Gary McBride
New Member
- 1
I have a target file but I'm testing a new package/message. I want to make sure that the test and control files are similar and at least do not significantly differ from each other for my key variables.
The Test & control files will be the same size splitting the total file using a random number generator.
I then plan to measure the key variables using T-Test for each of the two files to make sure they pass the critical value 0.05. (95% significance).
However, I have been asked to either
a) ensure proportions are maintained, i.e gender (Stratified Sampling) or
b) make each file almost the same by sorting on the key variable and selecting every 2nd record as the Test file
Queries:
1) Can I still use T-Tests to measure the similarity / difference between the key variable within the file or do i now need a different stats technique and if so what should i use?
2) Surely option b should not be used as it adds bias? Is it sfe to use option b?
I would appreciate your thoughts.
Thanks Gary
The Test & control files will be the same size splitting the total file using a random number generator.
I then plan to measure the key variables using T-Test for each of the two files to make sure they pass the critical value 0.05. (95% significance).
However, I have been asked to either
a) ensure proportions are maintained, i.e gender (Stratified Sampling) or
b) make each file almost the same by sorting on the key variable and selecting every 2nd record as the Test file
Queries:
1) Can I still use T-Tests to measure the similarity / difference between the key variable within the file or do i now need a different stats technique and if so what should i use?
2) Surely option b should not be used as it adds bias? Is it sfe to use option b?
I would appreciate your thoughts.
Thanks Gary
