Analysis of a factorial design

[nonw1988]

Hey guys,

I started using Minitab lately and I have a question, probably rather stupid.

My first design worked out fine, it was a 3? fullfactorial design.

The second one had also three factors but different amount of levels, in total 2*3*4 = 24 experiments.

The program wasn't able to show me the residual graphs and I didn't really understand why this was the case. Same with the ANOVA. There was a note that the MSE might be zero, but that didn't help me.

Is it becase the different factors don't have the same number of levels?

 

[Miner]

To be certain, you would need to attach your file. The most common reason is that you have a fully saturated model and no replicates. This leaves zero degrees of freedom for the MSE. The easiest solution beyond running the replicates is to remove high order interactions (3+ -way) from the model.

 

[nonw1988]

And again you saved my ass Miner. Thanks a lot. After removing the ABC interactions Minitab was able to deliver the figures. I am still not exactly sure why it's working now but I have the information I wanted, which is the important part.

 

[Miner]

When you have a saturated design, there are only enough degrees of freedom to estimate the terms in the model and the overall mean. There are no remaining terms to estimate the experimental error, which is required to calculate the F-ratio and p-value.

There are two ways to deal with this issue. One is to run at least one replicate. This doubles the available degrees of freedom, which solves the issue. The second is to assume that the higher order interactions are unlikely to be significant and to remove them from the model. This frees up the degrees of freedom used by those terms and reallocates them to the error term.

 

[nonw1988]

Alright. I think I got your point. Thanks again.