Wolfram Language

Robust Dispersion Measures in the Presence of Outliers

When data is contaminated by outliers, non-robust measures of dispersion, such as standard deviation, do not reflect the dispersion of the uncontaminated data. Robust measures like trimmed variance or dispersion give a more reliable result.

You can model contaminated data by adding to the original sample a small normal component with high variance. Then analyze the dispersion.

show complete Wolfram Language input

Visualize the reliability of these measures by performing a Monte Carlo simulation.

show complete Wolfram Language input

For data without outliers, dispersion is close to the standard deviation.

show complete Wolfram Language input

Related Examples

de es fr ja pt-br zh