The Normal Distribution and Z-Scores

The normal distribution is the single most important shape in statistics. It is the smooth, symmetric bell curve that appears when many small independent influences add together, and it is the foundation of confidence intervals, z-scores and most hypothesis tests. Understanding it turns a standard deviation from an abstract number into a precise statement about how likely a value is.

What makes the normal distribution so useful is its predictability. Once you know only two numbers — the mean and the standard deviation — you know the probability of landing in any range. This page covers the empirical rule, z-scores, and how analysts actually apply them.

What the bell curve is

A normal distribution is fully described by two parameters: the mean (μ), which sets the center, and the standard deviation (σ), which sets the width. The curve is symmetric about the mean, so the mean, median and mode all coincide there. Values near the mean are common; values far out in either tail are rare, and the tails thin out smoothly.

Change the mean and the whole curve slides left or right. Change the standard deviation and it gets narrower (small σ, tight cluster) or wider (large σ, more spread). Every normal distribution is the same shape, just rescaled — which is exactly what makes z-scores possible.

The 68-95-99.7 empirical rule

The empirical rule turns the standard deviation into a probability ruler. For any normal distribution:

within 1 standard deviation of the mean  ->  ~68% of values
within 2 standard deviations             ->  ~95% of values
within 3 standard deviations             ->  ~99.7% of values

Suppose adult heights in a population are approximately normal with mean μ = 170 cm and standard deviation σ = 10 cm. Then:

• About 68% of people are between 160 and 180 cm (170 ± 10).
• About 95% are between 150 and 190 cm (170 ± 20).
• About 99.7% are between 140 and 200 cm (170 ± 30).

This is why a value beyond three standard deviations is treated as a genuine anomaly: under a normal model, only about 0.3% of values ever get that far out.

Z-scores: a universal ruler

A z-score restates any value as its distance from the mean measured in standard deviations:

z = (value - mean) / standard deviation

A z-score of 0 is exactly average, +1.5 is one and a half standard deviations above the mean, and −2 is two below. Because the z-score strips away the original units, it lets you compare values from completely different scales — a test score and a height, say — on one common footing.

def z_score(value, mean, std):
    return (value - mean) / std

# height of 185 cm, mean 170, std 10
print(z_score(185, 170, 10))   # 1.5

# height of 155 cm
print(z_score(155, 170, 10))   # -1.5

A height of 185 cm has z = (185 − 170) / 10 = 1.5, meaning it sits 1.5 standard deviations above average — taller than most people but not extreme. Converting a variable to z-scores produces the standard normal distribution, a normal curve with mean 0 and standard deviation 1, which is the reference every normal calculation ultimately uses.

Turning z-scores into probabilities

With SciPy you can go beyond the round-number empirical rule to exact probabilities using the cumulative distribution function.

from scipy.stats import norm

# P(height <= 185) when mean=170, std=10  ->  z = 1.5
print(round(norm.cdf(1.5), 4))          # 0.9332

# P(height between 160 and 180) = within 1 std
print(round(norm.cdf(1) - norm.cdf(-1), 4))   # 0.6827

The second line recovers the 68% figure precisely: 0.6827. The first says about 93.3% of people are 185 cm or shorter, so only about 6.7% are taller. This is how percentile ranks on standardized tests are produced.

How analysts use it

The normal distribution underlies the machinery analysts rely on daily. Confidence intervals use z-values (1.96 for 95%) drawn from the standard normal. Outlier rules often flag points beyond ±3 standard deviations. Standardizing features to z-scores is a routine preprocessing step so that variables on different scales are comparable. And crucially, even when raw data is not normal, the central limit theorem makes sample means approximately normal — which is why normal-based tests work so widely.

Common mistakes

Assuming your raw data is normal. Income, durations and counts are usually skewed. The normal distribution's real importance is for sample means, not necessarily for the underlying values. Plot a histogram before assuming a bell shape.

Misreading the empirical rule as exact. The 68-95-99.7 figures are rounded and apply only to a truly normal distribution. Real data approximates them at best.

Forgetting z-scores can be negative. A negative z-score simply means below the mean. Treating the sign as an error loses half the information.

Confusing standard deviation with standard error. Standard deviation describes spread of individual values; standard error describes spread of the sample mean. They differ by a factor of the square root of the sample size, a distinction covered in the central limit theorem tutorial.

In interviews

The normal distribution is a favorite interview topic. Expect "State the 68-95-99.7 rule" and "What is a z-score and how do you compute it?" A common applied question gives you a mean and standard deviation and asks what percentage of values fall in a range, which you answer with the empirical rule. You may also be asked "Is real data usually normal?" — the strong answer distinguishes raw data (often not) from sample means (approximately normal by the central limit theorem). Knowing z = 1.96 for a 95% interval is a nice detail to have ready.

Where this fits in your learning path

The normal distribution ties together variance and standard deviation, which supply the σ it depends on, and the broader family in probability distributions. It leads straight into the central limit theorem, the result that explains its dominance. All of it forms the inference backbone of the data analytics learning path, and it recurs throughout the statistics work in the data analyst roadmap.