On 11 Jun 2003 16:18:52 -0700 [EMAIL PROTECTED] (Serge) wrote: > Hello, > > I am posting to report my results and to thank those that have > helped with this. > Thanks again! Basically, I grouped the data into 9 different weight > groups as I described previously. I then thew out groups 1, 8, and 9 > because there wasn't enough data in each group for the mean to be > truly representative of the population of all cars in that weight > group (ex. group 9 had only 2 cars). I then plotted weight group (x) > vs each of the 4 ratings (y). I found that there was a very strong > positive linear relationship between weight group and each of the side > ratings (in each case, r was > .95). Using a hypothesis test I showed > that this was significant with respect to the population. However, the > realtionship between weight group and the front ratings did not seem > to be linear. In part, this is because front tests are conducted by > crashing the car against the wall and thus, the ratings should be > compared in exactly that sense ... not in the sense of a car crashing > head on with an "average weight car." For my sample (187 cars), the > pattern was well fitted by a concave down parabola (ax^2 + bx + c; > with a<0). The critical point was above and between groups 3 and 4. I > used Maple to get two specific equations. The equation fit the points > for the 6 groups very well for the front driver rating. The fit was ok > but not as good for the front passenger rating. I then wanted to see > if there was signifcant evidence that imported cars (non-US) were > safer than domestic cars in each of the 4 crash types. I split the > data and found the mean rating for each type for imported and > domestic. Each imported mean score turned out to be higher. I > performed 4 significance tests and all 4 showed that the difference > was significant with respect to the population. Thus, my study seemed > to favor heavier imported cars. I welcome any questions or comments. > > Cheers, > Serge
Grouping in this fashion causes a number of serious problems. From what you've said why not just estimate nonparametrically (e.g., loess smoother) the relationship between the two continuous variables? --- Frank E Harrell Jr Prof. of Biostatistics & Statistics Div. of Biostatistics & Epidem. Dept. of Health Evaluation Sciences U. Virginia School of Medicine http://hesweb1.med.virginia.edu/biostat . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
