I have a process which has various categories of input and success rate for each category. I have collected a sample and executed a chi square test for determining if there is an association between categories of inputs and success rate for my process. My p value is coming as 0.00 indicating that there is an association. 

 

Now, I want to know how much is the extent of this association i.e. which category is expected to give more success. How do i use the sample data to determine which category of input should i increase more and more to get higher success rate in my process.

Hi Ashish, 

 

A p value of zero indicates that the two extreme categories (the one with lowest success rate and the one with highest success rate) have a significant difference. You may like to remove one of these and run the test again to see if others show a different result. If you again get a zero p value, you can continue removing an extreme value and re-run the test one-by-one and draw conclusions. 

 

Hope this helps. 

Hi Vishwadeep,

 

Thanks for replying.

 

I think I have not been able to explain it rightly in my post. I have a process whose output is binary - Pass or Fail. This process has an input which has various categories ( 8 in nos.), these categories when processed can either result into either pass or fail output. I wanted to know if there is association between the input category and the process output, so i conducted chi square test on the sample data. The p value of the test has come as 0.00 indicating there Alternate Hypothesis is true and there is an association between input category and the process output.

 

Now, i wish to establish statistically that which input category is beneficial for increasing the pass% in the process i.e. if the existing 8 categories of inputs in the process are controlled in right proportions or only some of the 8 categories are allowed as input then the process can achieve higher efficiency.

 

For this, i need to know by how many units should I increase / decrease a category and by how much units the process would increase or decrease by bringing this change.

 

Hope it clarifies.

Thanks for your response. Discussion in a forum is likely to have some risks of misunderstanding.  To be sure that I understand clearly, I request you to provide details of the process and the explanation of categories. 

Sure Vishwadeep,

 

An exam is conducted where the candidates can either pass or fail - this is the output - binary in nature

 

These candidates come from various backgrounds - viz. Profession (8 levels), Age, Area (2 levels), Gender (2 levels)

 

Hypothesis related to all these x's have come with p values lower than 0.05 suggesting association / dependency.

 

Now, I have to find which candidate is likely to pass the exam with highest probability.

 

Hope this helps.

I assume that you have carried out these tests one by one (also called OFAT - one factor at a time). Due to feasibility of interactions (combined complex effects), it is not advisable to combine these results to reach a generalized conclusion. To be able to find a combination in a candidate that has the highest probability of passing, it is advisable to design experiments that use various combinations. Also, in such an experiment it is better to take the output as exam score instead of just a pass/fail outcome. Hope this helps.