Help using MANOVA for a confused biologist

Hi all,
I have run an experiment where I measured 2 leaf surface characterists, under 2 environmental conditions: normal and stress. These characteristcs, my DVs, appear to have an inverse relationship to one another. When I altered the environmental factor, the data points appear to cluster together. I still have an inverse relationship, overall, but the environmental factor seems to affect where on the trend line the points fall. Visually they "cluster" together by environmental treatment. I have run some MANOVA tests using MiniTab, at someone's recommendation, to try and test where this apparent "clustering" is really due to the environmental factor or just part of the normal distribution of points (there is quite a lot of scatter, as with many biological systems). The MANOVA appear to uphold what I can see visually on the graphs. But I still don't really feel like I understand clearly what the MANOVA is doing to write confidently about it. Could someone explain in layman's terms what exactly MANOVA does? And do people think this the most appropriate test?

Also, I did not define either of my characterists as co-variates because I don't think they really are. Or at least they are both co-variates of each other because both characteristic influence the other. I was able to tell Minitab to perform the MANOVA on both characteristic verses my environmental factor simultaneously. However this was a bit of a guess as I don't really know what I'm doing and there are no expertises on MANOVA in my uni department. Does anyone know if this was the right thing to do?

Any help would be much appreciated!


New Member

MANOVA is a statistical technique that allows you examine the effects of categorical variable on multiple dependent variables (hence MANOVA). So you look to see whether this a significant multivariate difference or effect by looking at the Wilks Lambda and p-value. If so, you have to look to see exactly where the significant effect lies, it may be on a single dependent variable and not the others or on a few, or all. However, you can look at separate p-values (but you have to adjust the alpha criterion level from .05 to .05/k where k is the number of dependent variables; so if you have two DVs, your alpha will be .025, and you assess each dependent variable's p-value against this criterion (p-value must be lower).