Description
Describe the location, spread, skewness and tail weight of skewed data with stable linear combinations of sorted values instead of conventional moments.
Conventional skewness and kurtosis use cubes and fourth powers of deviations, so a few extreme task times or spend values dominate them and samples give erratic estimates. L-moments are linear combinations of the sorted data. The first is the mean; the second, L-scale, is half Gini's mean difference; the ratios of the third and fourth to the second, L-skewness and L-kurtosis, are bounded between -1 and 1. They exist whenever the mean exists, are nearly unbiased in small samples and are far less sensitive to outliers. Because their values are tabulated for many distributions, plotting a sample's L-skewness against L-kurtosis on an L-moment ratio diagram helps choose a distribution family. The normal distribution has L-kurtosis of about 0.123, and L-kurtosis tracks departures from normality better than classical kurtosis.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Latent
- Stage
- Evaluation, Post-launch
- Data
- Quantitative
- Effort
- Hours
Steps
- Sort the sample.
- Compute the first four sample L-moments from weighted sums of order statistics (probability-weighted moments), using standard software or formulas.
- Report mean, L-scale, L-skewness and L-kurtosis.
- Plot L-skewness against L-kurtosis on an L-moment ratio diagram with curves for candidate distributions (normal, lognormal, gamma, generalised Pareto).
- Choose the family whose curve passes closest and, if needed, fit it by matching L-moments.
- Compare groups or releases on L-scale and L-skewness when classical moments are unstable.
Sources
- Order Statistics (David & Nagaraja, 2003)
Also known as
- L-moments
- L-skewness
- L-kurtosis
- L-moment ratio diagram
- Gini mean difference
- O-statistics
- L-moments (and O-statistics) as robust distribution shape summaries
Categories
- Analytics & Behaviour
- Statistics & Modelling