You really don't want to include a "zero mean zero variance" treatment in your analysis: it will generate a main effect of treatment, and the "zero variance" part will cause your error mean square to be underestimated, leading to what one might call spurious sensitivity to treatment effects. And it isn't as if you needed to find out from the analysis that this treatment differs from the others -- you know that already.
For the treatment that was "poorly colonized", is that true of all four blocks? If so, I'd recommend eliminating that treatment from the analysis. You have essentially no useful information about it, in terms of the variable you're analyzing. (Except, perhaps, that the treatment is lethal.) If the treatment really does have, as you write, zero mean and zero variance, you don't need a formal ANOVA to tell you that it differs from the other treatments. In a 3-by-4 ANOVA you can then assess the relative effects of the other treatments. If, instead, the "zero mean zero variance" phenomenon applies only to one of the 16 block-by-treatment conditions in your experiment, there is a way of adjusting for that particular condition while pursuing effects in the rest of the experimental design. But this is not a standard form of analysis, and will take some explaining. I'll wait for clarification before trying to describe it. On Thu, 15 Apr 2004, Lenaick MENOT wrote: > I have conducted a colonization experiment invloving four treatments > each with four replicates, the replicates were randomly placed in four > blocks. I would like to use a two way ANOVA without replication to > test the effects of treatments but one of the treatment was poorly > colonized and most of the taxa are lacking. > > My question is could I perform an ANOVA including the treatment with > zero mean, zero variance, should I exclude the treatment with zero > mean zero variance from the analysis or is there any special procedure > to deal with zero mean zero variance in an analysis of variance? > > Thank you for your help. > > L�naick Menot ------------------------------------------------------------ Donald F. Burrill [EMAIL PROTECTED] 56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
