Is it possible to do weighted fitting with asymetrical errors? (plus and minus different)
Details: I measure probabilities, my data is from 0 to 1, thus the error bar for 0 values is only positif (and the one for 1 values is only negatif)
When fitting I choose "Control parameters", weighting method: arbitrary Dataset. But which error to choose? The positif? The negatif? A combination of both (average)?
I think the average is an approximate solution, but the asymetry is lost and the confidence band will become negatif in a certain range... not very satisfying.
Is there a better method (even if more complicated)?