3 ms·
If I really have to go with a single number I’d go with a trimmed mean, this comes after a long stint both in academia and trying to generate metrics pipelines
by ramraj07 4y ago
If I really have to go with a single number I’d go with a trimmed mean, this comes after a long stint both in academia and trying to generate metrics pipelines in industry.
- superb-owl 4y agoCorrect me if I'm wrong, but I think Trimmed Mean does well with normal distributions with some outliers. Weath doesn't follow a normal distribution even approximately, so variations on the mean will often give counterintuitive results.
- ramraj07 4y agoI’ve used it in many contexts and compared mean median and trimmean in many different distributions (even bimodal ones); trimmed mean works amazingly well and is very robust at representing the data’s central tendency in many contexts. Depending on how much you trim it, it can essentially get close to a mean or a median anyway. So consider it more as a mean or median but with a bit less outliers or a bit more extra context respectively.