You closed the previous lesson with a warning: two stores can have the same average sales and behave in completely different ways. Marta now turns that into an actual assignment: "Zaragoza-Centro and Bilbao-Casco take in the same on average, about €44,000 a day, but the operations director swears Bilbao is impossible to plan for. Prove it to me with numbers." To answer, you need the measures of dispersion, which quantify how far the data stray from their center. In this lesson you will learn the range, the variance, the standard deviation (and why the sample version divides by \( n-1 \)) and the coefficient of variation, which lets you compare the variability of things measured on different scales. And you will see why, in business, dispersion almost always translates into a single word: risk.

1. Why the mean is not enough
2. The range
3. Deviations from the mean
4. The variance
5. The standard deviation
6. Population (\(\sigma\)) versus sample (\(s\)): the reason for n−1
7. The coefficient of variation
8. Reading dispersion like a manager

These are the daily sales (in euros) of the two stores over the same 5-day working week:

Check the means: \( 220{,}000 / 5 = 44{,}000 \) euros in both cases. If you only looked at the mean (or even the median, which is also €44,000 for both), you would call the stores twins. But look at the data: Zaragoza moves at most €1,000 up or down; Bilbao swings by €6,000. The operations director is right, and your job is to put a number on that difference.

A measure of central tendency plus a measure of dispersion form the minimum honest summary of a quantitative variable. Either one without the other is incomplete information.

The simplest measure of dispersion is the range: the difference between the maximum and minimum values.

\[ R = x_{\max} - x_{\min} \]

• Zaragoza-Centro: \( 45{,}000 - 43{,}000 = \text{€}2{,}000 \)
• Bilbao-Casco: \( 50{,}000 - 38{,}000 = \text{€}12{,}000 \)

It already tells the two stores apart, and it takes two seconds to compute. But it has serious limitations:

• It uses only two data points (the extremes) and ignores everything else: it cannot tell whether the remaining values hug the mean or spread out.
• It is extremely sensitive to an outlier: a single exceptional day (a big snowstorm, a viral promotion) sends the range through the roof.
• It tends to grow with sample size: the more days you observe, the easier it is to catch an extreme, so comparing ranges of samples of different sizes is unfair.

Use it as a quick first look, never as a definitive measure. (There is a robust variant, the interquartile range, which you will meet in the lesson on Measures of Position and Outliers.)

The natural idea for measuring dispersion is to ask: how far is each data point from the mean? That distance is the deviation: \( x_i - \bar{x} \).

Let's compute them for Zaragoza-Centro (\( \bar{x} = 44{,}000 \)):

First temptation: average the deviations. But their sum is always exactly zero — that is the "balance point" property of the mean you saw in the previous lesson — so their average always comes out to 0 and measures nothing. The negative deviations cancel the positive ones.

The classic solution is to square each deviation before averaging: squares are always positive, so they no longer cancel, and on top of that they penalize large deviations more heavily (a deviation of 2,000 weighs four times as much as one of 1,000, not twice as much). That is where the variance comes from.

The variance is the average of the squared deviations. For a population of \( N \) values with mean \( \mu \):

\[ \sigma^2 = \frac{\sum_{i=1}^{N} (x_i - \mu)^2}{N} \]

For a sample of \( n \) values with mean \( \bar{x} \) (note the denominator — we will justify it shortly):

\[ s^2 = \frac{\sum_{i=1}^{n} (x_i - \bar{x})^2}{n-1} \]

We treat the 5 days as a sample of how the store behaves (we want to talk about the store in general, not just that week).

Step 1 — Deviations and their squares:

Step 2 — Divide by \( n - 1 = 4 \):

\[ s^2 = \frac{2{,}500{,}000}{4} = 625{,}000 \text{ €}^2 \]

\[ s^2 = \frac{90{,}000{,}000}{4} = 22{,}500{,}000 \text{ €}^2 \]

Bilbao's variance is 36 times Zaragoza's. The difference could not be clearer… but what does "625,000 euros squared" mean? Nothing interpretable: when we squared the deviations, the units got squared too. The variance is a fundamental technical building block (it will keep reappearing in inference), but it is not the measure you communicate to Marta.

The standard deviation undoes the squaring: it is the square root of the variance, and it is back in the original units.

\[ \sigma = \sqrt{\sigma^2} \qquad \text{(population)} \qquad \qquad s = \sqrt{s^2} \qquad \text{(sample)} \]

• Zaragoza-Centro: \( s = \sqrt{625{,}000} \approx \text{€}790.6 \)
• Bilbao-Casco: \( s = \sqrt{22{,}500{,}000} \approx \text{€}4{,}743.4 \)

Practical interpretation: the standard deviation is, roughly, the "typical" distance of any given data point from the mean. A normal day in Zaragoza strays about €800 from its €44,000; a normal day in Bilbao, almost €4,750. Now you have the sentence for Marta: "Both stores average €44,000 a day, but Bilbao's standard deviation (≈ €4,743) is about six times Zaragoza's (≈ €791): same mean, very different risk."

Two notes:

• The standard deviation is never negative, and it equals 0 only if all the data points are identical.
• When the data follow the famous "normal curve", the standard deviation supports very concrete rules of the form "95% of days fall within 2 deviations of the mean". You will see that empirical rule in detail in the Normal Distribution lesson; for now, hold on to the "typical distance" reading.
• Because it is built on the mean, it inherits the mean's sensitivity to extreme values: a single wild data point inflates \( s \) considerably. Keep that in mind when we deal with outliers in the next lesson.

You will have noticed the denominator by now: \( N \) for the population, \( n-1 \) for the sample. It is not a whim.

When to use each:

The intuition behind n−1. When you work with a sample you do not know the true population mean \( \mu \); you use the sample mean \( \bar{x} \) instead, computed from the very same data. And \( \bar{x} \) is, by construction, the value that makes the squared deviations as small as possible: the data are always closer to their own mean than to the true mean \( \mu \). The result: if you divide by \( n \), you systematically underestimate the population variance. Dividing by \( n-1 \) enlarges the result by just enough to correct that bias (which is why \( s^2 \) is called the sample variance, or corrected variance).

Another way to see it: of the \( n \) deviations, only \( n-1 \) are "free", because once the mean is known, the last deviation is forced (they must sum to zero). We say there are \( n-1 \) degrees of freedom, a concept that will resurface in Module 5.

With our Zaragoza data: \( \sigma = \sqrt{2{,}500{,}000/5} = \sqrt{500{,}000} \approx \text{€}707.1 \) versus \( s \approx \text{€}790.6 \). With \( n = 5 \) the difference is noticeable (about 12%); with \( n = 500 \) it would be irrelevant. Rule of thumb: in business you almost always work with samples → use \( n-1 \) by default (it is what spreadsheets do with their "sample" functions); the difference only matters with small samples.

The standard deviation is expressed in the variable's own units, and that — an advantage for interpretation — becomes a problem when you want to compare the dispersion of variables with different scales or units. Which varies more: receipt amounts or member satisfaction? Comparing €21.50 with 1.6 points is comparing apples and oranges.

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

\[ CV = \frac{s}{\bar{x}} \times 100 \]

• Zaragoza-Centro: \( CV = 790.6 / 44{,}000 \times 100 \approx 1.8,% \)
• Bilbao-Casco: \( CV = 4{,}743.4 / 44{,}000 \times 100 \approx 10.8,% \)

Here the means were equal and the CV adds little. Its real value shows when the scales differ.

With NovaMarket's consolidated figures:

Reading: although the standard deviation of daily sales (€6,200) is huge compared with that of receipts (€21.50), in relative terms receipts are far more variable (66% of their mean versus 14%). It makes perfect business sense: a single receipt can be €3 or €300, but a whole store averages thousands of purchases a day and its totals are proportionally more stable.

Cautions with the CV:

• It only makes sense for variables on a ratio scale (a true zero and positive values): euros, times, units. With temperatures in °C or with 0–10 scales its interpretation is debatable (satisfaction is in the table as an illustration; treat it with care).
• If the mean is close to zero, the CV blows up and stops being interpretable.

Always translate dispersion into the language of the business. With the Zaragoza/Bilbao case:

• Staff planning: Zaragoza can size its checkout and shelf-stacking teams almost on autopilot; Bilbao needs flexible shifts to avoid paying idle hours on quiet days and being overwhelmed on strong ones.
• Stock management: demand variability forces Bilbao to hold more safety stock (more capital tied up and more shrinkage in fresh produce) to avoid stockouts on peak days.
• Cash flow and targets: with the same mean, Bilbao will miss its daily target on many more days (and beat it on just as many). Judging its managers by individual days would be unfair.
• Risk and service quality: in e-commerce delivery times, the mean matters less than the dispersion: a customer does not experience "the 24-hour average", they experience their delivery; high dispersion means many customers with bad experiences even if the mean looks good.
• A signal worth investigating: high dispersion is often a question in disguise: what makes the good and bad days (or stores, or orders) different? Spotting it is the first step towards the analyses that come later (group comparisons, regression…).

The sentence to engrave in your memory: same mean, different dispersion = different risk and different management.

• Reporting a mean without its dispersion. "Average transaction value: €32.40" tells half the story; "€32.40 with a standard deviation of €21.50" describes the actual business.
• Interpreting the variance directly. Its squared units (€²) mean nothing to the business; always communicate the standard deviation and keep the variance for the calculations.
• Mixing up \( \sigma \) and \( s \), or using the population formula on a small sample: you will underestimate the variability. When in doubt, \( n-1 \).
• Comparing standard deviations of variables with very different means or units. That is what the coefficient of variation is for.
• Forgetting that \( s \) is sensitive to outliers. A single erroneous value (a €3,240 receipt that was really €32.40) can multiply the standard deviation. Inspect the extremes before computing (next lesson).
• Tip: always check that the deviations sum to 0 (or nearly 0, due to rounding) before squaring; it is the cheapest arithmetic check there is.

Daily sales at Valencia-Ruzafa over the same week (in euros): 40,000 — 42,000 — 44,000 — 46,000 — 48,000.

1. Compute the mean, the sample variance \( s^2 \) and the standard deviation \( s \).
2. Place the store between Zaragoza-Centro (\( s \approx \text{€}791 \)) and Bilbao-Casco (\( s \approx \text{€}4{,}743 \)): which one does it resemble more in terms of stability?

Two logistics operators bid for the e-commerce delivery contract. In a trial with 60 orders each, both achieve a mean delivery time of 24 hours, but operator A has a standard deviation of 3 hours and operator B of 9 hours.

1. Without doing any arithmetic: what does that difference mean for the customers' experience?
2. Compute each operator's CV. Which one would you recommend to Marta, and with what management argument?

Receipt amounts at two stores: Madrid-Centro has a mean of €34.10 and a standard deviation of €24.50; Cuenca has a mean of €27.80 and a standard deviation of €12.30.

Which store has more variable receipts in absolute terms? And in relative terms? Compute both CVs and interpret.

Exercise 2.1:

1. Mean: \( 220{,}000/5 = \text{€}44{,}000 \). Deviations: −4,000, −2,000, 0, +2,000, +4,000 (they sum to 0 ✓). Squares: 16,000,000 + 4,000,000 + 0 + 4,000,000 + 16,000,000 = 40,000,000. Sample variance: \( s^2 = 40{,}000{,}000/4 = 10{,}000{,}000 \) €². Standard deviation: \( s = \sqrt{10{,}000{,}000} \approx \text{€}3{,}162.3 \).
2. With \( s \approx \text{€}3{,}162 \) (CV ≈ 7.2%), Valencia-Ruzafa sits between the two, considerably closer to Bilbao's volatile profile than to Zaragoza's stability. Common mistake: dividing by \( n = 5 \) instead of \( n-1 = 4 \) (which would give \( s \approx \text{€}2{,}828 \)); these are 5 sampled days, not the population.

Exercise 2.2:

1. With the same mean, operator B will have many more very early and, above all, very late deliveries: more customers waiting 33, 40 or more hours. The mean does not capture those bad experiences; the dispersion does.
2. \( CV_A = 3/24 \times 100 = 12.5,% \); \( CV_B = 9/24 \times 100 = 37.5,% \). Recommendation: operator A, because at equal cost and equal mean it offers a predictable service: fewer incidents, fewer complaints, and the possibility of promising customers reliable delivery windows. Common mistake: concluding that "they are the same because the mean is the same" — exactly the mistake this lesson teaches you to avoid.

Exercise 2.3:

In absolute terms, Madrid-Centro (€24.50 > €12.30). In relative terms too: \( CV_{Madrid} = 24.50/34.10 \times 100 \approx 71.8,% \) versus \( CV_{Cuenca} = 12.30/27.80 \times 100 \approx 44.2,% \). Interpretation: Madrid-Centro's customer base is more heterogeneous (it mixes small grab-and-go purchases with large shops, including hospitality businesses), while Cuenca has a more homogeneous purchasing pattern. Note: here both verdicts agree, but that is not always the case: when the means differ a lot, only the CV allows a fair comparison.

You now know how to measure what the mean hides: the range as a quick glance, the variance as a technical building block, the standard deviation as the "typical distance from the mean" in real units (dividing by \( n-1 \) when you work with samples) and the coefficient of variation for comparing variability across different scales. And you know how to translate it into Marta's language: same mean with different dispersion means different risk, different stock, different staffing.

The mean and the standard deviation describe the center and the width of the data, but they still do not tell you where one specific data point falls within the set: is a €95 receipt merely notable or truly extraordinary? Beyond what delivery time is an order a rare case worth investigating? To answer, you need the measures of position — quartiles, percentiles, z-scores — and the tools for detecting outliers. That is the content of the next lesson: Measures of Position and Outliers.