sampling plan

[#1]

Hi

I have 4 different lots (A,B,C,D) and each lot contains 24 containers and each container is tested 3 times so I have 72 samples for testing. I am trying to come up with a sampling plan in which I have to test less number of samples so I want to compare statistical power of the following sampling plans:

Case 1: test all samples (n=24x3)

Case 2: test every other container (n= 12x3)

Case 3: test every 3rd container (n=8x3)

Case 4: test every 4th container (n= 6x3)

Case 5: test  alternate container twice (n=12x2)

What is the easiest way to solve this with JMP? Do I need JSL script for this ?

[#2 · Accepted solution]

JMP has a Compare Designs platform. The process is simple: design each candidate and make the data table for it. Select DOE > Design Diagnostics > Compare Designs. The current data table is the reference design. Select the other candidates and proceed. The power comparison is the first part of the evaluation.

[#3]

@Mark_Bailey thank you for your respone. I did compare the designs (file case1 and case2 attached) and it worked but when I am comparing the designs it gives me an error saying three of the columns are not compatible so cannot match those columns. I am unable to figure out why am I getting this error. Is there something I am missing or doing wrong when comparing the designs.

I wanted to compare designs and see what is the power of the design when only two test are done instead of three, so tried arranging the data in another form (file scenario1 and scenario2 attached), but when I compare these designs it shows no change in the power (above it is showing there is difference in power). Just wondering which is the correct way?

[#4]

First of all, these two JMP data tables appear to be from importing an external file, such as CSV format or Excel format. So only the values come in. There is a far amount of meta-data that is created when you design the experiment with JMP that is missing in both cases. I also changed the modeling type of Container to nominal, as it is not a continuous factor.

I am not sure, but I notice a few things that are not the same for both cases. The levels of Container do not match. That difference means that you have different sets of parameters so you cannot compare power.

JMP cannot determine the intended factors, response, or models without the missing meta-data.

[#5]

I have some comments regarding your situation.  Is it power that you really care about? What questions are you trying to answer? How representative is your study of future conditions? It seems like you have an estimate of the measurement system variation, the within lot variation and the between lot variation and the study looks nested.  It does appear you have biased the study to the within lot component, so you may be able to reduce the number of within lot samples in your study.