statsmodels.robust.scale.scale_tau

statsmodels.robust.scale.scale_tau(data, cm=4.5, cs=3, weight_mean=<function _weight_mean>, weight_scale=<function _winsor>, normalize=True, ddof=0)

Tau estimator of univariate scale

Experimental, API will change

Parameters:

dataarray_like, 1-D or 2-D

If data is 2d, then the location and scale estimates are calculated for each column

cmfloat, optional

constant used in call to weight_mean

csfloat, optional

constant used in call to weight_scale

weight_meancallable, optional

function to calculate weights for weighted mean

weight_scalecallable, optional

function to calculate scale, “rho” function

normalizebool, optional

rescale the scale estimate so it is consistent when the data is normally distributed. The computation assumes winsorized (truncated) variance.

ddofint, optional

Degrees of freedom used in the denominator of the variance computation. Default is 0.

Returns:

meanfloat or ndarray

robust mean

stdfloat or ndarray

robust estimate of scale (standard deviation)

Notes

Uses definition of Maronna and Zamar 2002, with weighted mean and trimmed variance. The normalization has been added to match R robustbase. R robustbase uses by default ddof=0, with option to set it to 2.

References

[1]

Maronna, Ricardo A, and Ruben H Zamar. “Robust Estimates of Location and Dispersion for High-Dimensional Datasets.” Technometrics 44, no. 4 (November 1, 2002): 307-17. https://doi.org/10.1198/004017002188618509.