Skewness and Kurtosis

Core Concepts: What Are Skewness and Kurtosis?

Skewness is a moment-based measure of the asymmetry of a distribution around its mean. A value of zero indicates symmetry; positive values point to a right-leaning (high-value) tail, and negative values to a left-leaning (low-value) tail. Kurtosis characterizes the peakedness and tail weight of a distribution relative to the normal. The normal distribution has a kurtosis of 3; in practice, excess kurtosis is reported by subtracting 3, so interpretation is centered on zero, making it easier to assess departure from normality.

How Are They Computed?

Skewness is defined as the third standardized moment about the mean: g1 = [n / ((n-1)(n-2))] × Σ[(xi - x̄) / s]³, where n is the sample size, x̄ is the sample mean, and s is the standard deviation. Kurtosis is based on the fourth standardized moment, with excess kurtosis obtained by subtracting 3. Standard errors for these statistics can be approximated as SE_skewness = sqrt(6/n) and SE_kurtosis = sqrt(24/n), enabling z-tests to assess whether departures from normality are statistically meaningful.

Common Misconceptions and Misuses

One of the most common misconceptions is that kurtosis measures peakedness; it primarily reflects tail weight. A visually peaked distribution does not necessarily have high kurtosis. Another misuse is evaluating skewness and kurtosis solely by numeric thresholds: with large samples, small departures become statistically significant yet practically negligible. Normality should be assessed holistically using Q-Q plots and tests such as Shapiro-Wilk, rather than relying on skewness and kurtosis values alone.

Why It Matters in Research Practice

Skewness and kurtosis play a decisive role in selecting the appropriate analysis method. With high skewness, the median better represents central tendency than the mean. Parametric tests requiring normality — such as t-tests, ANOVA, or Pearson correlation — can produce misleading results in the presence of severe skewness or extreme kurtosis. In such cases, data transformations (logarithmic, square root) or non-parametric alternatives such as the Mann-Whitney U test are preferred. These statistics thus serve as a fundamental checkpoint in assessing data quality and determining analysis strategy.

Sources