentage of Alonso’s points if they were in the Mercedes last year. You could probably say, given such large differences, that maybe A<BLauda>Reutemann>Andretti>Peterson just from the data(without taking into account Andretti’s No1 status) since you suppose consistent driver performance.

    4. Which brings me to my last point on the idea. That quantitative analysis doesn’t mean a lot without it’s supplementary qualitative one. The model ranks Rindt’s 1970 season as the best. But it doesn’t take into account that he inherited wins by a lot of cars in front retiring. You compare him with a rookie Fittipaldi running the out-of-date chassis in some races and Graham competing for another team altogether and not mention his let’s say “not Formula 1 material teammate” Miles. Just from the diagram of Clark-Stewart-Rindt it seems an anomaly to jump from a 7ppr driver according to your model to a 10 ppr driver. I doubt any pundit would rank this as the best performance ever.

    5. Also the quality or context of every teammate battle isn’t the same. Some battles have more meaning than others. For example one of the main goals of rookies is to at least beat their teammate. A battle between two drives in a dominant car is another instance of a pure teammate battle. Two established drivers’ goal won’t primarily be beating their teammate rather than make a title bid. You mention De Angelis and Senna being close although it wasn’t. I don’t think Senna’s goal was to beat De Angelis but to make a title bid. Especially considering Senna had double the number of De Angelis retirements and Elio never (or almost never I don’t remember exactly) run in front of Senna. Or Schumacher not really caring about Rubens whether they were in a dominant car or not and taking risks that would potential help his title challenge rather than comfortably beat his teammates.

    Ignoring everything above and the plausibility of the idea, regarding the model:

    1. I understand that with average finishing position it’s difficult to find an objective way to deal with retirements but why subjectively use a point system with a static 5-place exponential growth rather than a 1-place one? That way you gift different place gains different rates. I find that to be one of the main flaws of the model and I don’t understand the reasoning behind it.

    2. The biggest flaw I see is that you suppose Team performance stays the same at 61% through the years although not only is there is no reasoning behind it but it just isn’t true when the teams nowadays help their drivers way more(not only from car to pit radio but pit to factory radio, data gathering etc) and in the ‘50s for example they just assembled a car and maybe a change in the setup. The overall model may fit the data you inputted but 5=1+4=2+3 is not the same.

    3. By using the quality argument I mentioned before another flaw I see with the points per counting race is that you suppose that in each and every track a specific car will have the same chance of scoring points which isn’t true again in a lot of cases. Another thing that this misses is the classified finishes and retirements because of mechanical trouble which can’t be depicted in numbers.

    4. You say assigning blame on crashes can be subjective (although you have an article called “rules of racing”) but you penalize drivers of today who race more closely and it makes a big difference if drivers A points are x/10 rather than x/9.

    5. Conversely you subjectively take into account running out of fuel implying the danger for that to happen was the same in the ‘60s compared to the refueling era or even now when the drivers are coached through the radio with the help of 100% accurate fuel monitoring.

    6. I won’t try to guess how you calculated the competition effect but the fact that you found or suppose that the ‘80s drivers are lower than “expected” because of this while you have drivers from every other decade in the top10 seems weird. I hope you somehow included that competition isn’t the same through the field and through the years at the same time. Because you say that the most competitive year to date was 2011 but certainly Vettel had it easier than other frontrunners on previous years.

    7. You subjectively again change the rules and ignore that for 40 years they didn’t count all the races rewarding peak performance, by which you ironically rank them by, notably “affecting “ the championship contenders in 1964 and 1988 and non-contenders in more cases. Specifically in 1988 which was a pure teammate battle and Senna came out on top, by changing the rules Prost comes out on top judging from a diagram in your article. Don’t you find by yourself anything inherently wrong with that? It’s similar to the non-championships races drivers used to race in the ‘50s and ‘60s.

    8. Judging the “abnormal” ranking of James Hunt at 6th place by his 1973-1975 performance, or mainly by his 1973 performance it seems that you treat cases of customer cars or cases when the engine and chassis may be the same even for some races, as teammates. But how is that plausible when you don’t necessarily have the same body(especially at that time when everyone was experimenting with aerodynamics), don’t use the same tires ,same transmission, same wheels, maybe same fuel, and you don’t have the same mechanics to not only assemble but also setup your car, and maybe even more differences through teams which we don’t know of?

    9. Have you included skewed results of safety car races that didn’t finish under it?

    10. I suppose since it’s model fitting that the data are weighted.

    11. Lastly have you tried running the model in a closed period of time to see if the fit is better? Like before any of the current drivers were driving or upon Schumacher’s retirement in 2006?

    Overall I greatly enjoyed reading your article and appreciate immensely the amount of work put into it and the outcome itself. But at the same time it also made me appreciate the “experts’” subjective opinion, be it journalists or team principals who give chances to good talents in good teams, because in general it seems your model agrees with them. Excuse me for the long post.

    1. There are lots of good points here, so I’ll respond briefly to each. Regarding your points on the idea.

      2. The model quantifies performance rather than any particular component that defines performances, be it “skill”, “motivation”, “talent”, or any of a host of factors. While it would be fascinating to determine contributions of specific factors to overall performance, that wasn’t the goal here.

      3. The model is not making such direct inferences. Rather, it is simply finding the estimates of performance that best explain the data.

      4. There are two separate points here. One relates to driver skill changing over the career, which the model currently neglects, but I’m working on solving that. The second point relates to looking at race results in a more nuanced fashion (e.g., very carefully analyzing good/bad luck in every race). While there are advantages to that approach in certain circumstances, it’s untenable (and impossible with the data we have available) to perform such an analysis for all drivers in all races going back to 1950.

      5. This comes down to a subjective interpretation of driver motivations, the implication being that they coulda-woulda-shoulda performed differently with different motivations. These types of arguments are extremely speculative and unfalsifiable.

      Regarding the model.

      1. The exponential trend simply follows the trend that exists within the 10-6-4-3-2-1 points system. To wit, a factor of 10 decrease in points across 5 positions (1st to 6th). I am just extending that trend to positions beyond 6th, rather than dropping to zero for all positions beyond 6th.

      2. I’m not assuming that. The 61% is an overall estimate across all years. The spread of team performances and driver performances of course does vary between years.

      3. That certainly may be true, but it’s not statistically quantifiable from race results, because there is only one race at each track each year.

      4 and 5. Since drivers are directly compared only to their contemporaries, this should make no difference.

      6. I personally don’t have any problem with the results contradicting the sport’s pervasive narratives, including the idea that the 1980s was an era of titans. The competition effect takes into account the number of opponents, their strength, and how often they finish. It is an exact formula that can be mathematically derived (details in the paper if you’re interested). A large part of the reason for the dip in the 1980s is poor reliability — races had on average fewer finishes, making it easier to score points in counting races.

      7. I don’t ignore this, I mention it explicitly. This system had very little effect in practice, since drivers rarely had more points finishes than counted races, and even then it usually had little effect on the order of drivers in the championship. Only in two years did it actually determine the champion. Certainly this choice will have a small effect on the overall results, but not one worth worrying about.

      8. At some point, it’s necessary to make a choice as to what you consider teammates. This is the choice I made. You could if you wanted to go down to the level of the chassis number, the individual parts on the car, etc. If you want to be sufficiently pedantic, you may end up with no true “teammates” in history.

      9. I’m not sure what this means. The model analyzes the final results of races, regardless of how they occurred. All races are weighted equally.

      10. Each driver’s result for a given team in each season is weighted in proportion to the number of counting races they completed together that year.

      11. Only in a few cases, but there’s not much difference in the quality of fit. It is useful in some cases to separate the data though (e.g., treating Schumacher as two different drivers).

  18. Wow that was a fast response! I appreciate that.

    2. Yes, I just pointed out that I agree driver and team performance shouldn’t be dependent on each other (except for No2 drivers which is impossible to quantify).

    4. Oh I am impressed and glad to hear that you try and implement that. Subscribed! Although I haven’t been able to get my hands on and read through it yet, I believe a book called “Formula 1: All the Races: The World Championship Story Race-by-race: 1950-2011” if it is what it says it is, could help in a situation like that if you could get your hands on a cheap copy of it. I made this point because it was striking to me to name 1970 as the best performance since 1950.

    5. I agree but it doesn’t mean it is not true.

    Regarding the model:

    1. Although I concur that this point system is better than what we have had the past decade, when you try to compare the drivers by taking the team effect out of the equation, as I understand it from your model, a driver who beat his teammate by a factor of 10/6 for one place wouldn’t be the same as another who beat his teammate by a factor of 2/1. Be it 1-2 or 6-7 against 5-6 and 10-11. Since the performance is corrected with the competition effect (which as I understand it is different through the field) having a constant exponential growth would be more logical than rewarding some position gains different than others.

    2. Excuse me if I misunderstood this or worded it wrongly. So 61% is the average through the years and not constant, meaning your model can calculate what is the exact for each team and each year? That would be great to see.

    3. You are absolutely right about that. But since they have driven in the same circuits through the years do you think it would be sane to extend this research and calculate probabilistic results based on overall performances on each occasion and under the circumastances they retired under?

    4. Even when you claim that, by taking the latest example of Raikkonen-Ricciardo incident and assuming it was race ending for Raikkonen, his uncorrected ppr per race would be 52/5 rather than 52/4 thus far (if you would assign blame according to your “rules of racing” article with which I agree with). In my original post I wanted to also give the example that Prost and Lauda retiring on their own doesn’t reflect their skill, but since by performance they cost themselves points I guess I am torn with this.

    6. I don’t either but it seemed weird that you claimed drivers from the ‘80s get penalized because of the competition when drivers from eras with less competition appear at the top. Maybe it’s just because of their dominance and the competition effect doesn’t have that big of an effect on them.

    7. As I still believe that even the smallest differences could be amplified through the years, specifically when the rankings by ppr are so close, excuse me but I have to disagree with that (mind you, I say this reluctantly as I don’t know exactly the exact methodology you have used). It just doesn’t seem objective when we have the example that happened and doesn’t derive from a model right in front of our eyes and you change the result by changing the rules (which you might claim are fairer rules but it doesn’t change the fact they are different rules). For example imagine scientifically proving that in basketball a 3-pointer judged by it’s difficulty should count for 5 points. If you tried to rank players like that it would be almost certain that if they played under this rule they would try for more 5-pointers. I hope you can see my logic even if you find it flawed.

    8. I understand it’s a way of gathering more data but I find it wrong. For example Sir Stirling buying customer cars should have a slightly lower team performance percentage in his results than the works drivers. Of course the percentages won’t be big even with Hunt’s case. But when comparatively Hunt and Fittipaldi faired similarly against Mass, such a jump in ppr doesn’t seem natural, albeit neither does it seem impossible.

    9. Well up to ’96,’97 was it, even with the safety car the results were on aggregate and in my opinion fairer. I wouldn’t include races after that which had a safety car at some point in the race (unless it was the last laps and finished under safety car conditions). Just my personal opinion. Although an exiting race, i doubt Button would repeat his performance without so many safety car periods.

    11. Have you tried using more accurate polynomial or spline fitting models? I can’t think of a way to implement them at the moment but I was curious.

    Thank you for taking the time to answer.

  19. Thanks for this. Very interesting. Who do you have as top performer in 1981? I don’t think you say directly–if so, I missed it–but it might be inferred that you give that honor to Watson.

    1. It is indeed Watson, narrowly ahead of Prost!

  20. @NatxoVillar · · Reply

    Great job!
    Thank you very much for sharing us.
    Don’t give it up!

  21. […] I presented a mathematical model’s ranking list of drivers across the sport’s history. Rather than looking at the usual statistical metrics (wins, titles, poles), my model worked by […]

  22. Suphonamide · · Reply

    Congratulations on doing the most important thing – entertaining the reader (and making them think, even if only for ways to dispute the result). A question (having read for enjoyment rather than taken the time to analyse all the clever stuff you have done): for the Senna-Prost McLaren era, do they each lose points simply because they were not dominating an equally-brilliant team-mate? When they were both at their peaks, was there anyone who could get close to them for a sustained period? I think most people would assert that Senna could do things the nobody else was capable of – and would have destroyed a team-mate like Nico Rosberg or John Watson – like Jim Clark (and the romantic image of Gilles). Do they suffer compared to all of the top 15 drivers because they have no overlap with them and have somehow been unreasonably penalised by an inherent flaw (and I don’t mean that in an insulting manner) in the formula. The only overlaps seem to be a naive Prost vs Lauda in 84 and Senna vs Schumacher (in a illegal-ish Benetton). With no way to link these two through to the top 15, it strikes me as though there could be something wrong – as if they are a floating separate case and could just as easily be far above the top 15 as below it.

    Secondly, rather than use a driver as a comparator whose skills do not dimish (Rosberg beats Schumacher in the Mercedes and thus is top 10?) can the performace of that driver (within or outwith their 3-year top period) not be factored in – or is there simply not enough data to allow that?

    Apologies for suggesting that you haven’t already done this and are way ahead of me: I suppose for me (with a background of modelling biological data in relation to chemical structure) the main issue is that that the model needs to be refined continually, using training sets and test sets, until it makes sense whichever way you look at it. So remove some of the drivers when you create the model and see where they then come in. Remove a different batch and see if it all still matches up. The main issue is that the data has to match with what we know – but clearly there is no way to “know” whether Senna was better than Fangio, but few would doubt that he was better than Rosberg. It seems somewhat back to the Greek Philosophers and Medics – the proof says this and thus it must be true, whatever reality may say.

    With that in mind, can you go from empirical “evidence” and create a model which proves it – in this case, can you model what people think they “know” and get each of Fangio, Clark, Senna, Prost and Schumacher into a top 10 and not have anything ridiculous which makes the model look just as unreasonable as Hunt and Rosberg topping Senna and Prost?

    An as for John Watson! I started watching towards the end of the 1980 season and he was never anything more than a journeyman!

    Thanks again for the superb job – I shudder to think how long you have spent on this!

    1. Regarding Senna and Prost, there isn’t any inherent reason that the model should penalize top drivers who spend time as teammates. All that really does is help establish their positions relative to one another. It’s the comparisons with all their other teammates that lead to the conclusion that, while they were clearly the two top drivers of their era, they perhaps weren’t as singularly brilliant as some others in history. I don’t see any particular anomaly there in terms of linking their era to others past and present. The graph in Senna’s entry shows how they compared to the next best drivers of the era — they never had a sustained advantage over the rest of the field such as Clark’s, Stewart’s, Alonso’s, or Schumacher’s.

      Regarding changes in driver performance across the career, that is something I am working towards. I think that’s the most important next step in improving the model.

      Regarding Watson, he’s likely overrated somewhat for the reasons I’ve outlined here and in more recent articles, but I would consider it bizarre to call him a “journeyman”, given he was likely a top five driver for most of his career. Speaking subjectively, I would rank him lower, but probably would not consider ranking him behind Mansell.

      Personally, I’m not at all interested in reproducing what people “know” to be true. I agree that any model should pass a very basic sanity check (e.g., it shouldn’t have Katayama ahead of Mansell), and it is preferable for it to agree with rankings that are strongly supported by evidence-based reasoning (e.g., it would be preferable to rank Rosberg behind Hamilton). Beyond that, I’m quite happy for the model to outright dispute popular opinions if that’s what the data indicate. A lot of what people “know” about driver rankings is based on very flimsy evidence that has been gradually painted into an unassailable narrative. I certainly have no objective issue with Senna falling outside the top 10, nor Hunt falling inside it. Nor do I have a problem with the model concluding that the best drivers of today are superior to the best drivers of most prior eras.

      Thank you, it was indeed a very time-consuming effort! 🙂

  23. […] my mathematical model, we can ask which teams would have beaten their nearest rivals by the greatest margins if all teams […]

  24. AllHailRoot · · Reply

    Most of these points are mentioned, not covered, in the original post.
    Statistically these rankings have certain problems – the use of the standardized ranking points system will cause some distortion. This is as driver strategy towards the end of the season may adjust towards getting results for the WDC and therefore race strategy can be different.
    However I think the huge issue with these rankings is the effect of differential performance. Normally rankings are a very good predictor as there are lots of interfaces between the different competitors (such as with chess and elo rankings) and therefore an accurate ranking can be obtained. However, with F1 there are very few direct interfaces between the drivers which leads to large distortion. This is perhaps most noticeable with the ranking of Senna and Prost. The peak of both their careers coincided with each other and this is not reflected by the model. This lack of directly comparisons between drivers, combined with the high variance that occurs even in these conditions, are what makes rankings so difficult to create in F1. I’m not sure these are problems that are able to be circumvented.

    On a separate note I think subjective rankings of drivers display the general tendency in sport to moments of brilliance and charisma rather than consistency and effort. When watching sport we long for the spectacular and the unbelievable – the flash of inspiration over the years of graft. It is peak, rather than average, that many think of when they determine the best. To me, this is the reason of the elevation of certain drivers (Senna, Gilles Villeneuve) over others (Prost and maybe Schumacher).

  25. How do you handle classified non-finishes? Many times it can leave the driver with an unrepresentative “finishing” position. On the flipside, there was the case this year in Belgium of Vettel not finishing the race, but being classified higher than many lapped cars that did take the flag. I would say examples of the former shouldn’t count, but a classified non-finish that’s higher than other classified finishers should probably count.

    1. For non-classified finishes, I read race reports to find out the reason. If it was due to a mechanical problem that stopped the car in the closing stages, it is treated as a mechanical DNF. If a car is still running but just many laps down, it is treated as a finish. This is considered independent of whether the car was classified ahead or behind of others.

  26. Just to clarify, when I say a classified non-finish, I mean those instances of a driver not seeing the flag, but covering over 90% of the race distance. As opposed to a non-classified finish, where a driver sees the flag, but completes less than 90% distance (Bianchi at AUS 14 the last such instance of that).

    Also, I’d like to know how the scoring went this year for BUT-ALO, and VES-SAI? It seems to me that ALO wanted to retire from a few races this year, just ‘cos. He wanted to at ABU 15, but ended up finishing the race. Retiring or finishing in that race didn’t make any significant difference, but would in your model, especially as it’s the first year of a direct head-to-head between Alonso and Button.

    Similarly, Verstappen’s score will be negatively affected by his two DNF’s through Accidents. Is there something to be said for drivers that are perhaps more aggressive, but bring in the results through that approach, rather than a more conservative driver who “keeps their nose clean” but who scores lower average points per finish? At the end of the day if a driver can pull in more points through the former style, over the latter, then that’s the style they should employ.

    1. It depends on the reason the car stopped. If it stopped for mechanical reasons, it’s a non-counting race, even if the driver was classified in the points.

      Rankings for 2015 should be posted very soon!

  27. […] for all the 2014 season drivers. These rankings are derived from a mathematical model, described here, that statistically estimates the strength of driver and team performances in each year, as well as […]

  28. Your article was cited in this month’s Journal of Qualitative Sports studies by Bell, Smith, Jones, and Sabel and yours was the most mentioned it seems probably because it is one of the most extensive and recent. Interestingly they followed your point per race normalization but removed the non-driver finish exemption, the less than 3 races factor, and evaluate the entire career and not a 3-year peak. Doing so helped Nelson Piquet immensely getting him into the top 10 from outside the top 50 in yours and the Eichenberger and Stadelmann study you cite. Niki Lauda is #142 due to having only 5 years of good results in his career and Ascari at #10 is #76. And Christian Fittipaldi is their big outlier on the other side at #11. What are your thoughts?

    1. I’ll write a more detailed analysis of this in a coming blog post. To keep it short for now, I think it’s really fantastic to have another model in the literature, as it’s important to understand how the rankings are sensitive to different underlying assumptions. In this particular model, I’m not fond of the way DNFs have been handled (they are all treated as low finishing positions, regardless of cause). That is, I think, the reason for most of the rankings that seem obviously absurd, such as Lauda’s (see his horrible reliability in 1985).

  29. As a follower of yours I also wanted to write here to congratulate you on your citations from the recent work of Sheffield university (http://phys.org/news/2016-04-reveals-greatest-formula-driver.html).
    Interestingly, the 3rd best in both research works (yours and Sheffield’s) is a driver still active (F. Alonso), which means we are in a moment in history where we can try to predict and see.
    I mean, could he still improve to 2nd or even 1st if McLaren deliver a winnng car? Is this possible within your model predictions? Or perhaps not regardless of the car performance?
    Waiting for your next post on this issue!

  30. Andrija · · Reply

    Great work. Do you intend to make some list that would be regularly updated after every race and after every new information you obtain or every new update to the model? I’m talking about both greatest drivers list and current season list?

  31. […] I began this blog, I had just embarked on a major analysis project that ultimately generated model-based rankings of the best Formula 1 drivers across history. The essential difficulty in rating drivers (compared to rating athletes in many other sports) is […]

  32. Once again very good job with the model, was nice to read etc.

    But I also have a doubt about senna and prost: as a schumacher fan I don’t really mind senna being far behind him, however here I think something is wrong.

    While I clearly get the point that senna and prost weren’t as dominant compared to their teammates as schumacher and alonso, for example, a thing I can’t do anything but notice is the following:

    The way you used to say schmacher is in 4th position is simply taking his actual points per counting races in 1994, 1995, 1996 with the car exactly as it was (because I did a test and it comes incredibly close to his actual points per counting race written in his position rank).

    Therefore, the fact senna and prost were extrmely closely matched in their years took away a lot of points from them, to the point you didn’t use the years where both drivers were in mclaren, but the previous ones (most likely because they were the years they scored the most points per counting race, obviously).

    Let’s try to imagine a 1.988-1.990 (3 year peak) without prost then, senna would (as a guess) get a lot of points per counting race that year and thereby increase his rank consistently.

    So what I’m saying is that despite the comparison between team mates helps to understand how driver a would compare to driver c after knowing how driver a compared to driver b and driver b compared to driver c, it seems to me like the most important way used to determine driver rankings in this model is the 3 year peak, where obviously if you were dominating like schumacher did in a few of his ferrari years it would mean more ppr or a better rank (try taking only 2.001, 2.002, 2.004, probably he would get even more ppr), but when not dominating and prost and senna are absolutely the 2 top drivers who were in serious competition between each other, you share the wins, 2nd places etc. meaning lower ppr and lower rank.

    1. The Senna-Prost comparison is interesting. The fact that they were closely matched as teammates doesn’t in itself preclude the model from ranking them both below, equal to, or above Schumacher. It just gives the model confidence that Senna and Prost should be rated close to one another, which is sensible. Their relative rankings compared to other drivers really arise from the combination of all their other teammate comparisons.

      1. Ah, ok, you had also stated that above, and I had already read all post ofc, it’s just that based on what happened with a test I did with schumacher, it seemed like his ppr ranking was almost exactly considering the points he scored excluding obviously dq on technical grounds or mechanical DNFs, which, considering his peak includes 1996 where he drove a vastly inferior car compared to the williams, would lower his ppr compared to what he’d have scored if we could have taken only 3 years in which he dominated with ferrari.

        However the test I did now with senna makes me notice that the ppr take into account the car driven (which I had read but didn’t believe based on schumacher’s experiment I did), because:

        19. Ayrton Senna (1985-1987, 7.33 ppr)

        I tried, maybe not perfectly, to consider what races he must be counted and what mustn’t and I got 156 points in total in the 3 years mentioned which with 34 races only gave 4,59 ppr, even if not perfectly done it should be around that, so the fact you got a ppr ranking almost 3 points higher indicates you actually took his inferior car in consideration, which is good.

  33. […] of junior Achievement and Excitement scores with driver performance rankings in Formula 1 using my model of driver vs. team performance. I used every driver who debuted in Formula 1 since 1990 and used average whole-career points per […]

  34. […] from all years of the championship (1950-2016) are used to estimate driver performances (including all-time rankings), and from this fitting procedure we can obtain estimates of driver and team performances in […]

  35. This was an excellent read – thank you! Are you planning to update this now that the 2016 season is over?

  36. Will you update this till the 2.016 season anytime soon? As someone asked on the other thread it’d be interesting to see the changes in rosberg, hamilton, verstappen etc. places, especially since rosberg was compared with a schumacher over 40 for a while and now he’s had more years against hamilton.

    1. it was already published

      1. sorry I misunderstood

  37. Great article, very insightful. I do like the approach and the effort to achieve maximum objectivity, because as much as I enjoy fan arguments (yes, I do enjoy them), I still wonder who really was the greatest F1 driver ever.
    I have 2 more comments:
    1. I think James Hunt in 6th is an artifact of the method. Hunt was known for his aggressive driving style and proneness to accidents. It is possible that some of his mechanical failures could be attributed to that. If it was a factor that the drivers were conscious of, it would be understandable that they would prefer a more “careful” style of driving, less stressful on the cars. And even if they weren’t conscious of it, a less stressful style would still be more successful and therefore more viable in the long run. If that’s the case, than it would mean that Hunt was a less experienced, less skillful driver than this method makes him appear to be. I do understand though that it is practically impossible to determine to what extent a mechanical failure is a result of the driver’s choices, so if Hunt’s #6 is in fact a fluke, it is impossible to eliminate it.
    2. I don’t agree with the decision that drivers with the same car are considered belonging to the same team. Team =/= car. It’s typical for a car to be modified during the season, and I’m sure that when a team had bought a racing car, they didn’t just accept it as is and did some other modifications. Any such modifications would reflect the skill and expertise of the team, but not necessarily the quality of the car. So really, the equation is more like: performance = driver + team + car. Even in factory teams, I would imagine there’s a “home” team that designs and builds the car and its spare parts, and there’s a field team, which is somewhat different in the sum of their expertise, skill and facilities available to them. So if you want to focus on the driver’s performance and eliminate all the other factors as much as possible, different teams should be considered different teams regardless of which cars they used.

  38. OLIVER FELIX · · Reply

    i dont know whats all the holla balloos about when its really quite simple…like alonso said: evaluate each driver in equal playing fields where its all level…and see which driver has accomplished the most in the smallest amount of time given…that’s it!

    1. “…in the smallest amount of time given..” A lot of the F1 drivers of yesteryear didn’t get much time. Anyway, by that standard, it’s Jim Clark.

  39. Charlie Snowball · · Reply

    I have loved reading all your articles that use this mathematical model, and would love to see more data gathered from it.

    For instance you often reference the ranking of individual seasons, like saying Rindt in 1970 was the best individual performance in history. It would be amazing for you to release a list of the best individual seasons etc.

  40. By the way, what happens to the rankings if you treat Piquet before and after his 1987 crash as different drivers?

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: