#Top 100 models driver#
The performance model is a function of several predictors: Driver performance, Team performance, Customer car status (1950-1980), Season, Driver Age, and Driver Experience. Schematic representation of the updated f1metrics model. This is necessary because there are maximum and minimum possible scoring rates, so a linear model would not suffice. The scoring rate outcome variable is linked to the model’s performance variable via an S-shaped (sigmoidal) function. To allow comparisons across eras, a uniform scoring system is applied, which has a similar shape to the current 25-18-15-12-10-… system, but extends (exponentially decaying) beyond 10th place to allow performance differences between lower performing cars to be easily discriminated. Counting races include races where a driver finished, as well as races where a driver had a DNF for other reasons (e.g., they crashed, they gave up, or they had a DSQ on driver-conduct grounds). A counting race is defined as any race in which a driver did not experience a non-driver DNF (e.g., a mechanical DNF or a DSQ on mechanical grounds). To briefly summarize, the outcome that the model attempts to predict is a driver’s scoring rate in each season (points per race, or ppr for short) in their counting races. The statistical model used for generating driver rankings has been described in some detail in a recent post, where I documented the latest upgrades. I know this has been a highly anticipated and requested post.
#Top 100 models update#
Having made a series of upgrades to the model over the past 5 years, I feel it is time for a major update to the all-time rankings. There were fairly obvious steps available to improving the model, including the incorporation of age and experience effects, although the implementation was not altogether straigtforward.
#Top 100 models drivers#
The model I used at that time was fairly simplistic, accounting for only driver performance, team performance, and competition with other drivers in the same season. Another advantage is that models can be used to identify and challenge subjective narratives that do not really have any basis in fact.įive years ago, I used such a model to generate a top 60 all-time ranking list, which remains the second most viewed article on this blog, behind only my post explaining the rules of racing. One advantage of such an approach is that the results, and how they depend on the model itself, can be directly scrutinized. This approach is to use an objective statistical model to attempt to separate driver performance from car performance and other key factors. However, subjective rankings are prone to bias and dependent on each individual’s subjective mental model, shaped by the races they have or have not seen or analyzed.Īn alternative approach to driver rankings has developed over the past several years. Witness the fact that Stirling Moss (no world titles) is often placed near the top of such lists. Most subjective ranking lists already do this to some degree. If our objective is to rate driver performances, then we need to somehow make allowances or corrections for a driver’s car competitiveness. Some of the most talented drivers in the sport’s history won no titles, simply because they never raced in a sufficiently competitive car. To put it another way, F1 is not a level playing ground for drivers. As far as team performance goes, the driver is a relatively small contributor.
Yet, as F1 fans, we recognize that the question is far more complicated.į1 is a team sport. If the only aspect we care about is the number of titles or grand prix wins, then the answer is straightforward. Who was the greatest F1 driver of all time? It is an endlessly debated question amongst fans and pundits alike.
0 kommentar(er)
