Talk:Analysis of variance

From Wikiversity
Jump to navigation Jump to search

Sorry, I'm not a statistician, and don't really feel qualified to edit here, but I am unhappy about classing ANOVA as a technique of multivariate statistics. Isn't it helpful to keep the clear distinction between multivariate situations and multifactorial situations? ANOVA is a superb example of a multifactorial technique, where perhaps we have only measured one dependent variable, but we've done so with multiple treatments (experimental factors) on which our measured variable depends. This contrasts with multivariate experiments where we have measured many (perhaps hundreds or thousands) of variables, but may have done so in only two treatments (control and treated). The distinction isn't just a matter of words. The two situations are very different, and carry different risks (although there are parallels in obscure situations). In multifactorial experiments we don't have the curse of dimensionality, but we do have the problem that multiple pairwise comparisons would be likely to create false-positive results. In multivariate world we know that some combination of our measured variables is bound to correlate (purely by chance) with our treatment, which complicates interpretation. Sorry to be fussy about this distinction... 22:30, 12 April 2012 (UTC)[reply]