## F1 2016 vs 2017: mathematical/statistical comparison

Post here all non technical related topics about Formula One. This includes race results, discussions, testing analysis etc. TV coverage and other personal questions should be in Off topic chat.
whatthefat
66
Joined: Sat Sep 13, 2014 12:02 am

### Re: 2017 F1 Pre-season testing February 27 - March 2/ March 7-10

Author of f1metrics here. I've been doing some analysis of long runs for a pre-season f1metrics post, as I did last year. I just thought I'd share some interesting results so far, as a sort of preview.

Ideally, for testing, you want to examine long runs, rather than individual lap times, if you're hoping to learn about true pace. The reason for this is that individual lap times can be set under many different configurations (tyres, fuel, set-up, weather), and can be easily masked (e.g., if drivers back off in the last sector). In long runs, teams are usually running something closer to a race simulation, which can give better insights. I've been able to obtain some long stint data both for 2016 and 2017, mostly from @f1debrief on Twitter.

The main difficulty in comparing long runs is that they may be set at different fuel loads, and the difference between a long run done on full tanks and on a last-stint fuel load can be up to about 3 seconds. By careful analysis, however, it's possible to anchor some of these stints, e.g., by finding those that were set at a start of race fuel load and working from there.

This is essentially what I did for race simulations run by Ferrari and Mercedes. I first cleaned any outlier laps out of the runs (defined as any laps that are at least 1.5 seconds slower than the lap on either side), as these are usually due to traffic or other issues. I then fuel-corrected each lap in each stint, by calculating the equivalent time on full fuel tanks. For Barcelona, I'm finding a cost of ~0.06 sec per lap of fuel, which sounds about right given it was 0.09 sec in 2013, and the cars use about 2/3 as much fuel per lap these days].

I'm not yet at the point of comparing Mercedes and Ferrari's pace, but I can already say some interesting things about tyre degradation.

The fitted lines for each tyre compounds are quadratic functions, using a least-squares fit to the data.

Some key points to note:
1. 2017 times on race stints are averaging ~4-5 seconds per lap quicker than in 2016.
2. The 2016 Soft compound tended to degrade very quickly (especially at Barcelona, which is a tyre killer), resulting in a clearly nonlinear progression in lap times.
3. Within about 5 laps, this resulted in a cross-over point between lap times on a worn 2016 Soft and a fresh 2016 Medium. This was strategically important, because it allowed drivers to pit at that point onto a fresh medium tyre to attempt an undercut. Depending on the circuit, this cross-over point tended to occur around 5-15 laps into a stint.
4. Degradation rates for both the 2017 compounds are lower and much closer to linear. Based on data so far, the 2017 Soft compound degrades just slightly quicker than the 2017 Medium -- there is nothing like the difference in wear rates between compounds seen in 2016. The 2016 Medium could become faster than the 2016 Soft after a while, due to different degradation rates. The 2017 Medium seems to start slower and stay slower, at least over the range examined so far.
5. The cross-over point between a worn 2017 Soft and a fresh 2017 Medium now comes much later, around lap 12.
6. The relative benefit for an extra pit-stop is now also much smaller, meaning we should expect fewer pit-stops (somewhat stating the obvious). As a rule of thumb, a pit-stop costs around a net 20-25 seconds these day, due to the low pit speed limits. Making that up on tyres that lose ~0.08 sec per lap in 2017 is going to be challenging compared to making that up on tyres that were losing ~0.13 sec per lap in 2016.
There will be more detailed analysis in the blog when I have everything together after the second test, including a direct comparison of tyre degradation to the 2010 Bridgestones (I found some old testing data at Barcelona, including full race simulations!).

Henne
4
Joined: Fri Jul 11, 2008 3:29 pm

### Re: 2017 F1 Pre-season testing February 27 - March 2/ March 7-10

Completed Laps overview

turbof1
Moderator
Joined: Thu Jul 19, 2012 8:36 pm
Location: MountDoom CFD Matrix

### F1 2016 vs 2017: mathematical/statistical comparison

This thread will handle data and graphics concerning 2016 and/vs 2017 data, including laptimes, speeds, and all else collectable data for both years. Data is allowed to be processed into statistical regressions for graphical representation. The target is to get statistical insight into the differences between 2016 and 2017, as well as a statistical approach towards competitive differences in 2017.

There are stringent rules:
• Discussion about the presented data and its outcome only! This is an objective topic only. Subjective posts about for instance driver preference will be met with a formal warning.
• To avoid filling up the space with complaints about the data sets, we will not allow discussion about complaints in a direct way. Rather, if there are insufficiencies, directly PM me and I will check in with the author. The goal is to keep the topic nice and clean.
• Sources of data are allowed to be shared here.
• Anything not related to the data will be met with removal of your post and a warning.
RULES WILL BE APPLIED VERY STRICTLY IN HERE[/i]
#AeroFrodo

dans79
dans79
248
Joined: Sun Mar 03, 2013 6:33 pm
Location: USA

### Re: 2017 F1 Pre-season testing February 27 - March 2/ March 7-10

whatthefat wrote: The fitted lines for each tyre compounds are quadratic functions, using a least-squares fit to the data.
Can you share the coefficient of determination for the individual fits?

To my eye it looks like the 2017 fits are better than the 2016 ones. If so, that would lend some credence to the belief that 2017 tires are more consistent.
160 96 91 6

whatthefat
66
Joined: Sat Sep 13, 2014 12:02 am

### Re: 2017 F1 Pre-season testing February 27 - March 2/ March 7-10

dans79 wrote:
whatthefat wrote: The fitted lines for each tyre compounds are quadratic functions, using a least-squares fit to the data.
Can you share the coefficient of determination for the individual fits?

To my eye it looks like the 2017 fits are better than the 2016 ones. If so, that would lend some credence to the belief that 2017 tires are more consistent.
Sure! Here's what I have:

2016 Soft: R^2 = 0.75
2016 Medium R^2 = 0.73
2017 Soft: R^2 = 0.83
2017 Medium R^2 = 0.70

Note that I have fewer data points to fit the 2017 data at the moment too.

I think it's also worth noting that if you were to just look at those data, you shouldn't really be convinced that the 2016 Soft is a nonlinear function or a quadratic necessarily. The fit could be easily influenced by a single outlier. Having said that, I did look at a lot of Pirelli 2016 stints last year (including some unpresented work), and on that basis I think a nonlinear function (e.g., quadratic) is appropriate for describing their behavior on longer stints. That's true even for harder compounds. If you look, for example, at Verstappen's final race-winning stint in the 2016 Spanish GP, the lap-times have a definite positive curvature, with some serious deg in the last few laps.

DiogoBrand
95
Joined: Thu May 14, 2015 6:02 pm
Location: Brazil

### Re: F1 2016 vs 2017: mathematical/statistical comparison

Incredible analysis! If you were able to do that without any inside information about circumstances, you must make quite a good race engineer.

Anyway. Looks like we're gonna see a lot of one stoppers, and if the harder race compounds were hardly used in 2016, I expect them to not be used at all this year. If even the softs are so linear and durable in Barcelona, I don't know why teams would even use the harder compound throughout the year.

whatthefat
66
Joined: Sat Sep 13, 2014 12:02 am

### Re: F1 2016 vs 2017: mathematical/statistical comparison

So, I've also been doing a bit of analysis to try to get a handle on where the top teams are relative to each other. The short of it is that I wouldn't be confident ranking them for now, but here's what I have.

I've plotted all the stints I have for Mercedes (GREEN), Ferrari (RED), and Red Bull (BLUE). These are all from the tweets posted by @f1debrief on Twitter (if someone has another complementary dataset, I would love to hear it, because this one is very useful but clearly not exhaustive). I have cleaned these data by removing outlier laptimes (>1.5 sec deviation from laps on either side), but otherwise it's raw, i.e., there's no fuel correction.

The thick lines are from a race simulation run by Mercedes, so they can be used as reference points.

Soft compound:

The short, very fast stints for Ferrari can probably be ignored. I assume these are mostly qualifying runs. The interesting comparison for Ferrari is their long run, which almost exactly overlaps with Mercedes' first stint race simulation and is close to another Mercedes long run. If this is indeed a first-stint simulation, then Ferrari looks very close to Mercedes on current pace.

Red Bull have three long runs, two that go for 8 laps and pretty much overlap the final race stint for Mercedes, and another that goes for 12 laps and almost a second per lap slower. One of these is likely a final race stint, but hard to say which one. It's possible the 8-lap stint was run with only 8 laps of fuel, in which case it should be ~0.7-0.9 sec quicker than a true final race stint with >22 laps of fuel. That would imply Red Bull are still a bit behind Mercedes.

Medium compound:

Ferrari again have a long run that's very similar to the stint from the Mercedes race simulation. If these are equivalent middle stints, then again it looks very close between those two teams.

Ferrari have another stint that is 22 laps long (what you would expect for a stint in a 2-stop race), but it's around 3-4 seconds slower. This is a big difference, and difficult to account for with just fuel. However, this was one of Ferrari's first long runs on day 1, so it may have been under very conservative settings with completely full tanks.

Ferrari also have an 8-lap run that is about 2 seconds faster than Mercedes' simulated middle stint. That's almost exactly the time difference you would predict if Ferrari had only ~10 laps of fuel on that run and their other run was a middle stint (~44 laps of fuel). This increases my confidence that the closely-matched Ferrari/Mercedes stints are indeed equivalents on both softs and mediums.

Red Bull have one fairly long run, which is about 1 second faster than Mercedes' simulated middle stint. It's quite variable, and really difficult to know what it is.

bjuncek
bjuncek
1
Joined: Fri Mar 03, 2017 4:05 am

### Re: F1 2016 vs 2017: mathematical/statistical comparison

Theoretical question:
How would one (if at all) account for track friction/temperature when looking at tire degradation. One would expect tires behave quite differently in BCN, as opposed to e.g. RUS.

Is there enough race data from previous years to back up any type of parameter estimation for that?

marlosb
marlosb
0
Joined: Fri Jan 30, 2015 1:10 pm
Location: Sao Paulo, SP, Brazil

### Re: 2017 F1 Pre-season testing February 27 - March 2/ March 7-10

whatthefat wrote:
Great topic, good data and analysis.
Seems like soft compound is going be the best this season.

clipsy1H
-15
Joined: Sun Feb 12, 2012 1:21 am

### Re: F1 2016 vs 2017: mathematical/statistical comparison

just one thing.... i think Pirelli will change the tires characteristic few times this year. For sure the first race soft/medium/hard will not be the same with soft/medium/hard from race 10.

But nevermind, like others said you did an amazing job with this and a great amount of work and for this you deserve a bike LIKE !!! Well done.

Wass85
Wass85
-1
Joined: Wed Mar 01, 2017 9:11 pm

### Re: F1 2016 vs 2017: mathematical/statistical comparison

Could anyone on here find the long run stats for testing in 2008? It would be very interesting to see if the 2017 cars are really the fastest they've been around the updated circuit.

PlatinumZealot
474
Joined: Thu Jun 12, 2008 2:45 am

### Re: F1 2016 vs 2017: mathematical/statistical comparison

Great work as usual. I remember that great analysis you did last year. It was spot on. Looking forward to the analysis on the second test.
Last edited by PlatinumZealot on Sat Mar 04, 2017 9:39 pm, edited 1 time in total.

dans79
dans79
248
Joined: Sun Mar 03, 2013 6:33 pm
Location: USA

### Re: 2017 F1 Pre-season testing February 27 - March 2/ March 7-10

whatthefat wrote: Sure! Here's what I have:

2016 Soft: R^2 = 0.75
2016 Medium R^2 = 0.73
2017 Soft: R^2 = 0.83
2017 Medium R^2 = 0.70
Wow, they are closer than I thought they would be. The 2017 softs is a remarkably good fit considering how random the data naturally is. I'll be interested to see how the should change as you get more data for the 2017 tires.
160 96 91 6

Nuvolari
49
Joined: Thu Apr 07, 2016 1:10 pm

### Re: F1 2016 vs 2017: mathematical/statistical comparison

Thought I'd pick up on the comments that these race sims on a colder Barcelona pre-season is not representative of the pace in-season due to differences in tyre behaviour in the summer and even development in-season. Going back to your analysis of the 2016 preseason, I found that there wasn't a considerable difference in the pace shown at the Spanish GP versus the pre-season race sims you analysed. Specifically, looking at your analysis of Ferrari:

Spanish GP race stints for Raikkonen and Vettel:

http://en.mclarenf-1.com/index.php?page ... n%20Vettel

If anything the cars were staying at the slower end of the spectrum (possibly due to worser tyre wear), so I think it increases the confidence in your analysis. Even if the Mercedes cars were running in the race, I very much doubt they'd have run miles faster than the range suggested in the preseason analysis (low 1:27s to high 1:30s). Looking forward to your full analysis.

sosic2121
sosic2121
19
Joined: Wed Jun 08, 2016 11:14 am

@whatthefat