Measures of Dispersion

Understand range, variance, standard deviation, IQR, and coefficient of variation β€” how to measure the spread of data.

22 min read
Beginner

Why Central Tendency Isn't Enough

Consider two classes that both scored an average of 75 on an exam:

Class A: 74, 75, 75, 76, 75 β€” everyone scored about the same Class B: 30, 60, 90, 95, 100 β€” wildly different scores

Same mean. Completely different stories. The mean alone hides a crucial dimension: how spread out the data is. This is what measures of dispersion capture.

Range: The Simplest Measure

The difference between the largest and smallest values.

Range=xmaxβ‘βˆ’xmin⁑\text{Range} = x_{\max} - x_{\min}

Class A range: 76 - 74 = 2 Class B range: 100 - 30 = 70

Simple and intuitive, but terrible as a sole measure because:

  • It only uses two data points (the extremes)
  • A single outlier can massively inflate it
  • It tells you nothing about how data is distributed between the extremes

Variance: The Foundation

Variance answers: "On average, how far are the data points from the mean?"

But there's a subtlety. If we just average the raw differences from the mean, positives and negatives would cancel out (points above and below the mean would sum to zero). So we square the differences first.

Used when you have data for the entire population:

Οƒ2=1Nβˆ‘i=1N(xiβˆ’ΞΌ)2\sigma^2 = \frac{1}{N}\sum_{i=1}^{N}(x_i - \mu)^2

Used when you have a sample from a larger population. Note the n-1 in the denominator:

s2=1nβˆ’1βˆ‘i=1n(xiβˆ’xΛ‰)2s^2 = \frac{1}{n-1}\sum_{i=1}^{n}(x_i - \bar{x})^2
Computing Variance Step by Step

Data: 4, 8, 6, 5, 7 (sample)

Step 1: Find the mean: (4+8+6+5+7)/5 = 30/5 = 6

Step 2: Find each deviation from the mean:

  • 4 - 6 = -2
  • 8 - 6 = +2
  • 6 - 6 = 0
  • 5 - 6 = -1
  • 7 - 6 = +1

Step 3: Square each deviation: 4, 4, 0, 1, 1

Step 4: Sum the squared deviations: 4+4+0+1+1 = 10

Step 5: Divide by n-1 (sample): 10/4 = 2.5

Sample variance sΒ² = 2.5

Why n-1 instead of n? This is called Bessel's correction. A sample tends to underestimate the population's spread (sample points cluster closer to their own mean than to the true population mean). Dividing by n-1 corrects this bias. It's one of those beautiful mathematical details where a small tweak makes a big difference.

Why Variance Squares the Differences

Students often ask: "Why not just use absolute differences instead of squaring?" Great question. There are several reasons:

  1. Mathematical convenience β€” Squared functions are differentiable everywhere; absolute values create kinks that are harder to work with in calculus.

  2. Penalizes large deviations more β€” A point that's 10 units away contributes 100 to variance, not just 10. This makes variance more sensitive to outliers, which is sometimes desirable.

  3. Connects to geometry β€” Variance relates to the Pythagorean theorem and Euclidean distance in higher dimensions.

  4. Decomposes nicely β€” Total variance can be broken into components (this becomes crucial in ANOVA and regression).

That said, the alternative (mean absolute deviation) does exist and is sometimes used. It's just less common because variance plays nicer with the rest of statistics.

Standard Deviation: Variance in Human-Readable Units

Variance has a problem: its units are squared. If your data is in meters, variance is in metersΒ². That's hard to interpret.

Standard deviation fixes this by taking the square root of variance:

Οƒ=Οƒ2(population)s=s2(sample)\sigma = \sqrt{\sigma^2} \qquad \text{(population)} \qquad s = \sqrt{s^2} \qquad \text{(sample)}

From our example: s = √2.5 β‰ˆ 1.58

This means, roughly speaking, data points are about 1.58 units away from the mean on average. Same units as the original data β€” much more interpretable!

Standard deviation is the most commonly reported measure of spread. When someone says "the average is 100 with a standard deviation of 15," you immediately know that most values fall between about 85 and 115.

Interquartile Range (IQR)

Just like the median is a robust alternative to the mean, the IQR is a robust alternative to standard deviation.

  • Q1 (25th percentile): 25% of data falls below this value
  • Q2 (50th percentile): The median
  • Q3 (75th percentile): 75% of data falls below this value
  • IQR = Q3 - Q1: The range of the middle 50% of the data
IQR=Q3βˆ’Q1\text{IQR} = Q_3 - Q_1

The IQR ignores the extremes entirely, making it resistant to outliers. It's the foundation of box plots and the standard method for detecting outliers:

Outlier rule: Any value below Q1 - 1.5Γ—IQR or above Q3 + 1.5Γ—IQR is considered a potential outlier.

Coefficient of Variation: Comparing Apples to Oranges

How do you compare the spread of two datasets measured in different units? Enter the Coefficient of Variation (CV):

CV=sxˉ×100%\text{CV} = \frac{s}{\bar{x}} \times 100\%
Comparing Variability Across Scales

Heights of adults: Mean = 170 cm, SD = 10 cm β†’ CV = 5.9% Weights of adults: Mean = 70 kg, SD = 12 kg β†’ CV = 17.1%

Even though you can't directly compare centimeters to kilograms, the CV tells you that weight is relatively more variable than height in this population.

CV only works for ratio-scale data (where zero means "none"). It's meaningless for temperature in Celsius (0Β°C doesn't mean "no temperature").

Choosing the Right Measure

Dispersion Measures Compared
Measure
Pros
Cons
Best For
RangeDead simpleIgnores all but 2 valuesQuick overview
VarianceMathematical foundationSquared units, hard to interpretCalculations, formulas
Std DeviationSame units as dataSensitive to outliersGeneral-purpose reporting
IQRRobust to outliersIgnores 50% of dataSkewed data, box plots
CVUnitless comparisonOnly for ratio scalesComparing across scales

Test your knowledge

🧠 Knowledge Check
1 / 3

Why do we divide by n-1 instead of n when computing sample variance?

πŸ“Š

Statistics Fundamentals

Master statistical thinking from the ground up. Learn to reason about data, uncertainty, inference, and decision-making with an intuition-first approach.