Hamilton unlucky or not: how about a statistic model.

[SidSidney]

OK, I know this is going to attract the usual noise, but here's a serious question:

is Lewis Hamilton genuinely unlucky*, or is it just an impression?

Can we >>mathematically prove<< that what is happening to him in this season (and also in others given the sample size is so small in one season) is outside normal expected outcomes of a purely random distribution?

Please if you just have the impression or opinion he's the best driver ever and the unluckiest, keep that to one side - I'd really like to see numbers.

As a contribution, I am prepared to do the work to produce the basic standard deviations based on his results, but we need a comparison framework so this gets nailed once and for all.

*OED defines luck as "success or failure apparently brought by chance rather than through one’s own actions"

Some edits from the moderator:

so some ground rules...

*the intent behind this topic is to have a statistic model to see if Hamilton is either hit with bad luck or if there's something different going on.

*Some side talk is ok, but the priority is to get a pool of data, calculations around variances and standard deviations, and a distribution up.

*We understand that there are huge amount of factors that have influence on this, but getting a model up isn't going to hurt and might reveal some interesting things. Interpretations about it can always be made AFTER the model is up.

*No fanboyism and no complaining about fanboyism allowed. It's a number's topic. Any sort of idiocity inmediately warrants a warning.

*IF, and a big IF, this produces something meaningful and everybody cooperates, I'll write a front page headline, with all participant's efforts mentioned.

[JimClarkFan]

Overall, I think he has had more than his fair share of bad luck.

I'm not skillful to get into the mathematics, however there are a couple of things worth remembering that may have had knockon affects.

He has different brakes from Rosberg, so Canada and Germany may have been the result of a choice to have different brakes. This weeks fire, could have been a result of a rebuild problem from last weeks crash, which was a result of a brake failure- is that luck or is that the result of a poor choice of brake manufacturer?

[ScottB]

Over the years, we've seen a number of teams have the lion's share of 'bad luck' land predominantly on one driver:

Red Bull, Webber invariably got the mechanical gremlins while Seb was fine. This year he's had far more problems than Ricciardo.

Mclaren, I can recall DC retiring a lot more often than Mika did during their dominant years.

And now Hamilton at Merc.


Obviously, not suggesting a conspiracy at any of these teams, but certainly with Hamilton, he's had almost all the mechanical problems, seems to consistently have longer pit stops, plus general 'bad luck.'

I suppose it could be a case of an initial problem, be it a retirement or mistake that causes that side of the garage to fall behind, sees them pushing to catch up, taking risks and the like that can lead to more mistakes, component failures etc?

[Juzh]

Red Bull, Webber invariably got the mechanical gremlins while Seb was fine.

In 2013, and even that one is tight.

[SidSidney]

He has different brakes from Rosberg, so Canada and Germany may have been the result of a choice to have different brakes. This weeks fire, could have been a result of a rebuild problem from last weeks crash, which was a result of a brake failure- is that luck or is that the result of a poor choice of brake manufacturer?

OK this is why we need an agreed comparison framework. How can we objectively judge expected vs. unexpected outcomes? What are the units to measure luck / skill / expected outcome / something outside normal expectation?

My view is that the only way to accurately judge outcome is by the points a driver is awarded for results. How he and his team chose to get those results are irrelevant in the final analysis. There may be other views, like starting grid position, poles, pole conversion, whatever, and we need to plot some distributions.

But it has to be solid public data not "Hamilton is just unlucky because he chose x while Rosberg chose y" - no. How they got the results is irrelevant; we have to assume that these are great racing drivers in top racing cars, and that they are expert professionals who know what they are doing. If it is really luck then it should be equally balanced between good and bad in the long run. So all that matter is what results they were able to generate from that baseline.

If we see that Hamilton is far outside the norm on negative results without balancing success, then we have to explain that as either luck or lack of skill - I am not sure how to work that out yet but there is some brainpower on this forum somewhere .

If his bad outcomes are balanced by good ones, but the standard deviation is wider than others then we cna probably say that he is not unlucky, but there is something in the process that is probably within his / his team's control and going awry.

I really don't want a fistfight, I'd like to have a really straightforward data-based analysis. Is that possible?

[turbof1]

While I have a bit of an issue with upvotes for posts that really aren't contributing to the topic, I feel for now that the topic itself, courtesy to the way Sidsidney argumentated it, definitely has its place here.

So a statistic model has been proposed, which btw fits in with being technical. If the topic starts working with numbers and statistic distributions, I think fanboyism will be scared away.

Sidsidney, are you thinking of a normal distribution or a different one? If so, what would be the test sample? Don't forget t-tests work better when n>30.

[gray41]

I think it could be argued both ways, how about Hamilton enjoying all the glory at Silverstone?

F1 works in mysterious ways, moments like today add to the drama and story of the season and I think that is why it has so many fans.

[monsi]

I really don't want a fistfight, I'd like to have a really straightforward data-based analysis. Is that possible?

No. Or at least not without making gross simplifying assumptions.

In reality F1 cars are complex systems, run close to the limit. There is so much as outsiders we do not know.

1) Unknown fault tree. We don't know what all the component parts are let alone how they interact. 2) unknown individual component behaviour to model 3) unknown variables affecting any component - cycles of use etc.

Some things could be estimated from detailed knowledge of similar cars, if you had detailed inside knowledge, which I certainly don't. Even then there is so much unknown that I can't see any sensible analysis being likely. Of course if someone here does have access to a detailed model it would be fun to see what it says

Data only takes you so far. F1 cars are rolling prototypes. Lots of part and design changes. Failure profiles for changed components are going to have a lot of estimation and extrapolations from similar parts.

It would be more informative I think if we knew if there were differences in how close to the limit Hamilton's and Rosberg's cars are being run. Is Hamilton's being run to a higher risk profile in order to maximise performance ? Are we seeing knock-on errors because of enforced repair work ? Team too tired ? Anyone got a line to the inside ...

[matt_b]

Last year Lewis had 0 mechanical failures in all qualifying sessions and races, that is 0/38. Nico had 3 during the race last year including an engine failure here in Hungary. There is certainly no conspiracy, its motor racing and motor racing can be cruel, just ask Felipe when he was a few laps away from winning Hungary 2008.

[SiLo]

If someone were that way inclined they could get a table of DNF's per season for each driver, and a lot of drivers that have retired as well. Then maybe we can see which drivers had the most bad luck/caused the most problems and go from there?

Just a suggestion.

[SidSidney]

No. Or at least not without making gross simplifying assumptions.
In reality F1 cars are complex systems, run close to the limit. There is so much as outsiders we do not know.

Well let me give you an example from my posh business school course. We all had a large packet of Smarties, the outputs of a large, complex process. We counted the colours of the Smarties in a few hundred packets. From that sample we were able to show that there were more blue Smarties than expected, assuming normal distribution. This being a posh b-school they then brought in the CEO of Nestle' to discuss operational process control, Six Sigma, issues in running a global business etc.

My point: the output of a large, complex process is enough to indicate statistical imbalances within the process. This is the whole point of statistical process control. And I think that is all we need to say definitively, mathematically, if Hamilton is unlucky or has something else going on in his life that makes him unsuccessful. It just has to be framed correctly.

[turbof1]

I really don't want a fistfight, I'd like to have a really straightforward data-based analysis. Is that possible?

No. Or at least not without making gross simplifying assumptions.

In reality F1 cars are complex systems, run close to the limit. There is so much as outsiders we do not know.

1) Unknown fault tree. We don't know what all the component parts are let alone how they interact. 2) unknown individual component behaviour to model 3) unknown variables affecting any component - cycles of use etc.

Some things could be estimated from detailed knowledge of similar cars, if you had detailed inside knowledge, which I certainly don't. Even then there is so much unknown that I can't see any sensible analysis being likely. Of course if someone here does have access to a detailed model it would be fun to see what it says

Data only takes you so far. F1 cars are rolling prototypes. Lots of part and design changes. Failure profiles for changed components are going to have a lot of estimation and extrapolations from similar parts.

It would be more informative I think if we knew if there were differences in how close to the limit Hamilton's and Rosberg's cars are being run. Is Hamilton's being run to a higher risk profile in order to maximise performance ? Are we seeing knock-on errors because of enforced repair work ? Team too tired ? Anyone got a line to the inside ...

Monsi, it never hurts to gather and process data. How we'll interpret it, is an issue after it was processed in the first place.

[SectorOne]

Well let me give you an example from my posh business school course. We all had a large packet of Smarties, the outputs of a large, complex process. We counted the colours of the Smarties in a few hundred packets. From that sample we were able to show that there were more blue Smarties than expected, assuming normal distribution. This being a posh b-school they then brought in the CEO of Nestle' to discuss operational process control, Six Sigma, issues in running a global business etc.

My point: the output of a large, complex process is enough to indicate statistical imbalances within the process. This is the whole point of statistical process control. And I think that is all we need to say definitively, mathematically, if Hamilton is unlucky or has something else going on in his life that makes him unsuccessful. It just has to be framed correctly.

But F1 is not a Smarties-valve that does the same thing day in day out.
With that you have a fairly controllable thing to take measurements from to which you then can find out why there´s more blue ones then other ones or if the blue ones weigh more and find themselves lower down then the other colors.

It can´t be compared to something so complex and random as Formula 1 because there´s a billion more factors to take into account then simply 5 colors in a machine.

[SidSidney]

Sidsidney, are you thinking of a normal distribution or a different one? If so, what would be the test sample? Don't forget t-tests work better when n>30.

I was thinking normal dist initially, as luck should be normally distributed if truly random ie., when flipping a coin, in the long run, you are just as likely to get lucky as unlucky.

If we see a skew or wider SD spread in the data then this indicates something abnormal (literally). I was thinking that to remove the variable of the team/car/engineering conspiracy/whatever it has to be multiple seasons; I think it is fair to assume an "unlucky" driver was "unlucky" everywhere he drove?

And for comparison you need to take another set of drivers - say 20-25 to act as the baseline, to see if Hamilton's data is similar or different to their individual and average performance. Could also be from different eras, as again you would assume luck is truly random and independent of era.

That would be the basic analysis. I am not sure yet what to plot, any ideas?

[monsi]

Monsi, it never hurts to gather and process data. How we'll interpret it, is an issue after it was processed in the first place.

Fair enough. Agree entirely. I guess I am thinking of random rare events at the root cause level, hence (perhaps) poisson distribution types of tests, and six sigma type techniques when things get too infrequent.