Understanding Prediction Profiler Results Of Full Factorial With All Main Effects In Model

[#1]

Hello All,

I have a data set that is a full factorial of 3 categorical independent variables and 1 continuous dependent variable. This was not a designed test and contains no repeated data points. The Standard Least Squares model that I'd like to use contains all 3 main effects and 1 2-way interaction. My concern with the prediction profiler is that it is producing confidence intervals based on a sample size of 1 for each treatment. Are these intervals still okay to use based on how JMP looks at each main effect individually and the data set as a whole? Should I only look at the predicted values and ignore the confidence intervals? Is there a better approach that I should be using with this data set? Thanks in advance!

[#2]

First, welcome to the community.  I have the following thoughts and questions:

First, with the very limited information you provided and lack of any pictures of output or attaching the data table, we are extremely limited on advice we can supply.

1. You say "This was not a designed test".  What do you mean by this?  Is this an experiment you designed and ran?

2. Do you have any idea of the measurement system uncertainty?

3. Is there a practical amount of variation in the response variable? (did it change enough?) How do the results compare with your predictions?

3. Since you have no replication, the terms that were left out of the model are the basis for any statistical test.  This is often a biased test.  

4. Try saturating the model and evaluate the Normal and Pareto plots.  Did you look at the residuals?  Did you look at the interaction plot?  Do the answers make sense?