Hi Catherine, If you didn't get a response earlier, it's not because of your English, but the subject matter.
1. The nesting you've set up seems basically appropriate for a split-plot experiment such as this, although I don't understand why you're considering individual as a factor--isn't that your sample unit? 2. What's a 'base model'? What's required to run the model (lots of things will 'run') isn't necessarily what's correct...specify your statistical model and the Minitab model will come from that. 3a. I don't use Minitab, but it looks like post-hoc tests are straightforward to perform--they seem to have good online references. 3b. I think a mixed model, which doesn't assume all effects are constant across individuals, is very appropriate here. For example, your site variable can reasonably be expected to be correlated within foresttype, but the fixed-effect model assumes independence of observations. So I would treat site as a nested random effect. (That said, while the specification of random effects is good, and necessary, in my experience it's more important to make sure the whole- and split-error terms are being calculated correctly, because they frequently affect significance tests quite strongly.) Disclaimer: I'm NOT a statistician. If possible, see if Minitab runs a help forum or listserv for its users, or post your question (in general terms) to someplace like the R help list--lots of real statisticians lurk there. Hope this helps, Ben On 9/20/07 12:32 AM, "[EMAIL PROTECTED]" <[EMAIL PROTECTED]> wrote: > There are 2 forest categories (factor: foresttype) which are plantation and > natural forest. Each forest category has 2 sites nesting in. Each site > (factor: site) has 3 liana species (factor: species) which is common to all > sites. Lianas are located at edge and inside forest (factor: location) and > response variable is number of fruits. So fixed factors is foresttype, > species and location and nesting factors is site and liana individual > (random). I am carrying out nested anova with minitab using following model. > > Foresttype > Site(Foresttype) > Species > Individual(Site Species) > Location > Foresttype*Species > Species*Site > Foresttype*Location > Location*Species > Foresttype*Location*Species > > My question is: > 1. Is model correct (is nesting correct) > 2. Is base model one below? Are last 2 terms necessary to run model? > Foresttype > Site(Foresttype) > Species > Individual (Site Species) > Location > Foresttype* Species > Species*Site > > 3. How to do post hoc tests to see if location (edge or interior effects > fruit abundance) or species differ in result. I am also hearing of mixed > models as better than nesting anovas for this type data. What is this and how > to do?
