LVDell

Guys, this is not worth even discussing. The OP (which he and I have discussed this at great length offline several weeks ago) is looking to do something in the form of a self-report survey for car owners and informed the forum about it. It should have been left at that. More of an announcement if you will.

Whether it's a disgruntled owner wanting to feel "misery loves company" or a highly trained academic in statistics (like myself and the OP) trying to collect data, the simple fact is that the responses in this thread are the same as all the other threads........over and over and over again. It's going to get nowhere.

As one who enjoys crunching the numbers as much as the next stats geek, we are going to get nothing from this thread (like all the ones before). IMHO, this thread will do nothing for either side of the debate. The OP and I will have to agree to disagree on the validity of the study and the significance of any conclusion drawn from it. Nonetheless, I wish him well in his endeavor.

BruceP

Always fun when things get personal.

Anyway, since you don't know me, you can be forgiven for the absurd irony of that statement. I've created quite a lot of things, actually, including some things you've probably heard of. And I can tell you that if you're going to make a living 'creating things', you have to expect to defend your ideas sometimes. I don't care if your name is Spielberg, it goes with the territory.

mkaresh

It wasn't my intent to make things personal. I truly do not doubt your expertise.

The problem with defending what I do in a public forum is that the more detailed responses in such a debate get, the fewer people sign up to participate. Since the quality of my results is heavily based on the number of participants, such a debate ends up being a no-win situation for me. But this thread has been rendered ineffective for me at this point--there now aren't likely to be stats for the 996 any time soon, mission accomplished for anyone with this objective--so might as well debate on.

Here's an example of my results, for the Audi A4. Stats in terms of repair trips per 100 cars per year (lower is better):

2010: 30

2009: 64

2008: 48

2007: 48

2005.5-
2006: 78

2005: 154

2004: 135

2003: 119

2002: 159

Is there more information here than can be found in a steaming pile of BS? For me, a lot more.

I see a very clear difference between the B6 (2002-2005) and later generations. I also see higher repair rates for the first model year of each generation (2002, 2005.5-2006, 2009). The 2005 is a little higher than might be expected, until you realize that these cars were in the last year of their standard warranty--when people rush to get minor repairs done that they've been putting off. I see this "final year of warranty" blip often, especially for German cars.

I recently started providing a second set of stats that reports, unlike anyone else, the percentage of cars with no repairs in the past year, and the percentage with 3+ repair trips. These stats, especially the latter, require larger sample sizes than the average repair frequency, so I have them for fewer cars and advise against reading too much into the "lemon-odds" until we have more people involved.

The percentage of cars with no repairs during 2009 from the first eight cases (alphabetically, not cherry picking) where I have these stats for at least two consecutive model years:

Acura MDX
2008: 70
2007: 53 (first year)

Acura TL
2005: 84
2004: 75 (first year)

Audi A3
2007: 64
2006: 31 (first year, very common failure with sunroof sunshade clip)

Audi A4
2009: 52 (first year)
2007: 71
2005.5-
2006: 50 (first year)

BMW 3-Series
2008 335: 53 (common problem with fuel pump)
2007 335: 54 (common problem with fuel pump)
2007 328: 63
2006 E90: 53 (first year, no 335)
2004: 48
2001: 8 (worst among cars for which we have this stat)

Buick Enclave / GMC Acadia / Saturn Outlook / Chevrolet Traverse
2009: 74
2008: 53
2007: 41 (first year)

Chrysler 300 / Dodge Charger / Dodge Magnum
2006: 54
2005: 46 (first year)

Ford Five Hundred / Freestyle
2007: 65
2006: 64
2005: 44 (first year)

In all eight cases the stats form a clear pattern, with a lower percentage of cars with no repairs as cars get older, and especially for the first year in a design. Given the sample sizes, small in most cases, I'm been pleasantly surprised that the results are this clean. I would not expect most of these to be accurate within fewer than 10 points, but we're not seeing unexpected variation of more than a few points here.

So, again, no more than a steaming pile of BS?

Wellardmac

Yes, absolutely a steaming pile of BS. You're trying to extrapolate the results of double or low triple digit responses from a self-selected population to cars that are mass produced in hundreds of thousands of units per year.

This is not personal, we (and I know others have also tried and failed) are trying to get through to you that the methodology you're using is grossly flawed.

I know that you don't get it and I know that there's no changing your mind, so it's time to move on. I'm not a big fan of wasting my words where there is no chance of reconciliation - It's way more fun and effective to bang my head against a wall.

mkaresh

Why do you feel the need to be so insulting? Is this junior high? Edit: looks like Wellardmac's references to my being "thick-headed" and "moronic" have been removed. Thanks, I suppose.

As Bruce can no doubt confirm, when the size of the population is large it has absolutely no bearing on the required sample size. Population size isn't even part of the formula for determining the required sample size. This is stats 101. It's what makes survey research viable--samples are often a very small percentage of the total population.

What does matter: the variance. Which is why I initially focused on a metric with a low variance.

It's not relevant, but I might point out that few car models are produced in the hundreds of thousands per year. And certainly no Porsche is. Tens of thousands, certainly.

I'm aware that the source of my participants is far from ideal. Would I prefer a large, randomly selected sample with a high response rate? Sure. But no one has this, and there's a reason for it. Hard to do even if you have millions to spend, impossible if you need to have low enough costs to provide results to the general public.

J.D. Power spends a large amount of money to randomly survey car owners. In return they:
--have a target response rate of 23%, leaving them with a high risk of non-response bias, reducing the value of the random sample (the responses they receive are probably not random, even if the people asked to respond are)
--have a minimum sample size goal of 225, good, but not as large as many people think
--can survey vehicles at only two time points, 90 days and the third year
--cannot afford to provide a useful level of detail to the general public

I have found that useful results are quite possible even with the sample I do have. And I vastly prefer the results I have been able to provide to no results at all.

Sneaky Pete

Oh Christ..........

pweishaupt

Fascinating.

NIACAL4NIA

Have you ever noticed why antibacterial soap kills only 99.9% bacterias? Because the 0.1% helps the lawyers win the case. Scaremongering sells. My2002 996 6speed 84000 miles with original clutch and no rms/ims or major problem, so far sensors and a cracked coolant tank for a total of $200 DIY parts. My cousin's 2003 Mercedes SL500 87000 miles has over $20000 in repairs for leaky ABC shocks, flex plate, deff leak, hard top air pump, MAF, brakes, Alarm....etc

KoB

Nice math.

mkaresh

Small sample sizes. But I bet a good chunk of that $20k was for the ABC. Suspension parts wear out, and the more complicated the suspension is the more likely it is to fail and the more expensive it will be to fix. Mercedes's ABC system is about as complicated and as expensive as they get.

Unless the fancy bits make a huge difference to the car's handling, I always recommend conventional springs and dampers.

DBJoe996

I surprised someone hasn't said it yet, "There are lies. Then there are damn lies. Then there are statistics."

I rather like my dependable 99 daily driver w/130,000 miles. Not perfect and has busted a few parts, but hey...it's a car. Fix it and move on. I don't need statistics for that and I'm not going to use it. Now for another beer and back to the regular programming. Think I'll head to the garage and check my blinker fluid.....