Measures of Dispersion: Range, Variance, and Standard Deviation

Two classrooms score an average of 70 marks in a test. Sounds identical on paper. But in one class, everyone scored between 65 and 75. In the other, scores ranged from 20 to 100. The average alone can’t tell you this story – you need to look at how the data is spread around that average. This spread is what statisticians call dispersion, and it’s just as important as knowing the center of a dataset.

Why the average alone isn’t enough 🔗

Measures of central tendency – mean, median, and mode – tell you where the “middle” of your data lies. But they say nothing about how consistent or scattered the individual values are. Dispersion (also called variability or spread) fills this gap. A dataset with small dispersion has values clustered tightly around the center, indicating high uniformity. A dataset with large dispersion has values scattered widely, indicating less uniformity and, often, less predictability.

This matters in almost every real-world context. A factory producing bolts wants low variability in bolt diameter – high dispersion means inconsistent, unreliable products. An investor comparing two mutual funds with the same average return wants to know which one swings more wildly before putting money in. Dispersion is what separates “reliable” data from “unpredictable” data, even when the averages look the same.

Range: the simplest measure of dispersion 🔗

The range is the easiest measure of dispersion to calculate – it’s simply the difference between the largest and smallest values in a dataset. If exam scores run from 32 to 98, the range is 66.

Its simplicity is also its biggest weakness. The range only looks at the two extreme values and ignores everything happening in between, so two datasets can have wildly different internal patterns while sharing the exact same range. It’s also extremely sensitive to outliers – a single unusually high or low value can distort the picture entirely.

The coefficient of range 🔗

Because the range is an absolute number expressed in the same units as the data, it isn’t useful for comparing datasets measured in different units, such as marks versus rupees. The coefficient of range solves this by converting it into a relative measure:

Coefficient of range = (L − S) / (L + S), where L is the largest value and S is the smallest.

This ratio strips away the units, making it possible to compare the relative spread of two entirely different datasets on a common scale.

Variance and standard deviation: the real workhorses 🔗

While the range only uses two data points, variance and standard deviation use every single value in the dataset, which is why they’re considered far more reliable.

The standard deviation was introduced by the British statistician Karl Pearson in 1893. It’s calculated as the square root of the mean of the squared deviations of each value from the arithmetic mean – which is why it’s sometimes called the root-mean-square deviation. Squaring the deviations before averaging them ensures that positive and negative deviations don’t cancel each other out, and it also gives extra weight to values that are far from the mean.

The variance is simply the square of the standard deviation. So if you already have one, you have the other. Variance is useful in its own right for statistical modeling and further calculations, but because it’s expressed in squared units (like “marks squared” or “rupees squared”), it’s harder to interpret intuitively. Standard deviation, expressed in the original units, is easier to relate back to the actual data.

Population vs. sample: why n − 1 shows up 🔗

Here’s a subtlety worth understanding. When you calculate variance for an entire population, you divide the sum of squared deviations by N, the total number of observations. But when you’re working with a sample drawn from a larger population – which is almost always the case in real research – dividing by n tends to underestimate the true population variance.

To correct for this, statisticians divide by n − 1 instead of n when calculating sample variance. This adjustment, known as Bessel’s correction, accounts for the loss of one degree of freedom because the sample mean itself is estimated from the data rather than being a known, fixed value. The result is an unbiased estimate of the population variance – a small but important detail if you’re ever working with survey data or experimental samples rather than complete population data.

Coefficient of variation: comparing apples and oranges 🔗

Standard deviation has one limitation: it’s an absolute measure expressed in the original units of the data. This makes it impossible to directly compare the variability of, say, height (measured in centimetres) with weight (measured in kilograms). It also makes it hard to compare two datasets with very different means.

The coefficient of variation (C.V.) solves this by expressing standard deviation as a percentage of the mean:

C.V. = (σ / x̄) × 100

A higher C.V. signals greater relative variability – less consistency in the data – while a lower C.V. signals more uniformity, regardless of the units involved.

A real example: monsoon rainfall 🔗

Meteorologists use exactly this logic to compare rainfall reliability across seasons and regions. Data from the India Meteorological Department shows that the northeast monsoon, which affects south peninsular India between October and December, has a coefficient of variation of around 25 percent – noticeably higher than that of the southwest monsoon. In practical terms, this means northeast monsoon rainfall is far less predictable year to year, which directly affects agricultural planning and water resource management in the affected regions. This is the coefficient of variation doing real work: comparing variability across two rainfall systems with different average totals, on a single common scale.

Skewness: when a distribution loses its symmetry 🔗

Skewness describes the extent to which a distribution deviates from perfect symmetry. In a perfectly symmetrical distribution, the mean, median, and mode all coincide at the same point, and the curve looks identical on both sides of the center.

Real-world data rarely behaves this neatly. Distributions typically lean in one direction:

Positive skewness 🔗

A positively skewed distribution has a long tail stretching to the right. Here, Mean > Median > Mode. Income distribution in most economies is a classic example – a small number of very high earners pull the mean upward, while most people cluster at the lower end.

Negative skewness 🔗

A negatively skewed distribution has a long tail stretching to the left, where Mean < Median < Mode. Exam scores on a relatively easy test often show this pattern – most students score high, but a few very low scores drag the mean down below the median.

One of the most widely used measures here is Karl Pearson’s coefficient of skewness, calculated as (Mean − Mode) / Standard Deviation. Because a distribution’s mode isn’t always uniquely defined, this formula is sometimes adapted using the relationship Mode = 3(Median) − 2(Mean), which allows skewness to be estimated even when the mode is unclear. The sign of the resulting value tells you the direction of skew, while its magnitude tells you how pronounced it is – values typically range between −3 and +3 for real datasets, with most moderately skewed distributions falling closer to ±1.

Kurtosis: how peaked or flat is the distribution? 🔗

Knowing the center, spread, and skew of a distribution still doesn’t tell you everything about its shape. Kurtosis measures the relative peakedness or flatness of a distribution compared to a normal (bell-shaped) curve – essentially, how concentrated the data is around the mean versus how much of it sits out in the tails.

Distributions are classified into three types based on their kurtosis coefficient (β₂):

Mesokurtic: A normal distribution, where β₂ = 3. This is the reference point against which all other distributions are compared.

Leptokurtic: A distribution that is more peaked than normal, with β₂ > 3. Data is tightly concentrated around the mean, but the tails are “heavier,” meaning extreme values, while less frequent, are more pronounced when they do occur.

Platykurtic: A distribution flatter than normal, with β₂ < 3. Data is more evenly spread out, with lighter tails and fewer extreme outliers.

In essence, kurtosis is a measure of the combined weight of a distribution’s two tails relative to its peak. It’s particularly useful in fields like finance, where a leptokurtic return distribution warns of higher risk from rare but extreme market swings – something a simple standard deviation figure might not fully capture.

Putting the full picture together 🔗

None of these measures work in isolation. The mean tells you where the center is. Standard deviation and variance tell you how spread out the data is around that center. The coefficient of variation lets you compare that spread across datasets with different units or scales. Skewness tells you whether the data leans to one side. And kurtosis tells you how peaked or flat the shape actually is.

Two datasets can share an identical mean and standard deviation, yet look completely different once you factor in skewness and kurtosis – one could be a clean, symmetrical bell curve, while the other is lopsided with a sharp peak and heavy tails. A complete statistical analysis, whether it’s exam performance, rainfall patterns, or stock returns, needs all of these measures working together rather than any single one in isolation.

What do you think? If two Indian states report identical average annual rainfall but very different coefficients of variation, which one would you trust more for planning irrigation infrastructure – and why does the shape of a distribution (its skewness or kurtosis) matter just as much as its spread when making that kind of decision?

References

1. https://libguides.lib.miamioh.edu/data_analysis/dispersion
2. https://mausam.imd.gov.in/responsive/pdf_viewer_css/met1/Chapter-1%20page%205-9/Chapter-1%20page%205-9.pdf
3. https://egyankosh.ac.in/bitstream/123456789/73730/1/Unit-4.pdf
4. https://ebooks.inflibnet.ac.in/mgmtp15/chapter/measures-of-dispersion-skewness-and-kurtosis/
5. https://www.datacamp.com/tutorial/understanding-skewness-and-kurtosis