Inference for Proportions

Inference for proportions uses sample data to draw conclusions about an unknown population proportion

p

. Key result: the sampling distribution of

\overset{p}{^}

is approximately Normal with mean

p

and standard deviation

p (1 - p) / n

, which lets us build confidence intervals and run hypothesis tests for one or two proportions. Applies to categorical (yes/no, success/failure) data — polls, medical trials, quality control. Developed through the work of Laplace, de Moivre, and the early 20th-century statisticians who formalized hypothesis testing.

Inference for proportions is the branch of statistical inference concerned with estimating, and testing claims about, the fraction $p$ of a population that has some characteristic (votes yes, recovers from a disease, passes inspection). Because we rarely census an entire population, we take a random sample, compute the sample proportion $\overset{p}{^}$ , and use the known behavior of $\overset{p}{^}$ across repeated samples — its sampling distribution — to quantify uncertainty. This matters because virtually every poll, clinical trial, and A/B test reduces to a question about one or two proportions. The surprising fact: even though any single sample's $\overset{p}{^}$ is "wrong," the pattern of errors across all possible samples is so predictable (an approximately Normal curve) that we can attach precise probabilities to our conclusions.

The Core Insight

A proportion is just a mean of zeros and ones. Label each "success" as 1 and each "failure" as 0; then $\overset{p}{^}$ is the average of these indicators. This means the Central Limit Theorem applies directly: for large enough $n$ , $\overset{p}{^}$ is approximately Normally distributed. Everything else — confidence intervals, $z$ -tests — flows from knowing the center ( $p$ ) and spread ( $p (1 - p) / n$ ) of that Normal curve.

Formal Statement

Population and Sample Proportion

The population proportion $p$ is the true fraction of all individuals in a population possessing a given attribute. The sample proportion is

\overset{p}{^} = \frac{X}{n}

where $X$ is the number of successes observed in a random sample of size $n$ , and $X \sim Binomial (n, p)$ .

⟳Count successes, divide by sample size; each trial is an independent Bernoulli(

p

) outcome.

Sampling Distribution of $\hat{p}$

If a Simple Random Sample of size $n$ is drawn from a population with proportion $p$ , then $\overset{p}{^}$ has

μ_{\overset{p}{^}} = p

⟳

μ_{\overset{p}{^}} = E [X / n] = (1/ n) E [X] = (1/ n) (n p) = p

using

E [Binomial] = n p

σ_{\overset{p}{^}} = \frac{p ( 1 - p )}{n}

⟳

Var (X / n) = (1/ n^{2}) Var (X) = (1/ n^{2}) n p (1 - p)

; take the square root.

Moreover, if $n p \geq 10$ and $n (1 - p) \geq 10$ (the Large Counts Condition), the distribution of $\overset{p}{^}$ is approximately Normal:

\overset{p}{^} \approx N (p, \frac{p ( 1 - p )}{n})

⟳Apply the Central Limit Theorem to the mean of

n

Bernoulli indicator variables.

Standard Error

Because $p$ is unknown in practice, we estimate the spread using the observed $\overset{p}{^}$ . The standard error of the sample proportion is

S E_{\overset{p}{^}} = \frac{p ^ ( 1 - p ^ )}{n}

⟳Substitute the point estimate

\overset{p}{^}

for

p

in equation [3].

Confidence Interval for One Proportion

A level- $C$ confidence interval for $p$ is

\overset{p}{^} \pm z^{*} \frac{p ^ ( 1 - p ^ )}{n}

where $z^{*}$ is the critical value capturing the central $C$ proportion of the standard Normal curve (e.g. $z^{*} = 1.96$ for 95%).

⟳Invert the statement

P (- z^{*} \leq (\overset{p}{^} - p) / S E \leq z^{*}) = C

to bracket

p

One-Sample $z$-Test for a Proportion

To test $H_{0} : p = p_{0}$ , compute the test statistic

z = \frac{p ^ - p _{0}}{\frac{p _{0} ( 1 - p _{0} )}{n}}

⟳Standardize

\overset{p}{^}

under

H_{0}

; crucially the SE uses the

null value $p_{0}$ , not $\overset{p}{^}$ .

then find the $p$ -value from the standard Normal distribution according to the alternative hypothesis.

Two-Sample $z$-Test and Interval

For comparing two independent proportions $p_{1}, p_{2}$ , the difference $\overset{p}{^}_{1} - \overset{p}{^}_{2}$ has standard error (for a confidence interval)

S E_{diff} = \frac{p ^ _{1} ( 1 - p ^ _{1} )}{n _{1}} + \frac{p ^ _{2} ( 1 - p ^ _{2} )}{n _{2}}

⟳Variances of independent quantities add; take the square root of the sum.

For a test of $H_{0} : p_{1} = p_{2}$ , use the pooled proportion $\overset{p}{^}_{c} = \frac{X _{1} + X _{2}}{n _{1} + n _{2}}$ in the standard error:

z = \frac{p ^ _{1} - p ^ _{2}}{p ^ _{c} ( 1 - p ^ _{c} ) ( \frac{1}{n _{1}} + \frac{1}{n _{2}} )}

⟳Under

H_{0}

both samples share a common

p

, best estimated by combining all successes over all trials.

Why It Holds / Derivation

From Bernoulli Trials to the Normal Curve

Step 1. Define indicator variables

Y_{i} = 1

if individual

i

is a success,

0

otherwise. Each

Y_{i}

is Bernoulli(

p

) with

E [Y_{i}] = p

and

Var (Y_{i}) = p (1 - p)

. Step 2. The total successes

X = \sum_{i = 1}^{n} Y_{i}

, so

\overset{p}{^} = \frac{X}{n} = \frac{1}{n} \sum Y_{i} = \overset{ˉ}{Y}

. A proportion is literally a sample mean. Step 3. By linearity of expectation,

E [\overset{p}{^}] = \frac{1}{n} \sum E [Y_{i}] = \frac{1}{n} (n p) = p

. This gives equation [2]. Step 4. Because the trials are independent, variances add:

Var (X) = \sum Var (Y_{i}) = n p (1 - p)

. Then

Var (\overset{p}{^}) = Var (X / n) = \frac{1}{n ^{2}} n p (1 - p) = \frac{p ( 1 - p )}{n}

, giving equation [3]. Step 5. Since

\overset{p}{^}

is a mean of

n

i.i.d. variables, the Central Limit Theorem guarantees its distribution approaches Normal as

n

grows. The Large Counts condition (

n p \geq 10

n (1 - p) \geq 10

) ensures

n

is large enough relative to how skewed the underlying Bernoulli is. Step 6. Standardizing,

Z = \frac{p ^ - p}{p ( 1 - p ) / n}

is approximately standard Normal — the engine behind both the interval [6] and the test [7].

■

Pierre-Simon Laplace | 1749–1827

Laplace generalized de Moivre's Normal approximation to the binomial and applied it to real proportions — most famously estimating the proportion of male births in Paris and computing the probability that it truly exceeded the female rate. His Théorie analytique des probabilités (1812) effectively created the confidence-interval reasoning we now use for proportions, framing it as "inverse probability."

Building Intuition

Imagine a giant jar of red and white marbles where the true red fraction is $p$ . Each scoop of $n$ marbles gives a slightly different red fraction $\overset{p}{^}$ . If you imagine all possible scoops, their $\overset{p}{^}$ values pile up into a bell curve centered exactly at $p$ .

Three levers control the width of that bell:

Sample size $n$ : it sits under a square root, so to halve the margin of error you must quadruple $n$ . Diminishing returns are baked in.
The true proportion $p$ : the spread $p (1 - p)$ is largest at $p = 0.5$ (maximum uncertainty) and shrinks toward the extremes. A near-unanimous population is easy to estimate; a 50/50 split is the hardest.
Confidence level: demanding 99% certainty instead of 95% widens the net (larger $z^{*}$ ), trading precision for security.

Sampling distribution of $\hat{p}$: centered at the true $p$, with the 95% confidence band shaded.

<svg viewBox="0 0 520 320" xmlns="http://www.w3.org/2000/svg">
<rect width="520" height="320" fill="#0a0e17"/>
<!-- axes -->
<line x1="40" y1="270" x2="490" y2="270" stroke="#94a3b8" stroke-width="1.5"/>
<line x1="40" y1="270" x2="40" y2="30" stroke="#94a3b8" stroke-width="1.5"/>
<!-- bell curve centered at x=265 -->
<path d="M 60 270 C 150 270, 195 60, 265 60 C 335 60, 380 270, 470 270"
      fill="none" stroke="#d4a853" stroke-width="2.5"/>
<!-- center line -->
<line x1="265" y1="270" x2="265" y2="60" stroke="#94a3b8" stroke-width="1" stroke-dasharray="4 3"/>
<!-- 95% band markers -->
<line x1="180" y1="270" x2="180" y2="150" stroke="#94a3b8" stroke-width="1" stroke-dasharray="3 3"/>
<line x1="350" y1="270" x2="350" y2="150" stroke="#94a3b8" stroke-width="1" stroke-dasharray="3 3"/>
<!-- labels -->
<text x="258" y="288" fill="#e2e8f0" font-size="13" font-family="serif">p</text>
<text x="150" y="288" fill="#e2e8f0" font-size="12" font-family="serif">p − 1.96·SE</text>
<text x="320" y="288" fill="#e2e8f0" font-size="12" font-family="serif">p + 1.96·SE</text>
<text x="200" y="170" fill="#e2e8f0" font-size="12" font-family="serif">95% of all p&#770;</text>
<text x="220" y="45" fill="#d4a853" font-size="13" font-family="serif">distribution of p&#770;</text>
<text x="495" y="285" fill="#94a3b8" font-size="12" font-family="serif">p&#770;</text>
</svg>

Worked Examples

Confidence Interval — Election Poll

Given: A random sample of

n = 1000

likely voters;

\overset{p}{^} = 0.52

support candidate A. Build a 95% confidence interval for the true support

p

. Step 1 — Check conditions. Random sample ✓. Large counts:

n \overset{p}{^} = 520 \geq 10

and

n (1 - \overset{p}{^}) = 480 \geq 10

✓. (10% condition: population of voters ≫ 10,000 ✓.) Step 2 — Standard error (eq. [5]):

S E = \frac{0.52 \times 0.48}{1000} = 0.0002496 \approx 0.0158.

Step 3 — Margin of error,

z^{*} = 1.96

M E = 1.96 \times 0.0158 \approx 0.0310.

Step 4 — Interval (eq. [6]):

0.52 \pm 0.031 = (0.489, 0.551) .

Answer:

p \in (0.489, 0.551) at 95% confidence

Because the interval includes 0.50, we cannot conclude candidate A has majority support.

Hypothesis Test — Defective Rate

Given: A factory claims its defect rate is

p_{0} = 0.05

. An inspector finds

X = 38

defects in

n = 500

items. Test, at

α = 0.05

, whether the true rate exceeds 5%. Step 1 — Hypotheses.

H_{0} : p = 0.05

vs.

H_{a} : p > 0.05

. Sample proportion

\overset{p}{^} = 38/500 = 0.076

. Step 2 — Conditions.

n p_{0} = 500 (0.05) = 25 \geq 10

n (1 - p_{0}) = 475 \geq 10

✓. Step 3 — Test statistic (eq. [7], uses $p_{0}$ ):

z = \frac{0.076 - 0.05}{\frac{0.05 \times 0.95}{500}} = \frac{0.026}{0.000095} = \frac{0.026}{0.009747} \approx 2.67.

Step 4 — $p$ -value (right tail):

P (Z > 2.67) \approx 0.0038

. Step 5 — Decision.

0.0038 < 0.05

, so reject

H_{0}

. Answer:

z \approx 2.67, p -value \approx 0.004 \Rightarrow reject H_{0}

There is strong evidence the true defect rate exceeds 5%.

Common Misconceptions

The SE Switch: $p_0$ vs. $\hat{p}$

A frequent error is using the same standard error formula for both intervals and tests. They differ on purpose:

Confidence intervals use $S E = \overset{p}{^} (1 - \overset{p}{^}) / n$ (eq. [5]) — we have no hypothesized value, so we estimate spread from the data.
Hypothesis tests use $p_{0} (1 - p_{0}) / n$ (eq. [7]) — we assume $H_{0}$ is true, so the spread is computed under $p_{0}$ .

Mixing these gives a wrong test statistic. Likewise, two-sample tests pool (eq. [9]) but two-sample intervals do not (eq. [8]).

What "95% Confident" Does NOT Mean

A 95% confidence interval does not mean "there is a 95% probability that $p$ lies in this interval." Once computed, the interval either contains the fixed (but unknown) $p$ or it does not. The 95% refers to the long-run capture rate of the procedure: if we repeated the sampling endlessly, about 95% of the resulting intervals would trap the true $p$ .

When the Normal Approximation Fails

Suppose a rare disease has true prevalence $p = 0.01$ and you sample $n = 200$ people. Then $n p = 2 < 10$ — the Large Counts condition fails. The sampling distribution of $\overset{p}{^}$ is strongly right-skewed (you'll often see 0, 1, or 2 cases, never negative ones), so a Normal-based interval like $\overset{p}{^} \pm 1.96 S E$ can dip below 0 — nonsense for a proportion. Here one must use the exact Binomial or a corrected method (e.g. the Wilson or Agresti–Coull interval).

Connections

Inference for proportions is one face of a unified framework:

It is a special case of Inference for Means — a proportion is the mean of 0/1 data, with the twist that the spread $p (1 - p)$ is determined by the center, so we use $z$ rather than $t$ .
It rests entirely on the Central Limit Theorem and the Normal Distribution.
The underlying success count obeys the Binomial Distribution, the exact distribution we approximate.
The two-sample test connects to the Chi-Square Test: a $2 \times 2$ table's chi-square statistic equals the square of the two-proportion $z$ -statistic.
The logic of $p$ -values and confidence intervals belongs to the broader theory of Hypothesis Testing and Confidence Intervals.

Real-World Applications

Political polling — Gallup, Pew, and election forecasters report candidate support as $\overset{p}{^}$ with a "margin of error" that is the $\pm z^{*} S E$ term of equation [6].
Clinical trials — comparing recovery proportions between a drug group and placebo group is a textbook two-proportion test (eq. [9]); regulators like the FDA require such evidence.
A/B testing in tech — companies (Amazon, Netflix, Google) test whether a new web layout raises click-through proportion versus the old one.
Quality control — manufacturers monitor the proportion of defective units against a target $p_{0}$ (Example 2 above).
Epidemiology — estimating disease prevalence or vaccine efficacy (a difference of infection proportions) from survey or trial data.
Market research — gauging the fraction of consumers who prefer a product or recognize a brand.
Public-opinion and census auditing — checking whether the proportion answering "yes" to a referendum question differs from 50%.

Historical Development

1713 — Jacob Bernoulli's Ars Conjectandi states the Law of Large Numbers, proving $\overset{p}{^}$ converges to $p$ — the philosophical foundation of using samples to estimate proportions.
1733 — Abraham de Moivre derives the Normal approximation to the binomial, the first version of equation [4].
1812 — Pierre-Simon Laplace generalizes de Moivre's result and applies "inverse probability" to real proportions (e.g. sex ratios at birth).
1908 — William Sealy Gosset ("Student") publishes the $t$ -distribution; though aimed at means, it sharpens thinking about small-sample inference relevant to proportions.
1920s — Ronald Fisher formalizes significance testing and the $p$ -value, giving the one-sample proportion test its modern form.
1927 — Edwin B. Wilson proposes the Wilson score interval, a more accurate alternative to the simple $\overset{p}{^} \pm z^{*} S E$ interval, especially for small $n$ or extreme $p$ .
1933 — Jerzy Neyman and Egon Pearson introduce the confidence interval concept and the hypothesis-testing decision framework (Type I / II errors).
1934 — Clopper and Pearson publish the exact binomial interval, the gold-standard guaranteed-coverage method.
1998 — Agresti and Coull popularize the "add-2-successes-and-2-failures" adjustment, dramatically improving the textbook interval's accuracy.

Subtleties & Limits

The "obvious" Wald interval $\overset{p}{^} \pm z^{*} \overset{p}{^} (1 - \overset{p}{^}) / n$ has a notoriously erratic actual coverage: for a nominal 95% interval, real coverage can dip well below 90% even for moderately large $n$ , oscillating chaotically with $p$ — a phenomenon documented by Brown, Cai, and DasGupta (2001). There is no single universally "best" interval: the exact Clopper–Pearson is conservative (over-covers), Wald is unreliable, and Wilson/Agresti–Coull are good compromises. Choosing the right method for small samples, extreme proportions, or clustered/non-independent data (where the variance formula [3] breaks down) remains an active practical concern, motivating bootstrap and Bayesian credible intervals as alternatives.