Guys, several of you make very good points. I didn't mean to start this thread to bash Toyota. Heck, I own a Toyota and I actually like Toyotas. I'd probably put my mom in a new Camry if she was buying a new car. But if you've followed the history of this board, there is a ton of VW bashing and an insinuation that a VW is guaranteed to leave you stranded whereas a Toyota will guarantee zero problems. My point is the gap is probably a lot smaller than that in the late 2000s. It's not like VW=armageddon and Toyota=utopia.
Originally Posted By: TooManyWheels
The pure engineering types will probably want to weigh in on this, but my statistics classes taught that this number was as low as ten, and as high as 30, maybe 35. Past that point, relatively little was added in accuracy. So if a person knows ten owners of the vehicle in question, the evident pattern is already representative of the larger population. By the time you have known thirty owners, it is almost exactly representative of the whole.
My guess is that you are remembering the speed at which an arbitrary distribution converges to a normal distribution. Yes, after 30 observations, even an arbitrary distribution will not deviate far from a normal. But what I am referring to is statistical significance. It refers to the probability that a mean from one distribution (e.g. avg number of unscheduled visits for VWs) will be "far" away from the mean of another distribution (e.g. visits for Toyotas). There is a test statistic for doing this test called a z-score for a sampling distribution. And it is a function lf the sample size. Given the means and standard deviations, you can back out the sample size that you need. It would be very easy for CR to publish means and standard deviations, but they don't.
Originally Posted By: TooManyWheels
The pure engineering types will probably want to weigh in on this, but my statistics classes taught that this number was as low as ten, and as high as 30, maybe 35. Past that point, relatively little was added in accuracy. So if a person knows ten owners of the vehicle in question, the evident pattern is already representative of the larger population. By the time you have known thirty owners, it is almost exactly representative of the whole.
My guess is that you are remembering the speed at which an arbitrary distribution converges to a normal distribution. Yes, after 30 observations, even an arbitrary distribution will not deviate far from a normal. But what I am referring to is statistical significance. It refers to the probability that a mean from one distribution (e.g. avg number of unscheduled visits for VWs) will be "far" away from the mean of another distribution (e.g. visits for Toyotas). There is a test statistic for doing this test called a z-score for a sampling distribution. And it is a function lf the sample size. Given the means and standard deviations, you can back out the sample size that you need. It would be very easy for CR to publish means and standard deviations, but they don't.
Last edited: