Understanding Skewness in Statistics: Positive Skew, Negative Skew, and Real-World Examples
Not every dataset follows a symmetrical, bell-shaped distribution. In many real-world situations, values tend to cluster around one end of the scale, while a smaller number of observations stretch out toward the other. This pattern is known as skewness.
Consider household income in a large city. Most people may earn within a relatively narrow income range, while a small group of high-income households earns considerably more. These unusually large values create a long tail on the right side of the distribution.
Skewness helps describe this imbalance. It measures the asymmetry of a probability distribution and indicates whether observations are concentrated toward the lower or higher end of a scale.
Understanding skewness is useful when analysing financial data, business performance, customer behaviour, examination results, and other datasets. It also helps determine whether the mean or median provides a more meaningful summary.
What Is Skewness?
Skewness is a statistical measure that describes the asymmetry of a data distribution around its mean.
A distribution with zero skewness is theoretically symmetrical. When the skewness value is positive, the distribution generally has a longer tail extending toward higher values. When it is negative, the longer tail extends toward lower values.
There are three commonly discussed distribution shapes:
- Positive skewness (right skew): A longer tail appears on the right side of the distribution.
- Negative skewness (left skew): A longer tail extends toward the left side.
- Zero skewness (approximately symmetric): Both sides of the distribution are reasonably balanced.
The direction of the tail is particularly important. A common mistake is to identify skewness based on where most observations are concentrated. Instead, skewness is named after the direction in which the longer tail extends.
Right Skew (Positive Skewness)
A positively skewed distribution has most observations concentrated toward the lower end of the scale, with a relatively small number of large values extending the distribution to the right.
Household income is a familiar example. Most households earn moderate incomes, while a relatively small proportion earn exceptionally high amounts. These high-income observations stretch the right tail and increase the arithmetic mean.
Other examples include:
- Residential property prices in major cities
- Personal wealth and investment returns
- Population sizes across cities
- Customer spending amounts
- Waiting times for certain services
One important characteristic of positive skewness is that the mean is often greater than the median.
For example, consider the following annual household incomes (in thousands of dollars):
35, 40, 42, 45, 48, 50, 55, 60, 250
The majority of households earn between $35,000 and $60,000. However, one household earns $250,000.
The mean is approximately $69,444, while the median is $48,000.
The unusually high income increases the mean substantially, making it less representative of what a typical household earns.
This explains why median household income is frequently preferred when reporting income distributions. It is less sensitive to extreme observations and provides a clearer picture of the middle household.
Left Skew (Negative Skewness)
A negatively skewed distribution has most observations concentrated toward the higher end, while a smaller number of low values create a longer tail on the left.
Examination scores can sometimes follow this pattern, particularly when most students perform well but a few receive considerably lower marks.
Other possible examples include:
- Scores in relatively easy examinations
- Retirement ages in populations with specific eligibility patterns
- Customer satisfaction scores when most customers report high satisfaction
- Product ratings when the majority of customers give favourable reviews
In a negatively skewed distribution, the mean is generally lower than the median because a few unusually small observations pull the average downward.
For example, consider these examination scores:
20, 65, 75, 78, 80, 82, 85, 88, 90
The mean is approximately 73.7, whereas the median is 80.
The lower score of 20 reduces the average, even though most students performed reasonably well.
A similar pattern can occur when a variable has a natural upper limit. Examination scores cannot exceed 100, for example. If most students score close to the maximum, the remaining lower scores can create a longer left tail.
However, the presence of an upper or lower boundary does not automatically produce skewness. The actual distribution depends on how observations are distributed within those limits.
Why Skewness Matters in Statistical Analysis
Skewness is more than a description of the shape of a histogram. It can influence how you interpret results, report descriptive statistics, and select analytical techniques.
1. Choosing Between Mean and Median
The mean is sensitive to extreme observations, whereas the median is more resistant to them.
For heavily skewed datasets, reporting the median alongside the interquartile range (IQR) can provide a more representative summary.
For example:
- Income analysis: Median income may better represent a typical household.
- Property valuation: Median property prices can reduce the influence of unusually expensive properties.
- Healthcare costs: Median expenditure can provide a useful summary when a small number of patients generate exceptionally high costs.
2. Selecting Appropriate Statistical Methods
Some statistical procedures rely on assumptions about normally distributed data or model residuals.
When skewness is substantial, these assumptions may require closer examination. However, skewness alone does not automatically invalidate a statistical test.
For example, a t-test can be reasonably robust to moderate departures from normality under suitable conditions, particularly with sufficiently large and balanced samples.
The appropriate approach depends on the sample size, study design, variance structure, and analytical method.
3. Applying Data Transformations
Transformations can help reduce skewness and make certain statistical relationships easier to model.
Common approaches include:
- Log transformation: Often useful for positive, strongly right-skewed variables such as income, transaction values, and financial costs.
- Square root transformation: Can reduce moderate positive skewness, particularly for count data.
- Box-Cox transformation: Provides a family of transformations for positive-valued variables.
- Yeo-Johnson transformation: Supports both positive and negative values, including zero.
A transformation should not be applied simply to make a skewness statistic look better. Always consider the analytical objective and whether the transformed values remain interpretable.
Quick Reference: Positive, Negative, and Symmetric Distributions
| Distribution | Tail direction | Typical mean and median relationship | Examples |
|---|---|---|---|
| Positive skew | Right | Mean greater than median | Income, wealth, property prices |
| Negative skew | Left | Mean smaller than median | Easy examination scores, high satisfaction ratings |
| Symmetric | Neither side has a longer tail | Mean approximately equal to median | Normal distribution, some measurement errors |
These mean-median relationships are useful general indicators, not strict mathematical rules for every possible distribution.
Interpreting Skewness Values
There is no universal threshold for deciding whether a distribution is sufficiently skewed to require a transformation. However, the following guidelines are commonly used for an initial assessment.
| Skewness value | General interpretation |
|---|---|
| Close to 0 | Approximately symmetric |
| Between -0.5 and 0.5 | Little or no skewness |
| Between -1 and -0.5 | Moderate negative skewness |
| Between 0.5 and 1 | Moderate positive skewness |
| Below -1 | Strong negative skewness |
| Above 1 | Strong positive skewness |
For example, suppose a household income dataset produces a skewness coefficient of 1.8.
This indicates substantial positive skewness. A histogram would likely show a concentration of observations at lower income levels and a longer tail extending toward higher incomes.
In this situation, reporting the median and IQR may be more informative than relying exclusively on the mean and standard deviation.
How to Identify Skewness in Your Data
You do not always need to calculate the skewness coefficient manually. Several simple exploratory techniques can help identify asymmetry.
Histogram
A histogram provides a visual representation of the distribution. Look for an extended tail on either side of the main concentration of observations.
Mean and Median Comparison
Compare the mean and median as an initial diagnostic:
- Mean greater than median: Possible positive skewness.
- Mean smaller than median: Possible negative skewness.
- Mean approximately equal to median: Possible symmetry.
Remember that equal mean and median values do not guarantee a symmetric distribution.
Box Plot
A box plot can help identify asymmetry through the position of the median, the relative lengths of the whiskers, and potential outliers.
Using these methods together gives a more reliable understanding of the distribution than relying on a single numerical measure.
Conclusion
Skewness provides a useful way to understand how observations are distributed around the centre of a dataset. It helps identify whether extreme values are concentrated toward the higher or lower end and whether the mean adequately represents the data.
For positively skewed distributions, the mean is generally higher than the median. For negatively skewed distributions, the mean is generally lower. When distributions are approximately symmetric, both measures tend to be relatively close.
For practical statistical analysis, examine the histogram, compare the mean and median, and calculate the skewness coefficient when necessary. These checks help you choose appropriate descriptive statistics, assess modelling assumptions, and make better-informed decisions about data transformations.
