correct interpretation of fit least square report

[#1]

I am trying to find if my process parameters have any significant influence at response variable. The fit least square report shows overall model (shown in the picture) for assay is not significant but the effect test summary shows factor a and interaction a*d to be significant. Is this correct interpretation of the above case: except factor a and a*d no other parameter have influence on assay?

[#2]

Please show the Leverage Plots for these two terms that you mention. Also, the Analysis of Variance table is useful.

[#3]

[#4]

Thanks for the leverage plots and the parameter estimates table.  Based on what can be seen you are correct that only a and a*d are important to the model.  You could/should remove all of the other terms except d because of effect heredity with the interaction term.  So, your final model would only include a, d and a*d.  Your actual by predicted and the leverage plots show something unusual as well in that the error bars encompass the mean line (blue line).  This is an indication of a poor fit model.  Your R-square looks okay, but the Prob > F should be closer to zero to be considered a good model.  Take out all of the other terms using the effect summary and show us what you get when you do that.  Right now your model is way over-fit.  

HTH

Bill

[#5]

Thank you @Mark_Bailey  and @Bill_Worley  for your response. Attached is jmp file with both original model (named fit least squaes) and reduced model (named reduced model).

[#6 · Accepted solution]

First and foremost, I would fit the two models seperately because a and d are important in both model, but b and c are only important in fitting the data for y1.   What is your goal for y1 and y2?  Based on your profiler it looks like you are trying to maximize both values.  You can save the fit models back to the data table and then make a combined profiler from there and then do the optimization.

Second, even with the most reduced model possible y2 has a lower R-square than you would like, but based on the data, that's the best you can do with the available inputs.  Are there other possible inputs that could/would influence the value for y2, something not in your DOE?  Is there any wiggle room to have a higher d value?  If maximizing your outputs is your ultimate goal then a higher d value seems to be the way to go.  You may want to augment your DOE with higher d values to see if this holds true for both y1 and y2.  Having a higher d value may not be feasible, but it is at least worth looking at to see what is possible if your goal is to maximize y1 and y2.

HTH

Bill

[#7]

Thank you @Bill_Worley for your response. My goal is to match target. But I agree with you there is something affecting y2 response and I'll have to investigate that. Thank you so much for discussion and your help.