Skewness & Kurtosis

When analyzing data, the mean and standard deviation tell us about the center and spread of a distribution. However, two distributions can have identical means and standard deviations but look entirely different. This is where skewness and kurtosis become essential. These measures quantify the shape of a distribution, revealing its asymmetry and the characteristics of its tails, which are critical for understanding data behavior and making informed decisions.

Understanding Skewness

Skewness measures the asymmetry of a probability distribution. A perfectly symmetrical distribution, like the normal distribution, has zero skewness. When a distribution is skewed, one tail is longer than the other, pulling the bulk of the data towards one side. This asymmetry indicates that extreme values are more prevalent on one side of the mean, influencing the relationship between the mean, median, and mode.

pythonCalculating Skewness with SciPy

📐 Pearson's First Coefficient of Skewness

While more complex moment-based formulas exist, a simple approximation for skewness, particularly useful when the mode is clear, is Pearson's first coefficient:

S_k = \frac{\text{Mean} - \text{Mode}}{\text{Standard Deviation}}

Another common measure, the moment coefficient of skewness, is based on the third standardized moment:

\gamma_1 = E\left[\left(\frac{X - \mu}{\sigma}\right)^3\right]

where $E$ is the expected value, $X$ is the random variable, $\mu$ is the mean, and $\sigma$ is the standard deviation.

Types of Skewness

Skewness can manifest in three primary ways. A positively skewed (or right-skewed) distribution has a long tail extending to the right, indicating that most data points are concentrated on the left side, with a few larger values pulling the mean higher than the median. Conversely, a negatively skewed (or left-skewed) distribution has a long tail extending to the left, meaning most data points are on the right, with a few smaller values pulling the mean lower than the median. A zero-skewed distribution is perfectly symmetrical, with its mean, median, and mode often coinciding.

Distribution Shapes and Skewness

This chart illustrates three common distribution shapes: a symmetric (normal) distribution, a positively skewed distribution, and a negatively skewed distribution. Notice how the tails extend differently, indicating the direction of skew.

Loading chart...

Key Insight: Positively skewed distributions have a longer tail to the right, while negatively skewed distributions have a longer tail to the left. Symmetric distributions have balanced tails.

Check Your Understanding

In a positively skewed distribution, what is the typical relationship between the mean and the median?

Interpreting Skewness

The direction and magnitude of skewness provide crucial insights. For instance, income distributions are typically positively skewed, meaning most people earn less than the average income, and a few high earners pull the average up. Understanding this helps avoid misinterpretations that might arise from solely relying on the mean. In such cases, the median often provides a more representative measure of the 'typical' value. Skewness also impacts statistical tests, as many assume normally distributed data, and highly skewed data can violate these assumptions, leading to inaccurate conclusions.

pythonMean vs. Median in Skewed Data

Understanding Kurtosis

Kurtosis measures the 'tailedness' of a distribution, indicating the presence of outliers. It describes how heavy or light the tails are relative to a normal distribution, and consequently, how peaked or flat the distribution is around its mean. High kurtosis implies more frequent extreme values (heavy tails) and a sharper peak, while low kurtosis suggests fewer extreme values (light tails) and a flatter peak. It helps us understand the probability of observing values far from the mean.

📐 Moment Coefficient of Kurtosis (Excess Kurtosis)

The moment coefficient of kurtosis is based on the fourth standardized moment:

\gamma_2 = E\left[\left(\frac{X - \mu}{\sigma}\right)^4\right] - 3

This formula calculates excess kurtosis, where 3 is subtracted to make the kurtosis of a standard normal distribution equal to 0. This allows for easier comparison to the normal distribution. If the value is positive, the distribution has heavier tails than a normal distribution; if negative, it has lighter tails.

Types of Kurtosis

Distributions are categorized into three types based on their kurtosis relative to a normal distribution. A mesokurtic distribution has kurtosis similar to a normal distribution (excess kurtosis of 0). A leptokurtic distribution has positive excess kurtosis, indicating heavier tails and a sharper, more pronounced peak, suggesting a higher probability of extreme values. A platykurtic distribution has negative excess kurtosis, meaning lighter tails and a flatter peak, implying fewer extreme values than a normal distribution.

Distribution Shapes and Kurtosis

This chart compares three distributions with different kurtosis levels: mesokurtic (normal), leptokurtic (heavy tails, sharp peak), and platykurtic (light tails, flat peak). All distributions are centered around zero for comparison.

Loading chart...

Key Insight: Leptokurtic distributions have higher peaks and fatter tails, indicating more extreme values. Platykurtic distributions have flatter peaks and thinner tails, suggesting fewer extreme values.

Check Your Understanding

A financial asset's returns distribution shows high positive excess kurtosis. What does this imply?

Interpreting Kurtosis

Kurtosis is particularly important in fields like finance and risk management. A leptokurtic distribution for asset returns, for example, signals that extreme price movements (both crashes and booms) are more likely than a normal distribution would suggest. This means higher risk. Conversely, a platykurtic distribution might indicate a more stable process with fewer unexpected large deviations. Understanding kurtosis helps in selecting appropriate statistical models and risk assessment strategies, as it directly impacts the perceived probability of rare events.

pythonCalculating Kurtosis with SciPy

Try It Yourself

Modify the leptokurtic_data generation in the code above. Change the df (degrees of freedom) parameter of np.random.standard_t to a higher value (e.g., 10 or 30). How does this affect the calculated excess kurtosis and the shape of the histogram?

python

Skewness vs. Kurtosis: Key Differences

Characteristic	Skewness	Kurtosis
What it measures	Asymmetry of the distribution	Tailedness and peakedness of the distribution
Key values	Positive (right-skew), Negative (left-skew), Zero (symmetric)	Positive (leptokurtic), Negative (platykurtic), Zero (mesokurtic)
Impact on Mean/Median	Pulls mean away from median towards the longer tail	No direct impact on mean/median relationship (primarily affects spread of extreme values)
Practical implication	Indicates direction of extreme values; median often better central tendency for skewed data	Indicates likelihood of extreme values (outliers); critical for risk assessment

A summary of the distinct roles skewness and kurtosis play in describing distribution shape.

Key Takeaways

Skewness quantifies the asymmetry of a distribution, indicating if one tail is longer than the other.
Positive skew means a longer right tail, with the mean typically greater than the median, pulled by high values.
Negative skew means a longer left tail, with the mean typically less than the median, pulled by low values.
Kurtosis measures the 'tailedness' and peakedness, indicating the presence and impact of extreme values.
Leptokurtic distributions (positive excess kurtosis) have heavy tails and a sharp peak, implying more outliers.
Platykurtic distributions (negative excess kurtosis) have light tails and a flat peak, implying fewer outliers.
Both measures are crucial for understanding data behavior beyond central tendency and spread, especially for risk analysis and model selection.

← All lessons in Descriptive Statistics