Having had a strong interest in Formula One racing for many years, I have come up with a driver rating system, which covers Grand Prix seasons starting in 1985 (I was born in 1984, so I thought it would be good to start with the first full year of my life).
The system that I have chosen is inspired by the recent spate of F1 sites that have given the drivers ratings out of 10 for each race and come up with seasonal averages, such as on the online Autosport, as well as the complex mathematical system used by the late Patrick O'Brien for Patrick O'Brien's Grand Prix Ratings and the adjusted driver ratings used in the Daily Telegraph. There are also model-based approaches to assessing F1 drivers, in particular F1metrics is well worth checking out.
Methods
Each driver starts with a default rating of 9/10 and points are then added and deducted based on the following factors:
Speed: Assessing how far off the pace a driver was, deducting 0.5 marks for each 0.25 of a second off the pace (thus a 1 mark deduction for each half a second off the pace). This takes into account qualifying and race pace, but with a higher weight given to race pace as that's when points are won and lost.
Errors: Up to 3 points can be deducted for making errors, depending on the severity of the error. Up to a further 3 points can be deducted for unsporting behaviour such as deliberately crashing into someone else. I tend to go easy on drivers for "racing incidents" where the distribution of blame is controversial.
Racecraft: Up to 3 points can be added for particularly strong performances and good wheel-to-wheel combat.
The maximum that a driver can score is 10, and the minimum is 0. In practice, very few drivers reach the ultimate speed so season-average ratings are invariably lower than 9, even for the very best drivers. "Good but not great" drivers tend to be in the mid to high 7s, the greats tend to reach the mid-8s.
I often refrain from giving ratings to drivers whose cars were several seconds off the pace, especially if they don't have proven team mates as a benchmark. To take an extreme example, it was impossible to give Perry McCarthy's performances in the 1992 season in the dreadful Andrea Moda-Judd a fair assessment. Drivers also don't get a season average rating unless they have competed in at least five races.
By "the pace", I mean the maximum amount that the very best drivers could reasonably extract from the cars that they've been given, and any deficit to that is chopped off the driver's rating. It's a similar concept to Patrick O'Brien's concept of the 100.0 rating where he assesses drivers against the fastest driver of a given era and expresses the difference as a percentage (e.g. giving a driver who is 0.6% slower than the fastest driver a rating of 100.6). I admit that my version is nothing like as mathematically rigorous as Patrick's, though it relies less on the assumption of the fastest driver of a given era being the benchmark.
The "Racecraft" factor rewards entertainment, offsetting the "Errors" factor which would otherwise penalise drivers who are aggressive but accident-prone and favour the consistent but unspectacular ones. A good example of the counterbalancing nature of the two is that Ayrton Senna (very fast, very aggressive but accident prone) and Alain Prost (unspectacular but very consistent, and fast when he needed to be) tend to come out very evenly matched on my system.
I admit that there's a fair amount of subjectivity involved with this system, and as with Patrick O'Brien's system, a possible "weak link" is mis-assessing how good the cars were relative to each other, leading to the drivers' performances relative to each other being mis-assessed. Comparisons between team-mates help to address this issue, but even they aren't foolproof as you can get a fair amount of variability. One good example from my childhood was that Jean Alesi was generally faster than Gerhard Berger in 1993, but slower than him in 1994, and then in 1995, he was generally slower than him in qualifying but faster than him in the races.
However, this system allows me to take into account a wide range of factors that aren't completely covered by either Patrick O'Brien's system or by complex models. Thus, it's not intended to replace any of the other systems out there, but rather offers an alternative take on it, and I am providing race-by-race tables giving how well I think drivers performed, so viewers can see some of the intricate details season by season and race by race.
Tables
The tables give each driver's rating out of 10 for each race of the season. An average rating is given for the first half of the season, for the second half, and for the season as a whole. Half season averages are given to one decimal place, full season averages to two decimal places. Bold figures denote my "Driver of the Day" for that particular race. The number of "Driver of the Day" awards serves as a tie break system for occasions when two drivers' season average scores are identical to two decimal places.
Sources
STATS-F1 - gives lap charts, results, best laps. Turbos and Tantrums - a blog giving in-depth coverage of each season and race starting in 1981 (four years before my analysis starts). Like my driver rating system, it's being updated in chronological order: as of April 2021 it covers up to and including the 1997 Belgian Grand Prix. Wikipedia list of Formula One seasons - produces season and race summaries of varying quality. grandprix.com More season/race summaries. F1 Rejects (archived) : 2001-2013 season/race summaries. Formula One Drivers and Speed Comparisons 1950-2013, from Patrick O'Brien's blog. Mostly a "sanity check", to compare my results against his. There are some differences but overall my results and his have so far been closely correlated.