AP STATISTICS • SAMPLING DISTRIBUTIONS

The Normal Distribution, Revisited

Understanding why the normal curve governs sampling distributions and underpins statistical inference.

Historical Context & Motivation

You first encountered the normal distribution as a model for individual measurements — heights, test scores, or measurement errors. Now we revisit it in a profoundly different role: as the shape that sampling distributions tend to follow when sample sizes grow large. This shift from describing raw data to describing the behavior of statistics like x̄ and p̂ is arguably the most important conceptual leap in introductory statistics, because it connects probability theory directly to inference. The story of how mathematicians discovered this remarkable convergence spans three centuries and multiple countries, and each milestone deepened our understanding of why averages are so much better behaved than the individual observations they summarize.

1733
De Moivre's Approximation
Abraham de Moivre showed that the binomial distribution could be approximated by a smooth bell-shaped curve as the number of trials increased, establishing the first version of the normal curve as a limiting form.
1809
Gauss and the Error Curve
Carl Friedrich Gauss used the bell curve to model astronomical measurement errors, deriving the method of least squares and lending the distribution his name — the 'Gaussian' distribution.
1810
Laplace's Central Limit Theorem
Pierre-Simon Laplace proved that sums (and therefore means) of independent random variables converge to normality regardless of the underlying distribution, establishing the Central Limit Theorem in its earliest form.
1901
Pearson and Modern Statistics
Karl Pearson formalized hypothesis testing and goodness-of-fit procedures that rely on normal-theory sampling distributions, embedding the normal curve at the heart of applied statistics.
1908
Student's t-Distribution
William Sealy Gosset ('Student') showed how sampling distributions differ from the normal when sample sizes are small, refining inference by accounting for the additional variability of estimated standard deviations.

The central question this lesson addresses is: Why does the normal distribution reappear when we study sampling distributions, and under what conditions can we rely on it? Answering this question will let you use normal-based calculations to find probabilities for sample means, sample proportions, and differences of statistics — the computational engine behind confidence intervals and hypothesis tests on the AP Statistics exam.

Core Principles & Definitions

Before diving into formulas, it is essential to separate two ideas that students often conflate. A population distribution describes the values of individual observations — it can be skewed, bimodal, or any shape at all. A sampling distribution describes the possible values of a statistic (such as x̄ or p̂) computed from every possible sample of size n drawn from that population. The normal distribution's reappearance in this course is about the second concept: under broad conditions, the sampling distribution of a statistic is approximately normal even when the population is not.

1

The Central Limit Theorem (CLT)

For a random sample of size n from a population with mean μ and finite standard deviation σ, the sampling distribution of x̄ approaches N(μ, σ/√n) as n increases, regardless of the population shape.
2

Normal Condition for Proportions

The sampling distribution of p̂ is approximately normal when both np ≥ 10 and n(1 − p) ≥ 10, ensuring sufficient expected successes and failures for the bell curve to be a good model.
3

Standard Error vs. Standard Deviation

The standard deviation of a sampling distribution is called the standard error. For x̄ it equals σ/√n; for p̂ it equals √(p(1 − p)/n). As n grows, standard errors shrink, concentrating the sampling distribution.
4

The 10% Condition

When sampling without replacement, the standard-error formulas apply approximately only if the sample is less than 10% of the population, so that draws remain nearly independent.
KEY TAKEAWAY
Think of the CLT like tasting a well-stirred soup: no matter how varied the individual ingredients (the population), a large enough spoonful (the sample mean) always yields a consistent, predictable flavor profile (a normal distribution). The spoon doesn't need the soup to be uniform — stirring and averaging do the normalizing work. Similarly, summing many independent random contributions washes out individual quirks, producing the smooth bell curve that enables inference.

Visual Explanation

The diagram below illustrates how the Central Limit Theorem transforms a clearly non-normal population into a nearly normal sampling distribution as sample size increases. On the left you see a right-skewed population distribution. To its right are three sampling distributions of x̄ for n = 5, n = 15, and n = 30. Notice how the shape becomes progressively more symmetric and bell-shaped, and how the spread narrows, reflecting the decreasing standard error σ/√n.

As n increases from left to right, the sampling distribution of x̄ becomes narrower and more symmetric, converging to a normal distribution centered at the population mean μ.

Several features of the diagram merit careful attention. First, the center of every sampling distribution remains at μ, the population mean; x̄ is an unbiased estimator. Second, the spread shrinks dramatically — the standard error σ/√n decreases as n grows, so larger samples produce more precise estimates. Third, the shape transforms from skewed to approximately normal. For populations that are already symmetric, even n = 10 or 15 may suffice; for heavily skewed populations, n ≥ 30 is a commonly cited guideline on the AP exam, though the underlying principle is that larger n always helps.

Mathematical Framework

The normal distribution's probability density function and standardization formula remain the same as when you first studied them, but the parameters now describe sampling distributions rather than individual observations. Below are the key equations you will use repeatedly in inference problems.

SAMPLING DISTRIBUTION OF x̄
x̄ ~ N(μ, σ / √n)
μ = population mean; σ = population standard deviation; n = sample size. Valid when the population is normal, or when n is large enough for the CLT to apply.
Z-SCORE FOR A SAMPLE MEAN
z = (x̄ − μ) / (σ / √n)
This converts x̄ to the standard normal scale. Each unit of z represents one standard error away from the population mean.
SAMPLING DISTRIBUTION OF p̂
p̂ ~ N(p, √(p(1 − p) / n))
p = true population proportion; n = sample size. Approximately valid when np ≥ 10 and n(1 − p) ≥ 10 (the Large Counts condition).
Z-SCORE FOR A SAMPLE PROPORTION
z = (p̂ − p) / √(p(1 − p) / n)
Used for probability calculations and hypothesis tests involving a single proportion. When p is unknown, replace it with p̂ for confidence intervals or with p₀ for significance tests.
📋 Conditions Checklist for Normal Approximation
On the AP exam, you must verify conditions before using a normal model. For means: (1) Random — data come from a random sample or randomized experiment; (2) Normal/Large Sample — population is normal, or n ≥ 30; (3) Independence / 10% condition — n < 0.10 × N (population size). For proportions, replace condition (2) with the Large Counts condition: np ≥ 10 and n(1 − p) ≥ 10.

Normal Model Applications in Sampling

The normal model for sampling distributions does not merely appear in one isolated context — it recurs across nearly every inference procedure in AP Statistics. The diagram below maps the three most common scenarios where you apply normal probability calculations to sampling distributions: sample means, sample proportions, and differences of two statistics. Each branch shows the relevant parameter, standard error formula, and the condition that justifies the normal approximation.

A decision map showing three branches of normal approximation for sampling distributions of means, proportions, and differences, along with the three universal conditions that must always be verified.
Summary of normal-model parameters for common AP Statistics sampling distributions
StatisticMean (Center)Standard Error (Spread)Normality Condition
x̄ (one sample)μσ / √nPopulation normal or n ≥ 30
p̂ (one proportion)p√(p(1 − p) / n)np ≥ 10 and n(1 − p) ≥ 10
x̄₁ − x̄₂μ₁ − μ₂√(σ₁²/n₁ + σ₂²/n₂)Both populations normal or both n ≥ 30
p̂₁ − p̂₂p₁ − p₂√(p₁(1−p₁)/n₁ + p₂(1−p₂)/n₂)Large Counts met in each sample

Worked Example

A bottling plant fills soda cans with a mean volume of μ = 355 mL and a standard deviation of σ = 4.2 mL. The distribution of individual fill volumes is approximately normal. A quality inspector selects a random sample of n = 36 cans from a production run of 10,000. What is the probability that the sample mean fill volume is less than 354 mL?

Finding P(x̄ < 354)
1
Step 1 — Verify ConditionsRandom: The sample is selected randomly. Independence / 10%: n = 36 is less than 10% of N = 10,000, so observations are approximately independent. Normal: The population is stated to be approximately normal, so the sampling distribution of x̄ is normal for any sample size.
2
Step 2 — Identify Parameters of the Sampling DistributionMean of x̄: μ_x̄ = μ = 355 mL. Standard error of x̄: σ_x̄ = σ / √n = 4.2 / √36 = 4.2 / 6 = 0.7 mL.
x̄ ~ N(355, 0.7)
3
Step 3 — Compute the z-Scorez = (x̄ − μ) / (σ / √n) = (354 − 355) / 0.7 = −1 / 0.7 ≈ −1.43
z ≈ −1.43
4
Step 4 — Find the ProbabilityUsing the standard normal table or a calculator: P(Z < −1.43) ≈ 0.0764. On a TI-84, this can be computed as normalcdf(−1E99, 354, 355, 0.7) which returns approximately 0.0766.
P(x̄ < 354) ≈ 0.0764, or about 7.6%
5
Step 5 — Interpret in ContextIf the true mean fill volume is 355 mL, there is approximately a 7.6% chance that a random sample of 36 cans will have a sample mean below 354 mL. This relatively low probability suggests that observing x̄ < 354 would be somewhat unusual but not extremely rare if the machine is correctly calibrated.

Strengths & Limitations of the Normal Model

The normal approximation is an extraordinarily powerful tool, but it has well-defined boundaries. Understanding when it works well and when it breaks down is essential for earning full credit on free-response questions, where condition-checking is scored explicitly.

Strengths and limitations of using the normal model for sampling distributions
StrengthsLimitations
Provides a simple, closed-form probability model — z-tables or calculator commands replace complex exact calculations.Approximation degrades for small n when the population is strongly skewed or has heavy tails.
Works for means, proportions, and differences — the same framework applies across many inference procedures.For proportions near 0 or 1, Large Counts may fail even at moderate n, requiring exact binomial methods.
Robustness increases with sample size: even modest departures from normality are smoothed out by the CLT.Assumes independent observations; clustered or autocorrelated data violate this assumption and inflate Type I error rates.
Foundation for constructing confidence intervals and conducting hypothesis tests throughout the AP curriculum.When σ is unknown and n is small, the t-distribution (not the normal) should be used for means.
KEY TAKEAWAY
The normal model for sampling distributions is like a GPS navigation system: it works wonderfully on well-traveled roads (large samples, reasonable conditions), but if you take it off-road (tiny n, extreme skew, dependent data), it will give you misleading directions. Always verify that the road conditions — randomness, independence, and adequate sample size — are met before trusting the route it calculates.

Connection to Advanced Theory

The normal approximation you use in AP Statistics is the entry point to a much richer family of results. In more advanced courses, you will see that many of the procedures you learned with the z-distribution are special cases of broader frameworks. Understanding where the AP-level treatment sits relative to these extensions helps you appreciate both its power and its simplifications.

AP Statistics treatment vs. advanced extensions of the normal model
FeatureAP Statistics TreatmentAdvanced Extension
Distribution of x̄ (σ unknown)Use the t-distribution with n − 1 dfDerive t from the ratio of a normal and a chi-squared random variable; explore exact finite-sample distribution theory
Rate of CLT convergenceGuideline: n ≥ 30 for 'large enough'Berry–Esseen theorem quantifies error bound as C × E[|X|³] / (σ³√n), giving precise convergence rates
Non-independent dataRequire 10% condition; otherwise avoid normal modelCLT variants for dependent sequences (e.g., mixing conditions, Lindeberg–Feller CLT) extend normality to broader settings
Multivariate statisticsNot coveredMultivariate normal distribution governs joint sampling distributions of vectors of sample means

For the AP exam, what matters most is the ability to recognize when the normal model applies, compute z-scores using the correct standard error, and interpret probabilities in context. The deeper theoretical machinery — moment-generating functions, characteristic functions, convergence in distribution — supports the same conclusions you draw from the CLT, but at a level of rigor reserved for mathematical statistics courses. Your AP mastery of conditions, formulas, and interpretation provides an excellent foundation for that future study.

Practice Problems

1
A population has a strongly left-skewed distribution with mean μ = 50 and standard deviation σ = 12. Random samples of size n = 40 are drawn. Which of the following best describes the sampling distribution of x̄?
2
In a large city, 68% of households recycle. A random sample of 150 households is selected. What is the approximate probability that the sample proportion p̂ of households that recycle is greater than 0.72?
3
A machine fills cereal boxes with a mean weight of 368 g and a standard deviation of 6 g. The distribution of individual box weights is roughly symmetric. An inspector weighs a random sample of 16 boxes. Between what two values does the middle 95% of possible sample means fall?
PROBLEM 4APPLIED
A researcher studies the commute times of employees at a large company (N = 5,000). The population of commute times has a mean of μ = 28 minutes with a standard deviation of σ = 9.5 minutes and is moderately right-skewed. The researcher takes a simple random sample of n = 50 employees. (a) Describe the sampling distribution of x̄, including shape, center, and spread. Justify the shape. (b) Find the probability that the sample mean commute time exceeds 30 minutes. (c) If the researcher instead took samples of n = 10 from this population, would it still be appropriate to use a normal model for the sampling distribution of x̄? Explain.
PROBLEM 5CRITICAL THINKING
Two independent random samples are drawn: Sample A consists of n₁ = 64 observations from a population with μ₁ = 120 and σ₁ = 16, and Sample B consists of n₂ = 49 observations from a population with μ₂ = 112 and σ₂ = 14. Neither population is known to be normally distributed. (a) Describe the sampling distribution of x̄₁ − x̄₂, including its shape, mean, and standard error. Verify all necessary conditions. (b) Find the probability that x̄₁ − x̄₂ exceeds 14. (c) Suppose the sample sizes were reduced to n₁ = 9 and n₂ = 9 with the same population parameters. Explain how each aspect of the sampling distribution (shape, center, spread) would change, and discuss whether the normal model would still be valid.

Summary

The normal distribution plays a fundamentally different role in this unit than when you first encountered it: it now describes the behavior of sampling distributions rather than individual data values. The Central Limit Theorem guarantees that when sample sizes are large enough, both sample means x̄ and sample proportions p̂ have approximately normal sampling distributions, regardless of the population shape. The key parameters are the standard error — σ/√n for means, √(p(1 − p)/n) for proportions — which quantifies how much variability the statistic exhibits from sample to sample.

Before applying the normal model, always verify three conditions: the data must come from a random sample, observations must be approximately independent (10% condition), and the sample must be large enough for the specific statistic (n ≥ 30 for means from non-normal populations, or np ≥ 10 and n(1 − p) ≥ 10 for proportions). Once these conditions are met, z-score calculations using the appropriate standard error allow you to find probabilities, construct confidence intervals, and conduct hypothesis tests — forming the computational backbone of statistical inference.

Varsity Tutors • AP Statistics • The Normal Distribution, Revisited