Statistics Glossary → Skewness
Skewnessconcept
A measure of whether a distribution is symmetric around its mean and in which direction its tail stretches.
Skewness is the mean of the cubed deviations from the mean divided by the cube of the standard deviation (g₁ = m₃ / m₂^(3/2)); software usually reports the small-sample adjusted version G₁. A symmetric distribution has a value near zero. With positive (right) skew the long tail stretches toward large values and the mean is usually larger than the median (income, waiting times). With negative (left) skew the tail stretches toward small values (scores on an easy exam).
Dividing skewness by its standard error gives a z value that is a rough indicator of departure from normality; but in large samples even trivial departures come out 'significant', so the size of |G₁| should be read together with plots. Skewness is strongly affected by a few extreme values. It should not be confused with kurtosis: skewness describes asymmetry, kurtosis the heaviness of the tails. With marked right skew, a log or square-root transformation can make the distribution more symmetric.
An everyday example
Looking at the salaries of 1000 employees in a city, most earn 25-40 thousand TL, but a few managers earn over 300 thousand TL. The mean salary comes out at 42 thousand TL, the median at 33 thousand TL, and skewness takes a large positive value such as 3.1. Here the median is the better way to describe the 'typical employee'.
In MerQur
In Descriptive Statistics the 'SES / SEK' checkbox gives the standard errors and z values of skewness and kurtosis; in the Wilcoxon signed-rank test the 'Symmetry of differences' checkbox reports the skewness (G1) of the differences with its SE and z. The descriptive output of the t-tests, one-way and repeated measures ANOVA, and the Mann-Whitney U, Kruskal-Wallis and Friedman tests gives the adjusted skewness (G1); MANOVA, correlation analysis, canonical correlation, confirmatory factor analysis and path analysis give Mardia's multivariate skewness test, and the Bland-Altman and Likert analyses report the skewness of the differences or the items.
Analyses where this concept appears
Descriptive StatisticsOne-Sample t-TestIndependent-Samples t-TestPaired t-TestOne-Way ANOVARepeated Measures ANOVAMANOVAMann-Whitney UWilcoxon Signed-RankKruskal-WallisFriedmanCorrelation AnalysisBland-Altman Agreement AnalysisCanonical CorrelationLikert Scale AnalysisConfirmatory Factor Analysis (CFA)Path AnalysisExponential Smoothing (Holt-Winters)