and the modes in multimodal data. If it's "gotta look good" data, then just be consistent.
Be prepared to explain/justify if you use something that the data consumer is not expecting. Maybe the harmonic average is a better representation? : )
Based on my lab experience you probably either have a mixing or washing issue. In any case, if the numbers have a greater tendency to spread on this substrate than on the others, it's less reliable, and that should be shown, since that implies that any measurements based on using it are less precise (greater error term).
Errors around the median tend to be a pain to calculate, and that is probably the main reason that the mean is typical. Computational tractibility and all. Which just doesn't really matter as much on datasets this small, so trim or use medians at whim. But it's still good to central_tendency()+/-2*stddev(). |
