I have plotted 95% Confidence Interval for the Mean. The results show that the confidence interval in one of the cases is higher than the Mean (The Mean is 0.18/ Confidence interval is 0.26).
I used the value of confidence interval for the lower and higher value of the error bar. The figure shows that the Error bar is below zero, which does not make sense at all.
Regards,
It does not make sense if the measured quantity cannot take negative values. In this situation, the lower limit would be zero and the upper limit should be determined based on the distribution that is not normal.
I second Hening: if you have used a sensible distribution model, then the interval makes sense. If the distribution model is not sensible, it does not make sense.
The standard error, itself, can only be a positive number. However, the error can be a larger number than a mean resulting in a confidence interval that encompasses zero. In cases were the measures cannot be below zero, you should just ignore the portion below zero.
I don't think that it is ok to "just ignore the portion below zero". This procedure will not give intervals with the desired level of confidence.
.
the construction of "correct" confidence intervals for bounded parameters is a quite hot topic in experiental physics : for sure, the neutrino mass is very very small (if it is not strictly 0), but it has to be positive !
.
see https://arxiv.org/ftp/arxiv/papers/0911/0911.4271.pdf
also
https://arxiv.org/pdf/physics/9711021.pdf (ref[2] in the note above)
http://cds.cern.ch/record/610464/files/0303111.pdf
https://projecteuclid.org/download/pdf_1/euclid.ss/1030550859
.
- Badges
- Science topic