Thank you -- I've been racking my poor brain, trying to remember how to get some means in SAS (SAS book in office, working at home). I 'googled' Proc Means and found the answer. DW
--- Rich Ulrich <[EMAIL PROTECTED]> wrote: > On 27 Jan 2004 01:59:46 -0800, [EMAIL PROTECTED] > (Iain Toft) wrote: > > > I'm no stats expert but would like to solve the > following problem. > > > > Given two sets of scores... > > > > S1 = [38.37388112, 38.37471969, 38.40682618, > 38.37863674, ..., > > 38.39506964] > > S2 = [22.40683282, 22.23267916, 22.08881316, > 24.96633041, ..., > > 21.8635192] > > > > I'd like some statistical evidence to show the > scores attained in S2 > > are inferior to those in S1. I can show the > mean,min,max,stdev but > > would like something with a little more clout. > > The Subject: line asks, "is this a t-test > problem." > > It would seem likely to be, if the numbers were not > weird. > > If one set of numbers ranges from 38.4- to 38.4+ > while the other ranges by 3 points or more, then > the *assumption* seems pretty invalid, that you > should imagine the same process as generating > both. Perhaps more should be said, in order to make > the problem seem "reasonable". > > Then, you go to say that they are *paired* scores. > > No explanation is offered for why the one set is so > near to a constant. If they are pairs, and one > score > is nearly fixed, is it better (more reasonable) to > ask > whether the *difference* is has some properties? > - the difference is a very popular starting point > when > scores are properly assorted by pairs. > > Suppose the numbers lump up in a few places -- That > is the prospect by which I can imagine that the two > examples > could overlap. When the *other* group is bigger, > the pairs of scores are merely reversed, and > intermediate > scores do not occur. > > You get a *robust* test by asking the simple > question, > Which one is higher? If the starting values are > indicator, > then the one sample is higher for all 50 values. > > Look more widely for examples, if you want to see > what > a t-test is ordinarily used for, and what it is best > for. > Those data are not a good example. > I find that google has pretty good success when > < FAQ tutorial > are combined with broader topics, > > such as "paired t test". > > -- > Rich Ulrich, [EMAIL PROTECTED] > http://www.pitt.edu/~wpilib/index.html > "Taxes are the price we pay for civilization." > . > . > ================================================================= > Instructions for joining and leaving this list, > remarks about the > problem of INAPPROPRIATE MESSAGES, and archives are > available at: > . http://jse.stat.ncsu.edu/ > . > ================================================================= __________________________________ Do you Yahoo!? Yahoo! SiteBuilder - Free web site building tool. Try it! http://webhosting.yahoo.com/ps/sb/ . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
