To find what variable affect one parameter the most

[#1]

Hi! I have 6 column and my 6th column is basically a formula of column 1-5. For example,

- column 1 : temperature A

- column 2 : temperature B

- column 3 : temperature C

- column 4 : temperature D

- column 5 : flow

- column 6 : flow * [(A-C)/(B-D)]

What's the best tool in JMP for me to see - which of the column between 1-5 is the most contributing factor to why column 6 changes. I hope my questions are clear...Thank you.

[#2]

Your question does not make sense to me.  Column 6 you created with a formula - the formula shows how each of the variables is included, so it doesn't make sense to ask how much they contribute to column 6.  Usually, there is some response variable that you are trying to explain - and the columns are potential factors you are using to predict that response.  Then, it is sensible to ask how much each contributes.

[#3]

Column 6 is your response variable and columns 1-5 are your factors.  

You are asking for the sensitivity of the response to the factors.  It can be determined by calculus and new variables can be calculated in new columns.  Then, the column that has the maximum value is the one that has the most effect.  The sensitivity is a function of the actual values, so it could be different at each point in the domain of the factors.

Sensitivity is determined by looking at the equation itself.  The equation is:   

Y = Flow * (A-C) / (B-D)

By calculus we get these sensitivity equations and you can create them as five new columns each having a formula.

SA = dY / dA =  Flow / (B-D), units are flow per degree

SB = dY / dB = -Flow * (A-C) / (B-D)^2, units are flow per degree

SC = dY / dC = -Flow / (B-D), units are flow per degree

SD = dY / dD = Flow * (A-C) / (B-D)^2, units are flow per degree

SFlow = dY / dFlow = (A-C)/(B-D), units are dimensionless

[#4]

In addition to the calculus-based sensitivity equations that have been described, you can detect how the variables will affect the 6th parameter based on how they themselves change (the spread from minimum to maximum possible value of the variable) and the shape of that spread.  So, you can also explore the relationships using histograms of each variable and histograms of the final result (your 6th column).

[#5]

Any time I've got an existing model (i.e., formula) for the relationship between columns I find the Profiler is the best tool for seeing what that relationship looks like.

I took your formula and made up some ranges for the temperatures and set the flow from -1 to 1.

I published the Profiler to JMP Public and have embedded it below.

I'm also attaching the data table here.

Learn more about The Prediction Profiler.