I am trying to calculate the sample size required to assess a process improvement that is expected to reduce the proportion of high values of a final measurement by 15% to 30%. I would like to know the number of samples required such that the process improvement compared to historical results is detected with various levels of confidence (e.g., 90%, 95%, etc.). The historical data distribution I am working with is not continuous data (ordinal) and is not normally distributed. A distribution showing sample data is attached.
Can anyone help me with this? Thanks.
One approach would be to treat the data as binomial, high and not high. Then you're looking for a binomial sample size which should be moderately easy, but like Reeza I don't know how to do that in JMP.