13 The Normal Distribution
The Binomial and Poisson distributions from the last chapter describe countable outcomes. Many real-world business measurements, transaction amounts, exam scores, delivery times, product weights, are continuous instead, able to take any value within a range. The Normal distribution (the familiar bell curve) is by far the most important continuous distribution in statistics: countless natural and business processes approximate it, and it is the foundation almost every technique in the rest of this book, sampling, estimation, and hypothesis testing, ultimately relies on.
13.1 Properties of the Normal Distribution
The Normal distribution is a smooth, symmetric, bell-shaped curve completely described by just two parameters: its mean \(\mu\) (the center of the curve) and its standard deviation \(\sigma\) (the spread of the curve).
- Symmetric: The left and right halves are mirror images; mean, median, and mode all coincide at \(\mu\). This is exactly what zero skewness (Chapter 7) and near-zero excess kurtosis (Chapter 8) describe.
- Bell-shaped: Most values cluster near the mean, with probability tapering off symmetrically in both directions.
- Defined by \(\mu\) and \(\sigma\): Changing \(\mu\) shifts the curve left or right; changing \(\sigma\) makes it narrower (small \(\sigma\)) or wider (large \(\sigma\)), without changing its bell shape.
- Never touches zero: The curve extends infinitely in both directions, though probability becomes vanishingly small far from the mean.
13.2 The Empirical Rule
For any Normal distribution, a fixed proportion of the data always falls within a given number of standard deviations of the mean, regardless of the specific values of \(\mu\) and \(\sigma\). This is known as the empirical rule (or the 68-95-99.7 rule):
\[P(\mu - 1\sigma < X < \mu + 1\sigma) \approx 68\% \qquad P(\mu - 2\sigma < X < \mu + 2\sigma) \approx 95\%\]
\[P(\mu - 3\sigma < X < \mu + 3\sigma) \approx 99.7\%\]
The empirical rule is why a value more than roughly 3 standard deviations from the mean is so often flagged as an outlier or anomaly: under a Normal distribution, fewer than 0.3% of values should ever land that far out, so seeing one is a signal worth investigating.
13.3 The Standard Normal Distribution and Z-Scores
Every Normal distribution has its own \(\mu\) and \(\sigma\), which makes them hard to compare directly. The standard normal distribution solves this: it is the one specific Normal distribution with \(\mu = 0\) and \(\sigma = 1\). Any value \(x\) from any Normal distribution can be converted into a z-score, its standardized distance from the mean, measured in standard deviations:
\[z = \frac{x - \mu}{\sigma}\]
Once standardized, a z-score’s probability can be looked up from the standard normal distribution, either from a printed z-table or, far more conveniently, computed directly with software.
Example
A retailer’s monthly customer transaction amounts are approximately Normal with mean \(\mu = ₹5{,}000\) and standard deviation \(\sigma = ₹800\).
What is the probability a random transaction exceeds ₹6,000? First standardize: \(z = \frac{6000-5000}{800} = 1.25\). Looking up \(P(Z > 1.25)\) gives:
\[P(X > 6000) = P(Z > 1.25) \approx 0.1056\]
About 10.6% of transactions exceed ₹6,000.
What is the probability a transaction falls between ₹4,000 and ₹6,000? These correspond to \(z = -1.25\) and \(z = 1.25\):
\[P(4000 < X < 6000) = P(-1.25 < Z < 1.25) \approx 0.7887\]
About 78.9% of transactions fall in this range, illustrating the empirical rule: \(\pm 1.25\sigma\) captures a bit less than the \(\pm 2\sigma \approx 95\%\) figure and a bit more than the \(\pm 1\sigma \approx 68\%\) figure, right where it should.
What transaction amount marks the top 5% of spenders? This asks for the value with only 5% of the distribution above it, the 95th percentile, which corresponds to \(z \approx 1.645\):
\[x = \mu + z\sigma = 5000 + (1.645)(800) \approx ₹6{,}316\]
13.4 Why the Normal Distribution Matters So Much
- It shows up naturally: Many real measurements, heights, exam scores, measurement errors, are approximately Normal because they result from many small, independent influences adding together.
- The Central Limit Theorem: Even when the underlying data is not Normal, the distribution of a sample mean becomes approximately Normal as sample size grows, a result covered in the sampling chapter later in this book, and the single biggest reason the Normal distribution underpins nearly all of inferential statistics.
- It approximates other distributions: A Binomial distribution with large \(n\), or a Poisson distribution with large \(\lambda\), both start to look approximately Normal.
Not all business data is Normal, and it is a mistake to assume it is without checking. Highly skewed data (Chapter 7), such as income or transaction amounts with a long right tail, often needs a transformation, or a different distribution entirely, before Normal-based methods can be trusted. Always compare the actual skewness and kurtosis of a dataset against 0 before assuming normality.
Recap
Chapters 9 through 12 have now built the complete probability foundation of this book: the rules for combining probabilities, conditional probability and Bayes’ Theorem for updating beliefs with evidence, and both discrete (Binomial, Poisson) and continuous (Normal) distributions for describing entire ranges of outcomes. Module IV, coming next, builds directly on this foundation: sampling distributions, estimation, and hypothesis testing, the tools that let a business draw reliable conclusions about an entire population from just a sample of it.
Summary
| Concept | Description |
|---|---|
| Core Idea | |
| Normal Distribution | A continuous, symmetric, bell-shaped distribution describing many real-world measurements |
| Symmetric and Bell-Shaped | Left and right halves mirror each other; mean, median, and mode all coincide at the center |
| Parameters (mu and sigma) | The mean sets the center of the curve; the standard deviation sets its spread |
| The Empirical Rule | |
| Empirical Rule (68-95-99.7) | About 68%, 95%, and 99.7% of values fall within 1, 2, and 3 standard deviations of the mean |
| Standardization | |
| Standard Normal Distribution | The one specific Normal distribution with mean 0 and standard deviation 1 |
| Z-Score | A value's distance from the mean, measured in standard deviations |
| Standardization | Converting any Normal value to a z-score so it can be compared using the standard normal distribution |
| Why It Matters | |
| Why the Normal Distribution Matters | Many real processes approximate it naturally, and it underpins nearly all inferential statistics |
| Central Limit Theorem (Preview) | Sample means become approximately Normal as sample size grows, even from non-Normal data |
| When Normality Does Not Hold | Highly skewed data may need transformation or a different distribution before assuming normality |