On 22 Sep 2002 01:11:05 -0700, [EMAIL PROTECTED] (Juhani) wrote: > Stan and others > > You are right, the experiment analysed the emissions from 12 trucks > using a Fuel A fuel under simulated conditions, sometime later the > trucks were tested with Fuel B. (This could have been some months > later- during which time some of the engines/vehicles had been > serviced, all had increased engine wear, and finally there would have > been different climatic/seasonal conditions etc. The point is that > the tests were done under slightly different conditions (you can never > repeat the same test). > > Does this mean the tests are not valid because the conditions are > different...? surely not.
I would rather talk about whether an experiment is credible or creditable. Or say that the evidence is robust, in favor of a difference that seems large or robust. "Slightly different conditions?" What does that mean? It might be good (in one sense) if the tests were done under widely varying conditions, where the same, wide range exists for Fuel A as Fuel B. Assume Fuel A was tested on one cool spring day, on a level road, in non-stop highway driving with light traffic; and assume the test was on one *other* day for Fuel B. Slightly different? Extremely different? - is the score a contrast of A versus B, or day1 versus day2? > I would argue the tests are independent > from each other (and not dependent) One important part of 'dependent' is that it helps with precision. Car S might score 10 and then 8, whereas car T scores 20 and 18. It is nice to be able to say that "every car changed by 2" or "by at least 1" instead of simply saying that the range was 8 to 18 versus 10 to 20. If the starting points are different, then it typically helps the statistical statement if the rating are paired. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
