Statistics › Section H · Confidence intervals › Topic H.4

Interval for a proportion

More than half of the country approves, says the poll - and the honest question is how much wobble that number carries. The proportion confidence interval answers it, and every news-story margin of error is its half-width.

The pollster's interval

A headline figure such as "55% support" comes from a few hundred or a few thousand people, not from everyone. For a population proportion \(p\), the confidence interval turns the sample proportion \(\hat{p}\) into a range of plausible values, with a stated confidence in the method. It is one of the most quoted constructions in statistics: every news-story margin of error is this interval's half-width, and every "within 3 points" claim is this formula evaluated in advance.

The formula

For a sample proportion, the interval is

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

- the estimate plus and minus a critical value times its standard error, with the unknown \(p\) inside the SE replaced by its estimate \(\hat{p}\). The critical value is the ordinary normal \(z^*\), not \(t\): proportions rest on the binomial-to-normal approximation from the CLT chapter, which is exactly why the conditions matter. For 95% confidence, \(z^* = 1.96\).

Conditions: counts must be large enough

Three conditions license the formula. A random sample. Independence - either sampling with replacement or a sample under 10% of the population, so draws do not influence each other. And the success-failure condition from the binomial-approximation topic: at least 10 successes and 10 failures in the sample, so the bell approximation holds. When counts run small, an adjusted method such as the plus-four interval - adding two successes and two failures before computing - keeps the coverage honest. With only 8 successes in 40 trials, plus-four uses \(\tilde{p} = 10/44 \approx 0.227\) and \(n = 44\).

Worked example: the voter poll

In a random sample of 400 voters, 220 support a measure: \(\hat{p} = 0.55\). 1. Conditions: 220 successes and 180 failures clear 10 easily, and the sample is random. 2. Standard error: \(\sqrt{\frac{0.55 \times 0.45}{400}} \approx 0.0249\). 3. Margin: \(1.96 \times 0.0249 \approx 0.049\). 4. Interval: \(0.55 \pm 0.049 = (0.501, 0.599)\). 5. Interpretation: we are 95% confident that between 50.1% and 59.9% of all voters support the measure - a majority, but only just, since the lower bound barely clears 0.5.

Worked example: clinic satisfaction

A clinic surveys \(n = 250\) patients at random; 190 report being satisfied: \(\hat{p} = 0.76\). 1. Conditions: 190 successes and 60 failures both clear 10. 2. Standard error: \(\sqrt{\frac{0.76 \times 0.24}{250}} \approx 0.027\). 3. 95% margin: \(1.96 \times 0.027 \approx 0.053\). 4. Interval: \((0.707, 0.813)\). 5. Interpretation: we are 95% confident that between about 71% and 81% of all the clinic's patients are satisfied. Check: the margin is about 5 points, slightly wider than the poll's, even though \(0.76 \times 0.24 = 0.18\) is smaller than \(0.55 \times 0.45 \approx 0.25\): the smaller sample costs more precision than the lopsided \(\hat{p}\) saves.

What 95% confident means

The same reading as for the mean intervals, repeated here because proportions are where it gets misquoted most. \(p\) is fixed; the interval varies. If many random samples were drawn and an interval built from each, about 95% of those intervals would capture the true proportion - the method's long-run success rate. It is not a 95% probability that this particular interval holds \(p\): any one interval either does or does not, and no re-examination of the study can say which.

The margin covers only sampling error

The margin of error prices random sampling variability - nothing else. It does not cover undercoverage, nonresponse, or misleading questions, which are design failures, not chance failures. Recall the Literary Digest poll from the sampling topic: 2.4 million responses, a margin of error near zero by formula, and still the wrong winner predicted - because the sample was biased. A tiny margin on a biased sample is precision about the wrong number: the interval is only as trustworthy as the design that produced it.

Confidence, width, and the population's size

The levers are familiar from the mean intervals. Higher confidence widens: \(z^*\) runs 1.645, 1.96, 2.576 for 90%, 95%, 99%. Bigger samples narrow: the standard error shrinks with \(\sqrt{n}\). And one non-lever: the population's size barely matters. A random sample of 1,000 measures a city of 100,000 and a nation of 300 million almost equally well - which is why the next topic can plan a survey's size without knowing the population at all.

Common mistakes to avoid

First, saying "there is a 95% probability that \(p\) is in this interval" - the parameter is fixed; the method is what succeeds 95% of the time. Second, believing a larger population needs a proportionally larger sample. Third, treating the margin as covering bias - it covers chance alone. Fourth, running the formula with fewer than 10 successes or failures. Fifth, borrowing one study's \(\hat{p}\) to plan an unrelated study's size without caution. Sixth, reading a boundary-touching interval as decisive: 0.501 clears a majority, but only just.

Key takeaways

The proportion interval is \(\hat{p} \pm z^*\sqrt{\hat{p}(1-\hat{p})/n}\) - estimate plus critical value times standard error - under three conditions: random, independent (under 10% of the population), and at least 10 successes and 10 failures, with plus-four as the small-count adjustment. The voters: \(0.55 \pm 0.049 = (0.501, 0.599)\) - a majority, barely. The clinic: \((0.707, 0.813)\). Confidence is the method's long-run capture rate; the margin covers sampling error only, never bias; and population size barely matters once large.

Flashcards

Flashcard

p̂ ± z*√(p̂(1−p̂)/n)

The confidence interval for a proportion

Estimate ± critical value × SE.

Test your understanding

Question 1 of 5

A poll of 400 voters finds p̂ = 0.55. What is the 95% margin of error?

Question 2 of 5

The proportion interval's success-failure condition requires at least:

Question 3 of 5

A poll's margin of error covers:

Question 4 of 5

A 95% interval for the support proportion is (0.501, 0.599). Which statement is correct?

Question 5 of 5

To keep the same margin of error, does a larger population need a larger sample?

← Previous topicNext topic →