On Jul 23, 8:36 am, Richard Ulrich <Rich.Ulr...@comcast.net> wrote: > On Sun, 22 Jul 2007 18:36:50 -0000, loom91 <loo...@gmail.com> wrote: > > [snip, some previous and replies. Just to reply to one > point] > > > > > We plan on taking 50 data points from each school, about 20-25% of the > > class size. So you say that we should take the same number of points > > even if one school has more students than the other? Also, do you mean > > that the sample size is insufficient to draw reliable conclusions > > about whether coed schools really lessen the gender differences? > > The earlier poster tried to make this point, and I will > try again. Whether you are comparing 50 or 500 > students from two schools, you are only comparing > *two* schools. Or six schools. Within the schools, > there is a lesser hierarchy, perhaps, of a dozen > teachers -- or whatever. If you see differences between > two schools, is it something idiosyncratic to those > two schools, or idiosyncratic to that selection of a > dozen teachers? > > If you want draw an inference about "schools", it is > almost necessary to have a large number of *schools* > represented. There is a statistical approach that tries to > test across multiple strata, but that quickly gets complicated. > > With two schools, or a few schools, you can easily > do comparisons, by t-tests or ANOVA, that tell you whether > these particular schools seem to "differ". That can be a > starting point for discussing all the *wrong* reasons that > might exist that could account for the differences. > Selective admissions? Selective attrition? Particular > teachers? > > Even when there are 50 or 500 *schools* being sampled, > the resulting "inference" (whether coed schools differ) can > be challenged for systematic biases. With a large number > of schools, it begins to be possible to argue that various > factors have been measured, and that they indeed do seem > to be balanced or equivalent between schools, or otherwise > accounted for. > > With a few schools -- try to keep the questions simple, and > try to keep the conclusions well-tempered by doubts. > > I'm sorry I can't be of more help to the questions that you > pose. > > -- > Rich Ulrich, wpi...@pitt.eduhttp://www.pitt.edu/~wpilib/index.html
Thanks, I think I get your point. When comparing types of schools, the data points are actually the schools themselves and in that case there are only 5 (possibly 8) data points. Well, we have to work within our abilities. It is clearly beyond our abilities to collect and analyse data from hundreds of schools. We will work with what we have and see what we can arrive at.