Post by News
FASTEST F1 DRIVERS SINCE 1983
Just my view on this as someone who has worked with IT for three
decades and work with researchers on Data Analytics, including using
As with so many studies, the devil is in the detail. This particular
study hasn't really been fully explained in the cnet article, so it's
hard to come to firm conclusions either on what they were trying to
analyse, let alone (precisely) how they did so.
There are clearly some challenges in making inter-generational
comparisons (which is why we go round this circle so often), and there
are various ways to _try_ to resolve them.
Similarly, you have to remember that most ML models and tools are
designed to spot patterns (often those that elude humans), but it's down
to other (typically human-based) processes to refine the results to sort
the correlation from the causation. Simply throwing the data at ML
tools will give you interesting data (and I'm not saying that this is
all that they've done, but they might have), but it doesn't necessarily
give you information or tell you what you think it does.
It looks as though the key elements here are:
- Various data points relating to raw speed in qualifying
- Relationships between overlapping careers of the different drivers
Right off the bat there are some potential issues that may (or may not)
have been ignored but which would have significant impact on the result
of the model:
1. Qualifying has changed in lots of different ways over the years:
- The nature of the session(s) themselves (single laps versus traffic,
knock-outs, length of session, etc.)
- The tyres available and how they can be warmed or generally played
- Fuelling strategies and fuel types (particularly before homologation)
- The cars (and configurations) available
- The tracks involved
- Many other factors which I'm sure the more knowledgeable in the group
know better than me
2. Implicit (I think) in the intergenerational link is an assumption
that it's possible to say that if A is faster than B and (later) B is
faster than C that you can assume that A is faster than C.
Okay, with (1) I guess the assumption is that everyone in a given season
faces the same challenges, but I don't think that necessarily evens-out
(particularly once that data is compared season-to-season) and a small
error-bar here could multiply up over time.
With (2), I think they face some significant issues.
- First, I think that over time you have to accept that other factors
affect performance and even if all other things (car, tracks, tyres,
refuelling, etc.) were the same over several years, drivers improve
and get worse (whether because of physical or psychological reasons,
because of experience, through loss of confidence, whatever) so if B
was on a bad year when facing A but had considerably developed when
they faced C, there is no way to know if A is better than C. The more
time that passes, the harder that link is to make.
- Second, all things aren't the same. You think of things like the
Active Suspension in the early 90s for Williams, the Benetton
electronics, the various dodgy ducts a decade ago, let alone the
question hanging over Ferrari over last year's engine. How do you
have a solid link between years across drivers and technologies?
Maybe they've done a very solid job in the study to try to allow for
these (and other) distorting factors, but that's not clear in the
report. If they haven't, it's an interesting bit of pattern-matching in
the data, but I don't think it tells us anything.
Of course, as a trigger for (even more) debate...