I built a model and i wanted to assess the effect of 6 different independent variables, including interdependency, on a certain output. Not sure how to do that but what i did is that i assigned a Low and High level for each variable to minimze the number of runs. Then i used the Fit definitive screening method in JMP and i ran the multi regression model the program built and i used the logworth table to rank the effect of each variable.
Now the question is, is the methodology correct? and if not, what will be the best way to build a multi-regression analysis in JMP and calculate the p value/logworth for each variable as well as the interdependency
When using statistically designed experiments (DoE for short), 'design' and 'analysis' should go hand in hand. To reflect this, (in JMP 14 at least) there are two menu options under 'DOE > Definitive Screening', though the principle is general.
So, the answer to your question about the analysis will depend crucially on what runs you actually performed (your design). You say you used 'a Low and High level for each variable to minimze the number of run', and that you are interested in interactions between the six factors. Maybe you could use 'Tables > Anonymize' and post up the result? Then folks will have more to go on . . .
Hi Ian Apologies for late reply, i guess now i have more understanding of what i really want to do
Basically i aim to do a sensitivity analysis to see which indp variable has the largest impact on my output. Now the thing is that my output comes from another model, in other words it is a deterministic model So first i assumed that the logworth value calculated by JMP give me a rank of most important factors But now i realize that my understanding is incorrect as the logworth is basically a p-value which give a probability that a factor is significant and not necessarily its impact So any recommendations of what i can do? Or even maybe JMP is not suitable?