Manova, groups and iterations

Hi folks,

First off, I have no training or education in statistics, so my knowledge thereof is fairly limited. I've been reading quite intensely on MANOVA and feel I have a reasonably good understanding of it and I've also managed to pick up some of the basic concepts of statistics, but again, my knowledge is quite limited. My main problem is basically how to apply it.

Here’s the description of my experiment:

I have a host program which executes other programs. I have an unmodified version of the host program and 3 versions, each with one modification. It is the effects of these modifications that I am interested in.

I have a set of 6 programs that I execute on the 4 host programs. Let’s call them input programs. The execution of these input programs consists of a large number of transactions. My DVs are values concerning these transactions. One execution of an input program on a host program, results in many thousands of measurements (one for each transaction) and the values for the DVs characterize the transactions – for example, duration or number of aborts. So, for each combination of a host program and an input program, I get a long list of the values for the DVs of each transaction in this single execution/invocation.

However, a single execution is not reliable, so I do several iterations of each combination of a host program and an input program – say 30. And this is where my trouble in applying MANOVA begins. It’s how to deal with these iterations.

I have the following questions:

MANOVA deals aot. with variance within groups and between groups. (Right?) My problem is that I can’t identify what constitutes a group in my case, where I have iterations of the same combination of an input program and a host program, and for each of the 30 iterations, get a dataset consisting of the measurements of the DVs for several thousands of transactions. Are the 30 iterations of a combination of a host and input program a group? If not, then what is?

A more broad question is whether or not I can actually do, what I want to do, with a single application of MANOVA on this massive amount of data: the thousands of transactions generated by the 30 iterations of each of the 6 input programs on each of the 4 host programs; or if I’ll have to split it in some way, and apply MANOVA multiple times, to get the information I’m interested in. For example, should I apply MANOVA to each of the datasets for a particular input program across the different versions of host programs?

I hope I’ve provided enough detail. If not, let me know.

A Googol thanks in advance,
Last edited: