On 11 Nov 2002 05:56:05 -0800, [EMAIL PROTECTED] (Gartland, Myles) wrote: > I am stuck on how to determine if there is a statistical difference between > two time series of aggregate percentages. Two groups (n=~80) were surveyed > each year for 7 years, and their aggregate median cost percentage was > reported. > > From 1993 to 1999- > > Group A median percentage- 51.91%. 52.73% ......54.17% > Group B median percentage- 60.13%, 57.03%, .....67.30% > > This is all the information I have other than the sample size for each year > for each group (they are about n=80 per group per year). I do not have > percentages from each individual respondent, just an aggregate percentage > for the group. > > I assume I can call these proportions since they are stated in percentages.
If you are hoping to call these 'proportions' so that you automatically know their Standard Errors - that seems unlikely. It *sounds* like each individual in the sample for a year was already represented by a percentage, so the "aggregate median cost percentage" is some sort of mean. It would not denote a "proportion" of the sort that readily fits a contingency table. Thus, I'd say, "they are numbers" and if you only have the group numbers, then you are in a tough spot for your eventual testing. That is: You have group comparisons of 8 years versus 8 years. And you might have no information at all about inherent accuracy for any given year. When I look at the numbers, I wonder, are these numbers being summed so that one year includes the previous? Or, is there another origin of the numbers that happens to require or explain why the numbers are gradually increasing? In short: Are these "serially correlated"? If that is true, then you can look at the "trends" with errors that are easy to compute, but you might be hard-pressed to defend your test on the group means. For instance: What matters? - the average overall, the average at the start, or the average at the end? > > My hypothesis question is if group B is statistically higher and increasing > at a statistically higher rate then group A. > > However, I am stuck on how to test. I would have assumed a chi-square, but > my limited background in chi-square (if that is appropriate) cannot figure > out what to do. With your lack of experience in statistics being too apparent, you need to discuss the possibilities with a consultant, where you can discuss all the details. -- 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/ . =================================================================
