Originally Posted by grogan545' post='318952' date='Aug 4 2006, 11:52 AM

Hi Zman .Just now read your post.I will need some time to digest all that you are saying.Sounds like you are back to your very analytical normal self.After I digest this post I will try to give an intelligent reply.

Originally Posted by Znod' post='318733' date='Aug 3 2006, 07:03 PM

Hi All:

An issue that comes up from time to time is "magazine racing"--using, for example, the test data provided by the "Big 3" auto magazines--the USbig3). The apparent problem in trying to use such results to assess the relative performance abilities of different cars is the formidable numbers of both intra-magazine and inter-magazine uncontrolled factors. I have done some work that suggests that averages can be used to do some pretty effective mag racing.

I have the following data for 31 different "fast cars" (fcs) from tests based on each of the USbig3: 0-60, 1/4 ET, and 1/4 TV. I won't provide the detail of my matching criteria, but I think they are pretty good.

Using these data, I performed matched pairs t tests for: (1) the mean difference in 0-60 for the 31 fcs for CD vs MT, CD vs RT, and MT vs RT. I did the same thing for 1/4 ET and 1/4 TV. One need not be concerned about my parametric test choice. The central limit theorem implies that the pertinent sampling distriblution of the mean is approximately normal given a sample size of approximately 30 or more. Thus, the results from my matched pairs t tests should be robust.

My results are:

.................................................. .......CD vs MT..........CD vs RT..........MT vs RT
One-tailed probability value: 0 to 60:
.................................................. .............23....................42............. .......28
One-tailed probability value: 1/4 ET:
.................................................. .............02....................09............. .......37
One-tailed probability value: 1/4 TV:
.................................................. .............36....................29............. .......42

So, the point to this point is that the mean differences for paired USbig3's usually are insignificant. In this regard, using standard statistical reasoning, the one significant difference is .02. Thus, it looks like there is a significant difference in the average 1/4 ET reported for the 31 fcs for CD and MT.

These finding suggest to me that the effects of the formidable uncontrolled for factors tend to cancel out across the USbig3. Thus, at the extreme, it often may not be too bad to decide tentatively that the IS350 is faster in the 1/4 than the 545i on the basis of only two tests--with one appearing in one USbig3 mag and the other appearing in another USbig3 mag. However, I would apply this approach with great skepticism if the two mags involved are CD and MT. Additionally, since .09 is consistent with less frequently used rejection level of .10, I also would be extra skeptical if mags involved are CD and RT. I'll emphasize the obvious caveat that the two-test approach to mag racing will result in, I think, significantly mores erroneous conculusions than if averages are used.

But, the main point, following from the large number of insignificant differences, is that mag racing is likely to be highly valid once USbig3 averages for the various fsc's begin to be used. So, for example, I would be very confident in deciding that the C6 Z06 is "faster" in the 1/4 than the 911 Turbo (480 hp) if the USbig3 average ET of the former is less than that of the latter. And, putting the two ideas together, the less and less I rely on averages the more I woul doubt the implications. Also, for example, I would be especially skeptical of the implications of 1/4 ET averages where, for example one average of 2 included CD and the other included MT (similarly with CD and RT).

Now, from an overall perspective, I have done nothing but provide something of an analytical foundation for the ways most of us would use mag tests in doing mag racing intuitively--use averages when possible. Two of my real contributions are finding that (1) we should be extra skeptical of the apparent implications of 1/4 ET comparisons if the mags involved are CD and MT or CD and RT and (2) mag racing is likely to be highly valid once USbig3 averages for the various fsc's begin to be used--especially if "problems" with CD vs MT and CD vs RT are avoided in the 1/4 ET arena. Another contribution not yet mentioned is the implication that USbig3 averages may come closer to "truth" than might have been expected since most of the differences tested are insignificant. I'll be doing at least one follow up "study" on the above ideas.



Any thoughts anyone or am I ?

Come on some one. At least make a comment that leads me to think that your read what I wrote.

Originally Posted by cobradav' post='319016' date='Aug 4 2006, 02:26 PM

Ah, the recovery process has begun. Low and slow with lots of time to THINK and ANALYZE. Good work Znod. I kinda like 5th order polynomials myself

Originally Posted by grogan545' post='319244' date='Aug 5 2006, 07:13 AM

Zman have you read the post on Dinan Engineering yet.He gives a very in depth analysis of computer engine ...

Originally Posted by Znod' post='319281' date='Aug 5 2006, 08:00 AM

Where did you see the post more specifically? I haven't been able to bump into it.

Originally Posted by jiio' post='319282' date='Aug 5 2006, 10:07 AM

Here you go Znod,
http://forums.e60.net/index.php?showtopic=28871.

Originally Posted by Znod' post='274952' date='Apr 27 2006, 05:23 PM

Four types of loss are possible, drive train (including rolling resistance), aerodynamic drag, altitude, and weather.