Art Kendall <[EMAIL PROTECTED]> wrote in news:[EMAIL PROTECTED]:
> last part part 4 > Sometimes the number of SD's of the raw scores is multiplied by 15 or > 16, sometimes the z score corresponding to the percentile is multiplied > by 15 or 16. > > Just think how many questions there would have to be to make fine > distinctions along the whole ability spectrum! Yep, a lot of people seem to have trouble realizing that a test with a fixed number of items isn't going to be very precise at either the very high or very low ends of the scale. Given that a test-taker is *at least* X deviations (on whatever scale) from the mean, the number of items available to determine just *how much* the taker is over X decreases as X increases. Imagine a test with 100 items, and imagine it being applied to a sample from population who routinely get 99 out of 100 items correct on similar tests. How many effective test items do you have in this case? . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
