Hamilton unlucky or not: how about a statistic model.

[turbof1]

you guys have short memory .
after all the restructuring we still have that team which was BAR some time ago ...and tell you something they had really worrying Stretches of only one Driver being affected with horrible reliability issues.
I remember Takuma Sato with such a bad string of results and also Michael Schumacher who had this 2011/2 years which was nothing short of outright horrible from a Team Performance side ..with missed dead lines to make a quali attempt to DRS failures ,lost wheels and and and ....
maybe one is a bit biased when it comes to counting the mishaps on one side of the Garage but it seems as if Lewis now has inherited Schumis luck in this Team -interestingly it was not like that last year at all..

Excuse me, can you explain what kind of pseudo scientific thing "inheriting" luck is? What's the correlation between whatever happened to one of Mercedes drivers during previous seasons and 2014 Australia ENGINE problem (new, different engine), Hungary ENGINE problem (new, different engine), Canada ENGINE/ERS problem, Canada brakes problem (new brakes), Germany brakes problems of specific brand (new different brakes).

First thing with assessing "bad luck" would be to check if there's bad luck at all.

Which we intent to do here. It's the purpose the topic creator intended.

[SidSidney]

Try as you might, you cannot have a valid statistical model for this.

Here's why.

Unless you can factor in how much, or how little, mechanical sympathy Lewis has when driving his cars, you will never be able to create any model. Keep in mind, you're also at the whim of whatever Mercedes claims is the cause of a mechanical failure. They may never announce what it actually is lest it point to a component that could be integral to their ability to field the car that they currently do.

For example, Nico was able to nurse his W05 home at Canada while Lewis was not able to do it. Does that speak to a lack of ability on his part to nurse failing equipment? Unless you can definitive answer to that, trying to create a statistical model is an exercise in futility as there are too many unknown factors.

The scenario above could be equally described as Lewis getting unlucky or Nico getting lucky. I would really like to move byond individual events and look at the overarching trend.

I just don't think that fluffy qualititative stuff or the detail of how the drivers did a or b or c matters at this stage. What matters is simply the output of the process, which is ultimate measured in championship points or something equally tangible.

We are trying to determine if someone has above or below average luck against others doing the same thing, perhaps with differnt input variables, but that is the whole point of a snapshot Monte Carlo simulation, which is an absolutely standard investigation tool in all walks of life.

So the first step is to determine a baseline curve of outcomes for a suitable pool of drivers for some tbd statistic, then compare Lewis Hamilton's outcome for the same statistic.

Questions:

1) Suitable pool of drivers? We have to weed out the crappy cars, so maybe only the top 4 drivers in last 30 championships?
2) Statistic choice? Points scored from available points? DNF per race? Points per race? Ideas?

I don't know. I can do the mechnics of the calulation but need some ideas on what is a good measure of performance.

Honestly, if we do a good job and find something interesting it will be newsworthy, as I think TurboF1 realizes.

[GitanesBlondes]

Try as you might, you cannot have a valid statistical model for this.

Here's why.

Unless you can factor in how much, or how little, mechanical sympathy Lewis has when driving his cars, you will never be able to create any model. Keep in mind, you're also at the whim of whatever Mercedes claims is the cause of a mechanical failure. They may never announce what it actually is lest it point to a component that could be integral to their ability to field the car that they currently do.

For example, Nico was able to nurse his W05 home at Canada while Lewis was not able to do it. Does that speak to a lack of ability on his part to nurse failing equipment? Unless you can definitive answer to that, trying to create a statistical model is an exercise in futility as there are too many unknown factors.

The scenario above could be equally described as Lewis getting unlucky or Nico getting lucky. I would really like to move byond individual events and look at the overarching trend.

I just don't think that fluffy qualititative stuff or the detail of how the drivers did a or b or c matters at this stage. What matters is simply the output of the process, which is ultimate measured in championship points or something equally tangible.

We are trying to determine if someone has above or below average luck against others doing the same thing, perhaps with differnt input variables, but that is the whole point of a snapshot Monte Carlo simulation, which is an absolutely standard investigation tool in all walks of life.

So the first step is to determine a baseline curve of outcomes for a suitable pool of drivers for some tbd statistic, then compare Lewis Hamilton's outcome for the same statistic.

Questions:

1) Suitable pool of drivers? We have to weed out the crappy cars, so maybe only the top 4 drivers in last 30 championships?
2) Statistic choice? Points scored from available points? DNF per race? Points per race? Ideas?

I don't know. I can do the mechnics of the calulation but need some ideas on what is a good measure of performance.

Honestly, if we do a good job and find something interesting it will be newsworthy, as I think TurboF1 realizes.

Again you would be creating a totally flawed model because there is too much unknown data that you cannot find out about what goes on. There is simply too much information you will never have access to, that would be needed to attempt such a model. Not sure why you're not understanding this.

Edit: Also need to clarify, the biggest problem here also assumes that both drivers have access to 100% equal equipment. Good luck trying to get a definitive answer on that.

[bhall II]

Could you expand what you mean by that exactly, what Bonnigton and Clear have to do with today's engine problem for example? Or Australia engine problem? Or Canada brakes problem? Or Canada ERS engine/ERS problems, or Germany Brembo brakes failure.

BTW: if one to believe Wikipedia: Clear "previously he was race engineer for Nico Rosberg from (2010 to 2012)". What kind of a correlation can there be

Unless you're looking toward voodoo for answers, I think the implication is pretty clear.

If two different drivers are beset with mechanical and strategic errors over two different time frames, but with the same two engineers in charge of mechanical integrity and strategy, I'd say it's probably a good idea to look at the common link between the two drivers, yanno? Otherwise, you'll be left with your head in the sand, trying to ascribe to "luck" what is really the consequence of varying levels of vigilance.

[turbof1]

Try as you might, you cannot have a valid statistical model for this.

Here's why.

Unless you can factor in how much, or how little, mechanical sympathy Lewis has when driving his cars, you will never be able to create any model. Keep in mind, you're also at the whim of whatever Mercedes claims is the cause of a mechanical failure. They may never announce what it actually is lest it point to a component that could be integral to their ability to field the car that they currently do.

For example, Nico was able to nurse his W05 home at Canada while Lewis was not able to do it. Does that speak to a lack of ability on his part to nurse failing equipment? Unless you can definitive answer to that, trying to create a statistical model is an exercise in futility as there are too many unknown factors.

The scenario above could be equally described as Lewis getting unlucky or Nico getting lucky. I would really like to move byond individual events and look at the overarching trend.

I just don't think that fluffy qualititative stuff or the detail of how the drivers did a or b or c matters at this stage. What matters is simply the output of the process, which is ultimate measured in championship points or something equally tangible.

We are trying to determine if someone has above or below average luck against others doing the same thing, perhaps with differnt input variables, but that is the whole point of a snapshot Monte Carlo simulation, which is an absolutely standard investigation tool in all walks of life.

So the first step is to determine a baseline curve of outcomes for a suitable pool of drivers for some tbd statistic, then compare Lewis Hamilton's outcome for the same statistic.

Questions:

1) Suitable pool of drivers? We have to weed out the crappy cars, so maybe only the top 4 drivers in last 30 championships?
2) Statistic choice? Points scored from available points? DNF per race? Points per race? Ideas?

I don't know. I can do the mechnics of the calulation but need some ideas on what is a good measure of performance.

Honestly, if we do a good job and find something interesting it will be newsworthy, as I think TurboF1 realizes.

Again you would be creating a totally flawed model because there is too much unknown data that you cannot find out about what goes on. There is simply too much information you will never have access to, that would be needed to attempt such a model. Not sure why you're not understanding this.

Edit: Also need to clarify, the biggest problem here also assumes that both drivers have access to 100% equal equipment. Good luck trying to get a definitive answer on that.

Again, let's try it and see later where the flaws are. Yes unknown data always is an issue, but it doesn't hurt either to have something simplified now and add other variables later on to see if we could get a bit closer to the truth. We'll see.

[marcush.]

ah statistics ..
please remember for statistical calculations the number of samples is a very important figure.
with 20 or so races per year i doubt you can make up any reasonable calculations ...but why not have a try and see what is the outcome.


Satos season with BAR in 2005-cluttered with own midfortune but also horrible Slips by the Team -like putting rediculous amounts of fuel into the car... :
http://www.f1network.net/main/s169/st86061.htm?print=1

as for inheriting bad luck:

I do not Claim it to be scientific - it´s not at all! it´s a human trait -and it´s made by humans who fail to live up to expectations .People make mistakes this is inevitable.The big issue is when you fail to realise you are the one who might have committed a mistake .We have seen this Happening again and again .

You call it bad luck if hamiltons car Ends in a blaze of fire after not even completing a single lap ...and one race before a brake rotor failure with almost zero mileage on it.
But thats quite impossible .Can you please Elaborate how a part can fail within minutes when the System has proven to be reliable in that area for considerable time before and also with the other car under nearly the same conditions? That´s clearly a human factor spitting into your soup here .Be it a Thing of negligeance or a case of wrong handling (lack of knowledge) playing a part in it.
I know a guy from Tilton who supplied Carbon clutches to everyone and his dog in european (also works) motorsports and he once said : you would not believe it but as soon as a Company or Team starts using carbon clutches you can bet you will have three clutches destroyed before you can start working properly with the Team.
The first one is destroyed by the mechanic who had no working instructions ,the second one by the engineer who gave wrong instructions to the mechanic for the second one and the third one is destroyed by the Team Manager who knows even more and advises how to handle the product.
Only then the Team Comes back to the supplier bad mouthing the rubbish part but then you can start Training the guys on the product ..luck anyone ?

[SidSidney]

Again you would be creating a totally flawed model because there is too much unknown data that you cannot find out about what goes on. There is simply too much information you will never have access to, that would be needed to attempt such a model. Not sure why you're not understanding this.

Edit: Also need to clarify, the biggest problem here also assumes that both drivers have access to 100% equal equipment. Good luck trying to get a definitive answer on that.

It doesn't matter. None of it matters. In the end you get a WDC for points scored, not feelings or being clever with tyres or whatever.

You absolutely do not need to know what goes on inside a machine to measure variability in the output of that machine from a designed specification. You just compare the output to the spec and you know if it is off or not, regardless of what goes on inside the machine or process. Also, all SPC methods like Six Sigma assume every part, even those supposedly identical, is different, so the whole 100% equilvalence argument is completely baseless.

All I want to see is, does the output of Hamilton's process vary signifcantly from others in a negatively skewed way, or is it balanced between positive and negative effects. If it is skewed he may be lucky or unlucky, depending on which way it falls. If it is equally balanced he is by definition neither lucky or unlucky. If he has a broader distribution than others ie., a higher standard deviation, then it would indicate a process issue. If he has a tighter distribution he has a better process.

That's it. Why it is like it is ie., the root cuase, is another level of detail, the one you allude to, and I just don't care about that. I just want to see if pure random chance is really a factor, and I think that it should be measureable. After all they do it in baseball without knowing what bat weight someone uses, what weights they are lifting, what size boot they chose.... Nate Silver makes a good living out of it.

[turbof1]

That's it. Why it is like it is ie., the root cuase, is another level of detail, the one you allude to, and I just don't care about that. I just want to see if pure random chance is really a factor, and I think that it should be measureable. After all they do it in baseball without knowing what bat weight someone uses, what weights they are lifting, what size boot they chose.... Nate Silver makes a good living out of it.

I think is this the main thing: the probability of it being a serie of random problems or being something else. I'm not talking about conspiracy theories here, but just if it could be 'something else', a structural problem.

[ringo]

The statistic must include competitive cars only; those that can win a race. Secondly it must be limited to points positions, podiums, wins and qualifying. These factors better express bad luck since there is something to gain.

[marcush.]

do you really think it is necessary to produce a statistical model for this simple plot?
Why not just list the simple Count of :dnfs ,mechanical failures ,Driver Errors ,Crew Errors ,and Errors which can be attributed to whatever ... Ithink you will soon realise there is a pattern to it no calculations needed just collecting the numbers and assigning them to a cause or a combination of icauses..

So a generally slower pitstop on Lewis side could be a Problem on Hamiltons Crew (but it is the same guys performing the wheelchange)or Hamilton placing the car less precise than Rosberg ,or whatever.

[Shrieker]

It could simply have something to do with Nico pushing the car less coming into the pits, and the temperatures in the rim/nuts are different to Hamilton's, or vice versa. It could have something to do with the mechanics liking Nico more and therefore are more motivated when servicing him. You never know these things for sure.

BUT ! In a professional environment, sooner or later someone will be pulling their ears if one driver is almost always getting faster pit stops for no explicable technical reason.

[GitanesBlondes]

You guys need to listen to what marcush is saying. He knows far more than a lot of people here about this stuff.

[bhall II]

The statistic must include competitive cars only; those that can win a race. Secondly it must be limited to points positions, podiums, wins and qualifying. These factors better express bad luck since there is something to gain.

That's sample bias. You can't just discard statistics merely because they don't fit within some arbitrarily idealized model of what it means to have "bad luck." For instance, Bianchi would be thrilled with a tenth-place finish, but Hamilton would undoubtedly consider the same result a bad day at the office.

This is racing: everything is relative.

The only reasonable way to do this that even approaches objectivity is to do it completely. That means someone is going to have to trudge through every result for every driver since 2007 and then compare those rates of "bad luck" to Hamilton's rate of "bad luck."

Even better would be comparing Hamilton's rate of "bad luck" with those of every driver his teams have ever fielded, because team culture tends to be the root of all failure ("bad luck"), be it mechanical, strategic, or otherwise. But, I seriously doubt anyone would undertake such an investigation.

I think it's all a fool's errand, anyway.

EDITS: clarity

[turbof1]

Even better would be comparing Hamilton's rate of "bad luck" with those from every driver his teams have ever fielded, because team culture tends to be the root of all failure. But, I seriously doubt anyone would undertake such an investigation.

I think that's the best thing to do: go through several generations and take of every generations an equal sample size. The total sample size should be very large, like n=200. You should include random drivers from random teams. A technical failure is a technical failure; it doesn't matter if you are in a points position, a win position or at the back of the grid. Then test Hamilton's samples to it.

Btw, it's probably easier then it looks. skim through wikipedia entries for a few hours and you'll probably have it.

[ringo]

The statistic must include competitive cars only; those that can win a race. Secondly it must be limited to points positions, podiums, wins and qualifying. These factors better express bad luck since there is something to gain.

That's sample bias. You can't just discard statistics merely because they don't fit within some arbitrarily idealized model of what it means to have "bad luck." For instance, Bianchi would be thrilled with a tenth-place finish, but Hamilton would undoubtedly consider the same result a bad day at the office.

This is racing: everything is relative.

The only reasonable way to do this that even approaches objectivity is to do it completely. That means someone is going to have to trudge through every result for every driver since 2007 and then compare those rates of "bad luck" to Hamilton's rate of "bad luck."

Even better would be comparing Hamilton's rate of "bad luck" with those of every driver his teams have ever fielded, because team culture tends to be the root of all failure ("bad luck"), be it mechanical, strategic, or otherwise. But, I seriously doubt anyone would undertake such an investigation.

I think it's all a fool's errand, anyway.

EDITS: clarity

It would not be biased, because that is the information that we want to be presented. It would be a waste of your time to group every dnf that every happened, as it will not tell you what was at stake in each case. You would have to further process that information. Bad luck in Lewis' context or any other driver, is if they were expecting a good result and then the dnf befell them. A driver like Sutil languishing at the back then his car taking itself out of it's misery is not necessarily veiwed as bad luck.