DSD: Recreate Estimates

[#1]

Hi everyone,

I'm new to jmp, therefore I have a propably very basic question.

I'm using jmp pro 15 and performed a definitive screening design with 7 factors (6 continous, 1 discrete).

To evaluate the results in other programs, I want to recalculate the response estimates manually.

For this, I used to following formula:

response = intercept + ...

                   sum_over_main_effects(factor * main_effect) + ...

                   sum_over_interaction_effects(factor1 * factor2 * interaction_effect) + ...

                   sum_over_quadrativ_effects(factor * factor * quadratic_effect)

Hereby, the intercept, the main_effects, interaction_effects and quadratic effects were given by the DSD.

The factors were normalized before multiplying them with the corresponding effects to a range of -1 and +1.

For the influence of the discrete factor, I added the factor estimate whenever the discrete factor is active.

However, I am not able to recreate the response estimates at all. 

Can you tell me the correct way to calculate the estimate respones out of the parameter estimates?

Thanks for your help!

[#2]

I believe you could build the same model using the custom design platform (fill out the Factors and Model sections the same way) and then check Simulate Responses before making the table.  That way a 'Y Simulated' column is created with the appropriate formula that you can copy and paste.

[#3]

Thanks for your reply!

I just tried this and it does strangely not seem to work.

However, it should in theory be really simple to calculate a simulated response of a linear model if you know the correct mapping.

Do you know, what the factor estimates of the DSD indicate? Is this the quotient between the response and the corresponding factors?

[#4 · Accepted solution]

The Fit Definitive Screening platform is really a model selection platform. You click Make Model or Run Model to complete the regression analysis in the Fit Least Squares platform. So the parameter estimates are just regression coefficients. The difference that you see is likely caused by the parameterization of the categorical factor. The continuous factor levels are 'coded' (not necessarily 'normalized' as you mentioned) and the categorical factor levels are explained here.