Subscribe Bookmark RSS Feed

How can you do a paired data review for hundreds fo paired samples with multiple parameters

chris_dennis

Community Trekker

Joined:

Nov 21, 2014

Problem comes from reviewing sampling plan.  Current sample based on 4 different inputs.  We think one of these inputs is not important or redundant to the sample and does not effect the result.  We did a Fit X by Y Anova, t-test analysis but results in hundreds of charts.  How can be make a summary analysis that is easier to understand big picture and present to management?

1 ACCEPTED SOLUTION

Accepted Solutions
Solution

I think you want to take a look at the Response Screening.  It allows you to run you hundreds of comparisons and then use the FDR stat to find which of the analyses are the ones you want to look at.  It creates a data table with all of the results, and it has an  embedded script that lets you select specific rows(parameters) from the table, run the embedded script and see the details of the analysis.

Analyze==Modeling==>Response Screening

Jim
3 REPLIES
msharp

Super User

Joined:

Jul 28, 2015

Your question is pretty vague without being able to see the accompanying data.

But as far as answering: "How can be make a summary analysis that is easier to understand big picture and present to management?"

I'm going to point you to the Graph>>Graph Builder platform.  While this won't give give you statistical data like a t-test, it is really great for presenting data in a way that is easily consumed by management.

Solution

I think you want to take a look at the Response Screening.  It allows you to run you hundreds of comparisons and then use the FDR stat to find which of the analyses are the ones you want to look at.  It creates a data table with all of the results, and it has an  embedded script that lets you select specific rows(parameters) from the table, run the embedded script and see the details of the analysis.

Analyze==Modeling==>Response Screening

Jim
chris_dennis

Community Trekker

Joined:

Nov 21, 2014

Jim, Thanks for your help.  I did some homework on the Response Screening method and applied it to my data.  I was able to identify only one of the measured parameters had a potential significance to the input I would like to remove and it was significant < 5% of the time.  Using the FDR PValue & PValue vs. Rank Fraction graph it is clear to show parameters that may have an effect.