On 29 Aug 2003 07:21:02 -0700, [EMAIL PROTECTED] (Damian Lidgard) wrote: > Dear all, > > Can someone advise me on the following stats. I have a behavioural data set > consisting of 7 variables one factor. I ran two MANOVAs on the data set > addressing two different questions. I need to adjust the p values of the > seven univariate tests within each MANOVA for the Bonferroni correction. > However, some of the variables are correlated.
Bonferroni is what you do *instead* of MANOVA. It gives you a p-value for the whole experiment, using each variable. The MANOVA gives a p-value for the whole experiment, for the assumption that correlation exists and might be important; or the inter-relations otherwise might be important. - It is conservative of you to want to do MANOVA followed by Bonferroni; I suspect that you ought to consult in person with a statistician so you can confide all your data and your expectations, and your hopes and intentions. Bonferroni correction assumes that the tests are independent; but it is not hurt by their correlation. > > Two questions: How do I determine the appropriate adjustment for the > Bonferroni correction on correlated variables, and because I ran a MANOVA on > the same data twice, do I need to adjust the P value for each test according > to the combined number of univariate tests, i.e. 2 sets of 7 tests = 14, or > for each set of 7 univariate tests in turn. What can you get away with? Are the questions really independent? - in that case, you start over with the 5% 'experiment' for each. > > Any advise will be most welcome. > With behavioral data, we usually have one or two questions that matter. A question might be selected from your seven items, or formed as a composite score. If you are really loose about what you are asking, and you are sure that the correlations *matter*, then you perform the MANOVA. If there's nothing in the MANOVA, then you are finished. No effect. You can go further to look at "exploratory" results, but you don't want to use Bonferroni correction for exploring. If there's an useful effect in the MANOVA, then there *is* some importance for the 7 variables. You are not acting properly if you drop any one of them; you have to describe all seven. Again, but for this different reason, the Bonferroni correction is not appropriate. Hope this helps. -- 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/ . =================================================================
