Normal Curve Outliers

hlsmith

Not a robit
#2
You get to pick. Plus or minus 2 std would include 95% of data and 3 std ~99%. Values will land out there but you should visualize them with histogram or boxplot to see if they really are an outlier.
 
#3
Thanks hlsmith. So you are saying that there is no consensus. I understand. Seems to me most choose z>2 (or z<-2)
Possibly because of the choice of popular 95% for confidence intervals. I'm curious whether the 1.5*IQR decision for
outliers would confirm 2 or 3. Thanks for your response.
 

obh

Active Member
#4
I prefer the Tukey's fences you mentioned.
you can calculate the confidence interval using 1.5*IQR under the assumption of normality (Q1=Z0.25, Q3=Z0.75) ...