Skip to main content
Announcements
Join us at Qlik Connect for 3 magical days of learning, networking,and inspiration! REGISTER TODAY and save!
cancel
Showing results for 
Search instead for 
Did you mean: 
Not applicable

Fractile formaula

Hello everyone,

I am a bit new to Qlikview so apologies if this is a bit basic.

I am using the fractile formula to work out a the 95th percentile of a set of data. The formula takes the difference betwenn two different timestamps and then applies a fractile to the whole dataset, which is adjustable externally to a day, month or year level. The formula itself return a time, but when I have compared the reported figure to a manual calculation of the dataset, the formula provides a figure that is higher than it should be.



Could anyone shed some light on why this might be happening? I have used some of the other fractile formulas (rangefractile, etc) with no success. The formula itself can be found below.

IF((scas_division)='OXBUCKS',(Fractile({<year={$(=max(year))},Primary_Result_Group={'Responded'}>}(First19MInRespTime-time_treatment_clock_start),0.95,(Fractile({<year={$(=max(year))},Primary_Result_Group={'Responded'}>}(First19MInRespTime-time_treatment_clock_start),0.95)))

Many Thanks,

Mark

2 Replies
Anonymous
Not applicable
Author

Hi Mark,

Don't have the answer for you but I have a suggustion; try to seperate the different parts of your formula into smaller formula and see what the results are of these smaller formula's.
This way you can see in which part of you code the outcome is different compared to your manual calculation.

Not applicable
Author

Thanks Dennis,

I have broken it down to internal subsets. We have four seperate areas that we work in, and when I run it for individual areas, the number tally up. It appear that it is only when I run it for the whole data set that the figures dont add up.

Would there be any reason for this? The number of records exactly matches the raw data, it just seems that the fractile doesn't work unless something is done to reduce the full data set.

Mark