Normal Distribution & z-Scores — Use normal distribution and z-scores for probabilities - Biostatistics

SECTION 1

Historical Context & Motivation

The normal distribution is arguably the most consequential probability model in the history of quantitative science. Its familiar bell-shaped curve underpins clinical trial analysis, epidemiological modeling, and quality control in pharmaceutical manufacturing. Yet the distribution was not born in medicine—it arose from the practical problem of reconciling discrepant astronomical measurements in the eighteenth century, and its mathematical elegance soon attracted attention from mathematicians, actuaries, and eventually biostatisticians who recognized that many biological variables cluster symmetrically around a central tendency.

1733 • De Moivre's Approximation

Abraham de Moivre derived the bell curve as an approximation to the binomial distribution for large sample sizes, establishing the earliest analytical form of the normal density function in his work The Doctrine of Chances.

1809 • Gauss & the Method of Least Squares

Carl Friedrich Gauss used the normal distribution to justify the method of least squares in astronomical orbit-fitting, leading to the distribution's alternate name—the Gaussian distribution.

1835 • Quetelet's 'Average Man'

Adolphe Quetelet applied the normal curve to human chest measurements and heights, pioneering the quantitative study of biological variation and the concept of the 'average man' (l'homme moyen).

1893 • Karl Pearson & Modern Statistics

Karl Pearson formalized hypothesis testing using the normal distribution and introduced the term 'standard deviation,' creating the inferential framework that biostatisticians still employ for comparing treatment groups and assessing population health.

1925 • Fisher's Statistical Methods

Ronald A. Fisher published Statistical Methods for Research Workers, cementing the normal distribution and z-scores as foundational tools for experimental biology, agriculture, and medicine.

The central question that motivated these developments remains pressing in modern biostatistics: given a set of biological measurements—blood pressures, serum cholesterol levels, gene expression values—how do we determine whether a particular observation is typical, unusual, or extreme? The normal distribution paired with the z-score transformation provides a principled, universally applicable answer.

SECTION 2

Core Principles & Definitions

Before computing probabilities, it is essential to internalize several foundational properties of the normal distribution and the logic behind standardizing raw data. These principles form the conceptual scaffold on which all subsequent calculations rest. A normally distributed variable is fully characterized by just two parameters—its mean (μ) and its standard deviation (σ)—and every normal curve, regardless of scale, can be mapped onto a single reference curve through the z-score transformation.

Symmetry & Unimodality

The normal density is perfectly symmetric about its mean. The mean, median, and mode coincide at the center, producing a single peak (unimodal). Approximately 68% of observations fall within ±1σ of the mean.

The Empirical Rule (68–95–99.7)

For any normal distribution, roughly 68% of data lie within ±1σ, 95% within ±2σ, and 99.7% within ±3σ. This rule enables rapid probability estimation before consulting a z-table.

Asymptotic Tails

The tails of the normal curve extend infinitely in both directions, approaching but never reaching zero. This means any value is theoretically possible, though extreme values have vanishingly small probability densities.

Standard Normal Distribution

The standard normal distribution is a special case with μ = 0 and σ = 1, denoted Z ~ N(0, 1). It serves as the universal reference for all z-score probability lookups.

z-Score as a Standardized Measure

A z-score expresses how many standard deviations an observation lies above or below the mean. Positive z-scores indicate values above the mean; negative z-scores indicate values below.

✦

KEY TAKEAWAY

Think of the z-score transformation as converting currencies. If one researcher reports systolic blood pressure in mmHg and another reports BMI in kg/m², you cannot directly compare how extreme each measurement is until you 'convert' both to the same currency—standard deviations from the mean. The z-score is that universal currency: a z of +2.0 always means 'two standard deviations above the mean,' regardless of the original measurement units.

SECTION 3

Visual Explanation — The Bell Curve & Empirical Rule

The standard normal bell curve with the 68–95–99.7 empirical rule regions shaded. The violet band marks ±1σ (68.27%), the pink band marks ±2σ (95.45%), and the full span to ±3σ captures 99.73% of all observations.

The diagram above illustrates the core geometry of the normal distribution. Notice how the curve is perfectly symmetric about μ (the mean) and how the probability density tapers smoothly toward the tails. The shaded regions correspond to the empirical rule: approximately 68% of a normally distributed population falls within one standard deviation of the mean, 95% within two, and 99.7% within three. In biostatistical practice, these percentages provide an immediate intuition about the rarity of any given observation. For instance, a laboratory value more than 2σ above the mean occurs in fewer than 2.5% of the reference population—a threshold frequently used to flag abnormal clinical results.

When we move from a general normal distribution X ~ N(μ, σ²) to the standard normal Z ~ N(0, 1), the x-axis labels change from raw units (e.g., mg/dL) to dimensionless z-scores. This is precisely the transformation that makes the z-table universally applicable: once you compute the z-score for any observation, you can read off its cumulative probability directly from a single table or software function.

SECTION 4

Mathematical Framework

The mathematical backbone of normal distribution probability calculations involves three key expressions: the probability density function (PDF), the z-score transformation, and the cumulative distribution function (CDF). Together, they allow us to convert any raw measurement on a normally distributed variable into a precise probability statement.

NORMAL PROBABILITY DENSITY FUNCTION

f(x) = (1 / (σ√(2π))) × e^(−(x − μ)² / (2σ²))

Where x is the observed value, μ is the population mean, σ is the population standard deviation, and e is Euler's number (≈ 2.71828). The PDF gives the height of the density curve at any point x, not a probability itself. Probabilities require integration over an interval.

Z-SCORE TRANSFORMATION

z = (x − μ) / σ

This formula converts a raw score x from any normal distribution N(μ, σ²) into the standard normal scale N(0, 1). A positive z indicates the observation is above the mean; a negative z indicates it is below. The magnitude tells you how many standard deviations separate the observation from the center of the distribution.

CUMULATIVE DISTRIBUTION FUNCTION (CDF)

Φ(z) = P(Z ≤ z) = ∫ from −∞ to z of (1/√(2π)) × e^(−t²/2) dt

The CDF Φ(z) gives the probability that a standard normal variable takes a value less than or equal to z. This integral has no closed-form solution and is evaluated numerically—via z-tables, software (R's pnorm(), Python's scipy.stats.norm.cdf()), or calculator functions.

💡

Deriving Tail & Interval Probabilities

Because the total area under the curve is 1, the upper-tail probability is P(Z > z) = 1 − Φ(z). For an interval, P(a < Z < b) = Φ(b) − Φ(a). By symmetry, Φ(−z) = 1 − Φ(z). These three identities, combined with the z-score formula, allow you to answer virtually any probability question involving a normal distribution.

SECTION 5

Using the z-Table — A Detailed Breakdown

The standard normal table (z-table) is a lookup tool that maps z-scores to cumulative probabilities Φ(z). Most tables are structured so that the left column lists the z-score to the first decimal place and the top row provides the second decimal place (the hundredths digit). The intersection gives Φ(z). Understanding this layout is critical because biostatistics exams and many clinical reference tools still rely on table lookups.

An excerpt from a standard normal z-table. To look up z = 1.00, find row 1.0 and column .00; the cell (highlighted in cyan) reads 0.8413, meaning 84.13% of the standard normal distribution lies at or below z = 1.00.

Common Probability Scenarios

Question Type	Formula	Example (z = 1.50)
P(Z ≤ z) — left tail	Φ(z) — read directly from table	Φ(1.50) = 0.9332
P(Z > z) — right tail	1 − Φ(z)	1 − 0.9332 = 0.0668
P(a < Z < b) — between	Φ(b) − Φ(a)	P(−1.00 < Z < 1.50) = 0.9332 − 0.1587 = 0.7745
P(Z ≤ −z) — negative z	1 − Φ(z) (by symmetry)	Φ(−1.50) = 1 − 0.9332 = 0.0668

Four standard probability query types and their solutions using the z-table.

SECTION 6

Worked Example — Fasting Blood Glucose

A large epidemiological study reports that fasting blood glucose (FBG) in a healthy adult population is normally distributed with μ = 90 mg/dL and σ = 10 mg/dL. A clinician wants to know the probability that a randomly selected individual from this population has an FBG above 110 mg/dL—a threshold sometimes flagged for pre-diabetes screening.

P(FBG > 110 mg/dL)

1
Step 1 — Identify Given Values

We are given μ = 90 mg/dL, σ = 10 mg/dL, and we want to find P(X > 110). The variable X represents fasting blood glucose, and we assume X ~ N(90, 10²).

μ = 90, σ = 10, x = 110
2
Step 2 — Compute the z-Score

Apply the z-score formula: z = (x − μ) / σ = (110 − 90) / 10 = 20 / 10 = 2.00. This tells us that 110 mg/dL lies exactly two standard deviations above the mean.

z = 2.00
3
Step 3 — Look Up Φ(z) in the z-Table

From the standard normal table (row 2.0, column .00), we find Φ(2.00) = 0.9772. This is the cumulative probability P(Z ≤ 2.00)—the area under the curve to the left of z = 2.00.

Φ(2.00) = 0.9772
4
Step 4 — Compute the Upper-Tail Probability

Because we want P(X > 110) = P(Z > 2.00), we subtract the cumulative probability from 1: P(Z > 2.00) = 1 − 0.9772 = 0.0228.

P(FBG > 110) = 0.0228 or 2.28%
5
Step 5 — Interpret in Context

Approximately 2.28% of healthy adults in this population would be expected to have a fasting blood glucose exceeding 110 mg/dL. The clinician can use this probability to assess whether an individual's result is unusually high relative to the healthy reference distribution, supporting decisions about further diagnostic testing.

About 1 in 44 healthy adults exceeds this threshold by chance.

SECTION 7

Strengths, Limitations & Common Pitfalls

Strengths	Limitations
Universality: The Central Limit Theorem guarantees that sample means approach normality for large n, even if the underlying population is not normal.	Assumption sensitivity: If the data are substantially skewed or heavy-tailed, normal-based probabilities can be misleading, especially in the tails.
Simplicity: Only two parameters (μ and σ) fully specify the distribution, making it parsimonious and easy to estimate from data.	Unbounded support: The normal model assigns nonzero probability to negative values, which is inappropriate for strictly positive variables (e.g., concentrations, survival times) without transformation.
Analytical tractability: Closed-form expressions for many derived statistics (confidence intervals, test statistics) depend on normality.	Outlier influence: μ and σ are sensitive to outliers; a single extreme observation can distort z-score calculations significantly.
Ubiquitous software support: Every statistical package provides normal PDF, CDF, and quantile functions, facilitating rapid computation.	Discrete data misfit: Count data and binary outcomes require binomial, Poisson, or other discrete distributions; applying the normal model directly introduces approximation error.

Strengths and limitations of the normal distribution and z-score approach in biostatistical practice.

⚠️

WHEN TO QUESTION NORMALITY

Before computing z-scores, always examine your data graphically (histogram, Q–Q plot) and numerically (skewness, kurtosis). In biostatistics, many variables—income data, survival times, antibody titers—are right-skewed. Applying a log or Box-Cox transformation can often restore approximate normality, allowing you to use z-based methods validly. Think of it like calibrating an instrument: the z-score formula is the instrument, but it gives accurate readings only when the normality 'calibration assumption' holds.

SECTION 8

Connection to Advanced Theory

The normal distribution and z-score framework serve as the gateway to more sophisticated biostatistical methods. Understanding where the standard normal fits within the broader landscape of inferential statistics clarifies when to move beyond simple z-based analyses.

Feature	z-Score / Normal Approach	Advanced Extension
Known σ	z-test: z = (x̄ − μ₀) / (σ / √n)	t-test (Student's t): used when σ is unknown and estimated by s; heavier tails for small n
Single variable	Univariate normal N(μ, σ²)	Multivariate normal N(μ, Σ) for correlated biomarker panels
Sampling distribution	Central Limit Theorem: x̄ ~ N(μ, σ²/n)	Bootstrap and permutation methods when CLT conditions are questionable
Probability model	Continuous, symmetric bell curve	Generalized linear models (GLMs) for non-normal outcomes: logistic, Poisson, gamma

Progression from z-score fundamentals to advanced biostatistical methods.

In clinical research, the z-score also reappears in standardized effect sizes (Cohen's d), growth-chart percentiles in pediatrics (CDC and WHO z-scores for height-for-age), and genomic studies where thousands of test statistics are compared to the standard normal to control false discovery rates. Mastering the z-score transformation is therefore not merely an introductory exercise—it is a skill that scales directly into advanced epidemiological and genomic analyses.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A researcher reports that a patient's serum creatinine level corresponds to a z-score of −1.50 relative to the reference population. Explain what this z-score means in plain language and state whether the patient's level is above or below the population mean.

Show answer

A z-score of −1.50 means the patient's serum creatinine is 1.50 standard deviations below the population mean. The negative sign explicitly indicates that the value is lower than average. In terms of the empirical rule, a z of −1.50 falls between −1σ and −2σ, placing this patient's creatinine in roughly the lower 6.7% of the reference distribution (Φ(−1.50) = 0.0668). Depending on clinical context, a low creatinine may reflect low muscle mass or other physiological factors.

Question 2: BASIC CALCULATION

Total cholesterol in a certain adult population follows N(200, 35²) mg/dL. What is the probability that a randomly selected individual has a cholesterol level below 165 mg/dL?

Show answer

First compute the z-score: z = (165 − 200) / 35 = −35 / 35 = −1.00. Using the z-table, Φ(1.00) = 0.8413. By symmetry, Φ(−1.00) = 1 − 0.8413 = 0.1587. Therefore, P(X < 165) ≈ 0.1587, or about 15.87%. Approximately one in six adults in this population has a total cholesterol below 165 mg/dL.

Question 3: INTERMEDIATE

Diastolic blood pressure (DBP) in a healthy adult cohort is normally distributed with μ = 80 mmHg and σ = 8 mmHg. Find the probability that a randomly chosen individual has a DBP between 72 mmHg and 92 mmHg.

Show answer

Compute both z-scores: z₁ = (72 − 80) / 8 = −1.00 and z₂ = (92 − 80) / 8 = 1.50. From the z-table, Φ(1.50) = 0.9332 and Φ(−1.00) = 1 − 0.8413 = 0.1587. The probability of falling in the interval is P(72 < X < 92) = Φ(1.50) − Φ(−1.00) = 0.9332 − 0.1587 = 0.7745. Thus, about 77.45% of individuals in this cohort have diastolic blood pressure between 72 and 92 mmHg. This interval covers a range from 1σ below to 1.5σ above the mean.

Question 4: APPLIED

A pharmaceutical company manufactures aspirin tablets labeled as 325 mg. Quality control data show that the actual aspirin content per tablet is normally distributed with μ = 327 mg and σ = 4 mg. A tablet is considered out of specification if it contains less than 318 mg or more than 336 mg. What proportion of tablets is expected to fail quality control?

Show answer

We need P(X < 318) + P(X > 336). For the lower bound: z₁ = (318 − 327) / 4 = −2.25, and Φ(−2.25) = 1 − Φ(2.25) = 1 − 0.9878 = 0.0122. For the upper bound: z₂ = (336 − 327) / 4 = 2.25, and P(X > 336) = 1 − Φ(2.25) = 1 − 0.9878 = 0.0122. Total out-of-spec proportion = 0.0122 + 0.0122 = 0.0244, or about 2.44%. This symmetric result occurs because the specification limits are equidistant (±9 mg) from the process mean. In a production run of 10,000 tablets, roughly 244 would be expected to fail QC.

Question 5: CRITICAL THINKING

In a genome-wide association study (GWAS), a researcher computes a test statistic of z = 5.20 for a candidate SNP. Using the standard normal distribution, compute the two-tailed p-value. Then discuss why GWAS studies typically use a significance threshold of p < 5 × 10⁻⁸ rather than the conventional p < 0.05, and comment on what this implies about how extreme a z-score must be to achieve genome-wide significance.

Show answer

The two-tailed p-value is p = 2 × [1 − Φ(5.20)]. From extended z-tables or software, Φ(5.20) ≈ 0.99999990, so 1 − Φ(5.20) ≈ 1.0 × 10⁻⁷, giving p ≈ 2.0 × 10⁻⁷. This is a very small p-value, but it does not reach the genome-wide significance threshold of 5 × 10⁻⁸. GWAS studies test approximately 1 million independent SNPs simultaneously, so a Bonferroni-type correction of 0.05 / 10⁶ ≈ 5 × 10⁻⁸ is applied to control the family-wise error rate. To achieve this threshold, the z-score must exceed approximately ±5.45. This example illustrates that the meaning of 'extreme' depends on the multiplicity context: a z-score that would be overwhelmingly significant in a single-test scenario (z = 5.20) may not survive the stringent corrections required in high-dimensional genomics.

SUMMARY

Lesson Summary

Summary

The normal distribution is a continuous, symmetric, bell-shaped probability model fully defined by its mean (μ) and standard deviation (σ). The z-score transformation z = (x − μ) / σ converts any observation on a normal variable to the standard normal distribution N(0, 1), enabling universal probability lookups via the z-table or CDF Φ(z). The empirical rule (68–95–99.7) provides a quick mental framework for gauging how unusual a measurement is, while precise tail and interval probabilities are computed using Φ(z), 1 − Φ(z), and Φ(b) − Φ(a).

In biostatistics, z-scores are applied to clinical reference ranges, pharmaceutical quality control, epidemiological screening, and even genome-wide association studies. However, the normal model carries assumptions—symmetry, continuous support, and sensitivity to outliers—that must be verified before applying z-based inferences. When these assumptions hold, the normal distribution and z-score framework provide an elegant, powerful, and computationally simple engine for probabilistic reasoning that generalizes naturally into t-tests, confidence intervals, and generalized linear models encountered later in the biostatistics curriculum.