Summing the squared deviations from the mean
Extremely sensitive to outliers as we are squaring the deviations
A more robust alternative
Because deviations are not squared, large deviations do not get a large impact
MAD <= STD for a given sample size
There is still a mean which can still be sensitive to outliers although less than STD
We can use median absolute deviation
medAbsDev = median(|xi - median(x)|)
(Arithmetic) Mean (L2 Central tendency) Median (L1 Central tendency) SD (L2 Dispersion) MAD (L1 Dispersion)
Last updated 1 year ago
