CI for a Proportion — Confidence Interval for a Proportion...

SECTION 1

Historical Context & Motivation

Statistical estimation has always grappled with a fundamental tension: we observe a finite sample, yet we wish to make claims about an entire population. When the parameter of interest is a population proportion — the fraction of individuals possessing some attribute — the challenge is particularly vivid because the underlying data are binary (success or failure, yes or no). The development of confidence intervals for proportions draws on several centuries of probabilistic and statistical thinking, from early work on the binomial distribution to the modern theory of interval estimation that underpins nearly every survey, clinical trial, and quality-control process today.

1713

Bernoulli's Ars Conjectandi

Jacob Bernoulli published his foundational treatise on probability, formally defining Bernoulli trials and the binomial distribution — the mathematical backbone of proportion inference.

1812

Laplace's Normal Approximation

Pierre-Simon Laplace demonstrated that the binomial distribution could be approximated by the normal (Gaussian) curve when the number of trials is large, providing the theoretical justification for what would become the z-interval.

1934

Clopper–Pearson Exact Interval

Clopper and Pearson published the first exact confidence interval for a binomial proportion, highlighting the distinction between exact and approximate methods and motivating continued refinement.

1937

Neyman's Confidence Interval Framework

Jerzy Neyman formalized the concept of a confidence interval as a random interval that captures the true parameter with a specified long-run frequency, establishing the interpretation still used in introductory statistics courses.

1998

Agresti–Coull Adjusted Interval

Agresti and Coull proposed a simple adjustment to the Wald (standard z-) interval that dramatically improves coverage probability for small samples, sparking a renewed debate about which proportion CI to teach.

Against this backdrop, a central question emerges: given a sample of size n in which we observe x successes, how do we construct an interval that, with a stated level of confidence, brackets the true population proportion p? The z-interval for a proportion provides the most widely taught answer, leveraging the Central Limit Theorem and the standard normal distribution to build a tractable, interpretable estimate.

SECTION 2

Core Principles & Definitions

Before diving into the mechanics of the z-interval, it is essential to ground ourselves in the foundational ideas that make the procedure valid. Each card below captures a principle that, taken together, forms the logical chain from raw binary data to a usable confidence interval.

Sample Proportion (p̂)

The sample proportion p̂ = x / n is the point estimator for the unknown population proportion p. It is unbiased: E(p̂) = p.

Sampling Distribution & CLT

By the Central Limit Theorem, for sufficiently large n the distribution of p̂ is approximately normal with mean p and standard deviation √(p(1 − p) / n).

Standard Error (SE)

Because p is unknown, we estimate the standard deviation of p̂ with the standard error SE = √(p̂(1 − p̂) / n), substituting the sample proportion for the parameter.

Critical Value (z*)

The critical value z* is the z-score that captures the central C% of the standard normal distribution. For 95% confidence, z* = 1.960.

Conditions for Validity

The interval requires (1) a random or representative sample, (2) independence of observations, and (3) np̂ ≥ 10 and n(1 − p̂) ≥ 10 to ensure the normal approximation is adequate.

✦ KEY TAKEAWAY

Think of the confidence interval as a net cast around a fish. The fish (the true proportion p) is fixed but hidden; the net (the interval) is random because it depends on the particular sample you caught. A 95% confidence level means that if you cast the net the same way over and over — each time drawing a new sample — about 95% of those nets would enclose the fish. The net does not move the fish, and the fish is not '95% likely' to be inside any single net; rather, the procedure succeeds 95% of the time in the long run.

SECTION 3

Visual Explanation — The Sampling Distribution and CI Construction

The bell-shaped curve represents the approximate sampling distribution of p̂. The central 95% of the area (shaded in cyan) is captured between p̂ − z*·SE and p̂ + z*·SE, forming the confidence interval. The red tail regions each contain 2.5% of the area (α/2).

The diagram illustrates how the z-interval is constructed geometrically. The sampling distribution of p̂, approximately normal by the Central Limit Theorem, is centered at the observed sample proportion (shown at 0.60 as an example). Moving z* standard errors in each direction from p̂ carves out the central region of the distribution whose area equals the desired confidence level. For a 95% interval, z* = 1.960, and the two tails each contain α/2 = 2.5% of the probability mass. The resulting interval — the green segment — is the set of plausible values for the population proportion p. Note that wider intervals (larger z* or larger SE) correspond to higher confidence or greater uncertainty, while narrower intervals arise from larger samples or proportions closer to 0 or 1.

SECTION 4

Mathematical Framework

The mathematical derivation of the z-interval for a proportion follows directly from the normal approximation to the binomial distribution. We begin with the sampling distribution result, formalize the standard error, and then invert a probability statement to obtain the interval.

SAMPLING DISTRIBUTION

p̂ ~ N( p, √[ p(1 − p) / n ] ) for large n

p̂ = sample proportion (x/n), p = population proportion, n = sample size. The notation ~ N(μ, σ) denotes an approximate normal distribution with mean μ and standard deviation σ.

STANDARD ERROR

SE(p̂) = √[ p̂(1 − p̂) / n ]

Since p is unknown, we replace it with p̂ to obtain the estimated standard error. This plug-in approach is valid when the sample size conditions (np̂ ≥ 10, n(1 − p̂) ≥ 10) are met.

CONFIDENCE INTERVAL FORMULA

p̂ ± z* × √[ p̂(1 − p̂) / n ]

Equivalently: ( p̂ − z*·SE , p̂ + z*·SE ). The critical value z* depends on the confidence level C: for C = 90%, z* = 1.645; for C = 95%, z* = 1.960; for C = 99%, z* = 2.576.

MARGIN OF ERROR

E = z* × √[ p̂(1 − p̂) / n ]

The margin of error E is the half-width of the interval. It determines the precision of the estimate; halving E requires quadrupling the sample size n.

The derivation proceeds by standardizing: because (p̂ − p) / √[p(1 − p)/n] ≈ Z ~ N(0,1), the probability statement P(−z* ≤ Z ≤ z*) = C can be algebraically rearranged to isolate p inside an interval centered at p̂. The substitution of p̂ for p in the standard error is what makes this the Wald interval — the most commonly presented form — although this substitution introduces slight inaccuracies at extreme proportions or small samples, a point we revisit in Section 7.

SECTION 5

Conditions for Validity & Common Critical Values

The z-interval is an approximate method, so its validity rests on several conditions. Violating these conditions can produce intervals with actual coverage probabilities substantially different from the nominal level. This section consolidates the requirements and provides a quick-reference table for critical values.

A visual checklist summarizing the three conditions — randomness, independence, and the success–failure condition — that must all hold before constructing a z-interval for a proportion.

Common critical values from the standard normal distribution

Confidence Level (C)	α = 1 − C	z* (Critical Value)
90%	0.10	1.645
95%	0.05	1.960
98%	0.02	2.326
99%	0.01	2.576

Each critical value z* is the positive z-score such that the area in the two tails of the standard normal distribution equals α. Equivalently, z* is the value where P(−z* ≤ Z ≤ z*) = C. In practice, these values are found using a z-table, software, or a calculator's inverse-normal function. Note how increasing the confidence level from 95% to 99% raises z* from 1.960 to 2.576, widening the interval and reducing precision — a fundamental trade-off between confidence and precision.

SECTION 6

Worked Example

A university survey office randomly selects 500 undergraduate students and asks whether they have used the campus mental-health counseling center in the past year. Of the 500 respondents, 145 answer "yes." Construct a 95% confidence interval for the true proportion of undergraduates who have used the counseling center.

95% CI for Campus Counseling Usage

Step 1 — Identify the given valuesWe have n = 500 students surveyed and x = 145 who answered "yes." The desired confidence level is C = 95%, so α = 0.05 and z* = 1.960.

n = 500, x = 145, z* = 1.960

Step 2 — Compute the sample proportionp̂ = x / n = 145 / 500 = 0.290. This is our point estimate: about 29% of the sample used the counseling center.

p̂ = 0.290

Step 3 — Verify conditionsRandomness: the sample was randomly selected. Independence: 500 is almost certainly less than 10% of the university's undergraduate population. Success–failure: np̂ = 500 × 0.290 = 145 ≥ 10 and n(1 − p̂) = 500 × 0.710 = 355 ≥ 10. All conditions are satisfied.

All three conditions met ✓

Step 4 — Calculate the standard errorSE = √[p̂(1 − p̂) / n] = √[0.290 × 0.710 / 500] = √[0.2059 / 500] = √[0.0004118] ≈ 0.02029.

SE ≈ 0.0203

Step 5 — Compute the margin of errorE = z* × SE = 1.960 × 0.02029 ≈ 0.03977.

E ≈ 0.0398

Step 6 — Construct the intervalLower bound = p̂ − E = 0.290 − 0.0398 = 0.2502. Upper bound = p̂ + E = 0.290 + 0.0398 = 0.3298.

95% CI: (0.250, 0.330)

Step 7 — Interpret in contextWe are 95% confident that the true proportion of undergraduates at this university who used the campus mental-health counseling center in the past year is between 25.0% and 33.0%. This means that if we were to repeat this sampling procedure many times, approximately 95% of the resulting intervals would contain the true proportion.

SECTION 7

Strengths, Limitations, and Alternative Methods

The Wald z-interval enjoys widespread use because of its simplicity and intuitive form, but decades of research have catalogued situations where its performance degrades. Understanding these trade-offs is essential for applied work and for choosing a method appropriate to the data at hand.

Strengths and limitations of the Wald z-interval for a proportion

Aspect	Strengths	Limitations
Simplicity	Easy to compute by hand; formula is symmetric around p̂, making interpretation straightforward.	Simplicity can be misleading: students may apply it without verifying conditions.
Coverage accuracy	Achieves near-nominal coverage when n is large and p is near 0.5.	Coverage can drop well below the nominal level when p is near 0 or 1, or when n is small (< 40). The interval may even produce bounds outside [0, 1].
Symmetry	Symmetric form p̂ ± E is easy to communicate and visualize.	Proportions near the boundary are inherently asymmetric; a symmetric interval is a poor approximation in such cases.
Alternatives	Serves as the foundation for more refined methods (Wilson, Agresti–Coull).	The Wilson and Agresti–Coull intervals have superior coverage properties and are recommended by many statisticians for routine use.

⚠ WHEN TO LOOK BEYOND THE WALD INTERVAL

Think of the Wald interval as a standard ruler: it works perfectly on a flat table but bows and gives misleading measurements on a curved surface. When p̂ is close to 0 or 1, the 'surface' is effectively curved (the binomial distribution is highly skewed), and the flat ruler — the symmetric normal approximation — no longer fits. In such situations, the Wilson score interval or the Agresti–Coull adjusted interval provides a better-fitting 'flexible ruler' that respects the curvature.

SECTION 8

Connection to Advanced Methods

The z-interval for a proportion is a special case of a broader family of interval estimators. Understanding where it sits in this landscape helps clarify when to use it and when more advanced tools are warranted. The table below contrasts the Wald z-interval with several related methods encountered in intermediate and advanced statistics courses.

The Wald z-interval in context: comparisons with alternative proportion CI methods

Method	Key Idea	When Preferred
Wald (z) Interval	Uses p̂ in both the center and the SE; symmetric form p̂ ± z*·SE.	Large n with p not too extreme; introductory courses for conceptual clarity.
Wilson Score Interval	Inverts the z-test; center is shifted toward 0.5, producing an asymmetric interval.	Small to moderate n; p near 0 or 1; recommended by many modern textbooks.
Agresti–Coull ("Add 2")	Adds 2 pseudo-successes and 2 pseudo-failures before computing a Wald-type interval (p̃ = (x+2)/(n+4)).	Quick, practical improvement over Wald with minimal extra computation.
Clopper–Pearson (Exact)	Based directly on the binomial distribution; guaranteed ≥ C coverage.	When exact coverage is required (e.g., regulatory settings); tends to be conservative.
Bayesian Credible Interval	Combines a prior distribution for p with the observed data via Bayes' theorem to produce a posterior interval.	When prior information is available or a direct probability statement about p is desired.

As you progress to courses in mathematical statistics, biostatistics, or survey methodology, you will encounter these alternatives in greater depth. The Wilson score interval, in particular, has gained momentum as the recommended default in many modern references because it maintains near-nominal coverage across the entire parameter space. The conceptual foundation you have built with the z-interval — point estimate ± margin of error, conditions checking, and interpretation — transfers directly to these more sophisticated methods.

SECTION 9

Practice Problems

PROBLEM 1 — CONCEPTUAL

A pollster constructs a 95% confidence interval for the proportion of voters who favor a ballot measure and obtains (0.42, 0.50). She states: 'There is a 95% probability that the true proportion is between 0.42 and 0.50.' Is this interpretation correct? Explain why or why not, and provide the correct interpretation.

PROBLEM 2 — BASIC CALCULATION

In a random sample of 200 smartphone users, 68 report using a mobile-payment app at least once a week. Construct a 90% confidence interval for the population proportion of weekly mobile-payment users.

PROBLEM 3 — INTERMEDIATE

A researcher wants to estimate the proportion of adults in a city who support expanding public transit. She wants a 95% confidence interval with a margin of error no greater than 3 percentage points (0.03). If she has no preliminary estimate of p, what is the minimum sample size she should use?

PROBLEM 4 — APPLIED

A pharmaceutical company tests a new vaccine on a random sample of 1,200 volunteers. Of these, 1,044 develop sufficient antibody levels. Construct a 99% confidence interval for the true efficacy rate (proportion developing sufficient antibodies). Then comment on whether the Wald interval is appropriate here, given that p̂ is close to 1.

PROBLEM 5 — CRITICAL THINKING

Consider two independent random samples: Sample A has n₁ = 400 with p̂₁ = 0.52, and Sample B has n₂ = 400 with p̂₂ = 0.46. Each yields its own 95% CI for its respective population proportion. A student claims that because the two intervals overlap, we can conclude the two population proportions are equal. Critique this reasoning. Under what conditions might this conclusion be wrong, and what procedure would be more appropriate for comparing two proportions?

SUMMARY

Summary — Confidence Interval for a Proportion

The z-interval for a proportion provides a method for estimating an unknown population proportion p using the formula p̂ ± z* × √[p̂(1 − p̂)/n]. The procedure hinges on three conditions: random sampling, independence of observations (verified via the 10% condition), and the success–failure condition (np̂ ≥ 10 and n(1 − p̂) ≥ 10). The critical value z* is determined by the desired confidence level — common values are 1.645 (90%), 1.960 (95%), and 2.576 (99%).

Proper interpretation is essential: a 95% confidence interval does not mean there is a 95% probability that p falls in the interval; rather, the procedure captures p 95% of the time across many repeated samples. The margin of error E = z*·SE quantifies precision; halving E requires quadrupling n. While the Wald z-interval is the standard introductory tool, more advanced methods such as the Wilson score interval and the Agresti–Coull interval offer improved coverage when p is near 0 or 1, or when n is small.