Why the Average is a Liar: A Guide to Measuring Data Spread

We rely on the average (mean) to summarise data because it is incredibly convenient to represent a whole group with just one number. But averages have a dark secret: they frequently lie by omission.

Imagine two students, John and Arun. They both have an average score of 62 across their subjects.

• John’s scores: 60, 65, 62, 61, 62. He is consistent; he’s a solid 62-level student across the board.
• Arun’s scores: 100, 20, 95, 30, 65. He is brilliant at math and programming, but terrible at chemistry and history.

Judging them strictly by their identical averages completely erases Arun’s extreme highs and lows. To get the real story, we need a second number: a measure of spread. Spread tells us how far the individual data points deviate from the center.

Here are the four ways we measure spread, from the simplest to the most bulletproof.

1. The Range: Quick but Fragile

The simplest way to measure spread is to subtract the absolute lowest value from the highest value.

• John’s range: 65 – 60 = 5
• Arun’s range: 100 – 20 = 80

This instantly tells us John is far more consistent. However, the Range is incredibly fragile. Because it only uses the two most extreme numbers, a single typo or freak outlier (like Arun scoring a 5 on one quiz because he was sick) will completely warp the result, making the data look wildly scattered when it really isn’t.

2. Variance: The Problem with Negatives

If we want a better measure, why not just find out how far every single score is from the mean, and then take the average of those differences?

Let’s try it. If the mean is 10, and a data point is 15, the deviation is +5. If a data point is 5, the deviation is -5.

Here is the problem: if you add up all those deviations, the positives and negatives will perfectly cancel each other out, leaving you with zero. It would falsely look like your data has no spread at all.

To fix this, statisticians square the deviations. Squaring a negative number makes it positive ($-5 \times -5 = 25$). When you take the average of these squared deviations, you get the Variance.

But in solving one mathematical problem, we created a logical one. What is a “squared test score”? What is a “squared Rupee”? The unit of variance is mathematically meaningless to humans.

Read this for more on why Variance is squared.

3. Standard Deviation: Returning to Reality

To bring the Variance back to a unit we can actually understand, we just undo the square we did earlier. We take the square root of the Variance.

This magical number is the Standard Deviation (SD).

If John’s average score is 62 with a Standard Deviation of 2, it means the vast majority of his test scores fall just 2 points away from his average. It is the gold standard for measuring dispersion because it uses every single data point, but brings the final number back down to reality.

4. Inter-Quartile Range (IQR): The Outlier-Proof Vest

Standard Deviation is great, but because it is based on the mean, it is still slightly vulnerable to massive outliers.

If you want a measure of spread that completely ignores extreme weirdness, you use the Inter-Quartile Range (IQR).

Imagine lining up all your data from smallest to largest and chopping it into four equal quarters (quartiles).

• The first 25% are your lowest scores.
• The last 25% are your highest scores.
• The middle 50% is the meaty center of your data.

The IQR is simply the range of that middle 50% (calculated by subtracting the 25th percentile from the 75th percentile). By completely throwing away the bottom quarter and the top quarter, the IQR essentially builds a fence that ignores outliers entirely.

Use this interactive sandbox to see how a single crazy data point affects these different measurements:

Spread Measures Explorer