What inspired this wish list request?
Thank you for adding Fieller's Method (ratio of means test) to Fit Y by X in JMP 19! There are some usability issues with this first iteration:
- Cannot specify which level is in the numerator or denominator of the ratio. The test does not recognize Value Order for the X values and just automatically selects the larger mean as the denominator by default with no way to change it. Users have to save the resulting report as a data table and take the reciprocal of all the fields to flip the ratio.
- No 1-sided test. Currently, users have to know to run a 90% confidence test to get the 1-sided 95% limits.
- No error handling or warnings for negative (mean) values or unstable confidence interval estimates
- It doesn't make sense to test the ratio of a positive to negative value (or vice versa), but the platform provides an output all the same.
- The confidence limits become unstable as the ratio approaches infinity (denominator approaches 0). This can often be addressed by taking the reciprocal of the ratio to get a more stable estimate close to 0.
- The mratios package in R has error handling for negative means and warnings for unstable confidence limits.
What is the improvement you would like to see?
I'd like to add some functionality to maximize its utility for users:
- Specify which level is in the numerator or denominator of the ratio.
- Add a 1-sided test.
- Add error handling or warnings for negative (mean) values and unstable confidence limit estimates.
Why is this idea important?
Being able to reliably and confidently say whether a product is X Times Better/Faster than another product is a powerful claim for any business to make.