Kurtosis is a statistical measure that describes the shape of a data distribution especially how heavy or light the tails are. It tells us whether a dataset has outliers than a normal distribution or if most data points stay closer to the average.
It gives analysts a clearer picture of how spread out or peaked the data really is beyond just the mean and variance.
Kurtosis quantifies the degree to which data points cluster in the tails or peak of a distribution. The formula is typically based on the fourth standardized moment:
\text{Kurtosis} = \frac{\mathbb{E}[(X - \mu)^4]}{\sigma^4}
Where μ is the mean and σ is the standard deviation. A normal distribution has a kurtosis of 3 i.e. mesokurtic.
Types of Kurtosis
Kurtosis can be classified as:
- Leptokurtic: Distributions with wide tails and positive kurtosis is called leptokurtic distribution.
- Mesokurtic: When the excess kurtosis is zero or close to zero is called mesokurtic distribution.
- Platykurtic: When the excess kurtosis is negative is called platykurtic distribution.
Interpretation of Kurtosis
Some of the ways to interpret kurtosis are:
- High Kurtosis (Leptokurtic): Data has extreme outliers, peaks sharply at the mean.
- Low Kurtosis (Platykurtic): Data is more evenly spread, flatter peak and lighter tails.
- Normal Kurtosis (Mesokurtic): Distribution resembles the normal curve, standard probability of extreme events.
Calculation of Kurtosis
To calculate kurtosis in statistics, you can follow these steps:
1. Compute the Mean (μ): Calculate the arithmetic mean of the dataset.
2. Compute the Variance (σ2): Calculate the variance of the dataset, which is the average of the squared differences from the mean.
3. Compute the Standard Deviation (σ): Take the square root of the variance to find the standard deviation.
4. Compute the Fourth Moment (μ4): Calculate the fourth moment of the dataset, which is the average of the fourth power of the differences from the mean.
5. Compute Kurtosis: The formula for calculating kurtosis is:
\text{Kurtosis} = \frac{\mu_4}{\sigma^2}
Sometimes, we might also see a version of kurtosis that subtracts 3 from this calculation. This is called excess kurtosis
6. It subtracts 3 because the kurtosis of a normal distribution is 3. So the formula becomes:
\text{Excess Kurtosis} = \frac{\mu_4}{\sigma^2} - 3
Example with Dataset
Stepwise example of calculating and interpreting kurtosis:
Step 1: Define Dataset
data = [10, 12, 23, 23, 16, 23, 21, 16]
Step 2: Calculate Kurtosis Using SciPy
from scipy.stats import kurtosis
kurt_value = kurtosis(data)
print("Kurtosis:", kurt_value)
Kurtosis: -1.3984375
Step 3: Interpret Results
- If kurt_value > 0, the data is leptokurtic (heavy-tailed).
- If kurt_value < 0, the data is platykurtic (light-tailed).
- If kurt_value ≈ 0, the data is mesokurtic (normal-like distribution).
Leptokurtic vs. Mesokurtic vs. Platykurtic Distributions
Property | Leptokurtic | Mesokurtic | Platykurtic |
|---|---|---|---|
Kurtosis | > 3 | =3 | < 3 |
Tails | Fat | Moderate | Thin |
Peak | Sharp | Normal | Flat |
Outliers | More Frequent | Moderate | Less Frequent |
Applications
Some of the applications of kurtosis are:
- Finance: Kurtosis helps measure risk by detecting heavy tails in stock returns, indicating potential extreme losses or gains.
- Quality Control: Identifies the likelihood of defects or outliers in production processes.
- Statistics and Data Analysis: Used alongside mean and variance to describe distribution characteristics more completely.
- Machine Learning: Helps detect anomalies in datasets and informs preprocessing decisions.
- Research and Social Sciences: Measures how extreme certain survey or experimental responses are.
Limitations
Some of the limitations of kurtosis are:
- Sensitive to Outliers: Extremely affected by a few extreme values in the data.
- Misleading for Small Samples: Small datasets may produce inaccurate kurtosis values.
- Doesn’t Show Direction: Kurtosis indicates tail weight but not whether data is skewed left or right.
- Needs Context: Should be interpreted with other statistical measures for meaningful insights.