- posted and e-mailed. On 20 Jul 2003 08:34:34 -0700, [EMAIL PROTECTED] (mac55) wrote:
> In the medical research I review, the outcome variables are often > rating scales. Examples would be a visual analogue scale for pain or > rating scales (usually from 0 - 10) of various other factors. These > data are often compared using parametric statistics. My question > is.....are these scales interval/ratio or are they ordinal. My > feeling is that these are ordinal data unless you can show me studies > that prove the an 8 is truly twice as much as a 4. Therefore I have > always used non-parametric statistics. Am I correct???? No, you doubly messed up. First: Ratio is not the same as interval. "Interval" means that 4 is just as far from 8 as from 0. Eight would be twice as much if it were ratio -- If "doubling" is mentioned a lot when you discuss the data, that suggests (to the statistical observer) that it might be useful to take the logs. Are the differences truly equal intervals? Well, after you perform the "rank-transformation", do you like the observed intervals better? - the intervals, after all, are what you are testing. A decent scale of any endurance will probably have better spacing than what your sample-in-hand provides from its ranks. I don't know if your Visual Analogue scale is using integers or not, but for integers, you can have more distortion introduced by lumping in categories, if the N is not tiny. You might read Agresti (An introduction to categorical data analysis) for his example that shows three different scorings for categories (pp 35) of alcohol consumption. Basically, there is little risk in using ANOVA on the raw scores of limited integer scales when there can't be any such thing as an "extreme outlier". Should you use the ranks? Well, you do find it (a little? thoroughly?) misleading to refer to the average of your scores? If the means seem misleading, *that* is a suggestion that perhaps you might want to use the ranks. If you do it both ways, it will usually be more informative to your readers if you report the observed means of your data; and the rank-tests can serve as confirmation (presumably) that the parametric tests are not misleading you in either direction. Hope this helps. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." Justice Holmes. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
