2 Comments
User's avatar
Christian_Z_R's avatar

I guess the next step might be finding the probability of n-1 out of n observations all being above (below) the median and then try to figure out how many observations you would need to have a 95% confidence in the median being between the second highest and second lowest observation.

I guess if you did this iteratively you could find an algorithm along the lines of: "With m independent observations the median is with 95% confidence between the k highest and k lowest observation."

Expand full comment
O.H. Murphy's avatar

Yeah, that would be pretty cool. I'm not sure, but perhaps if you assumed that the population distribution was symmetric, then the removal of the k highest and k lowest would have symmetric effects on the probabilities of being above/below the median for the remaining data, allowing them to stay 50-50. I'd have to work through the math though.

Expand full comment