Selecting Procedures for Proportions

When analyzing data involving proportions — such as the fraction of voters who support a candidate, the percentage of patients who recover from a treatment, or the rate of defective items in a manufacturing process — statisticians must decide which inferential procedure is most appropriate. The two main procedures are confidence intervals, which estimate the true population proportion within a range, and hypothesis tests, which evaluate claims about the population proportion. Selecting the correct procedure depends on the research question, the structure of the data, and whether the necessary conditions are satisfied. A proportion p is a value between 0 and 1 that represents the relative frequency of some characteristic in a population. We estimate p using the sample proportion p̂ = x/n, where x is the number of successes and n is the sample size. Because p̂ is itself a random variable, it has a sampling distribution, and the entire framework of inference for proportions rests on understanding that distribution. The Central Limit Theorem guarantees that, under certain conditions, this sampling distribution is approximately normal — a fact that underlies both confidence intervals and hypothesis tests. Choosing the right procedure is not automatic. You must first identify the type of question being asked: 'What is the true proportion?' calls for a confidence interval, while 'Is the proportion equal to some claimed value?' calls for a hypothesis test. You must also verify that the sample is random, that observations are independent, and that the large-sample normality conditions are met. Skipping these checks can lead to invalid conclusions, so the selection process is just as important as the calculation itself.

Key takeaways
  • Identify the Research Question
  • Check the Randomness Condition
  • Check the Independence Condition
  • Check the Large-Sample Normality Condition
  • Select the Specific Procedure
  • Determine the Alternative Hypothesis Direction (For Tests Only)

Identify the Research Question

Ask yourself: Is the goal to estimate the population proportion, or to test a specific claim about it? Estimation → use a confidence interval. Testing a claim → use a one-proportion z-test. For example, 'What proportion of students pass the exam?' requires estimation, while 'Is the pass rate different from 70%?' requires a hypothesis test.

Check the Randomness Condition

The sample must be obtained by a random process — a simple random sample (SRS), stratified random sample, or a randomized experiment. If the data come from a convenience sample or voluntary response, formal inference is unreliable. This condition ensures that p̂ is an unbiased estimator of p.

Check the Independence Condition

Individual observations must be independent. If sampling without replacement from a finite population, apply the 10% condition: the sample size n must be no greater than 10% of the population size N (i.e., n ≤ 0.10N). This keeps each observation's outcome from substantially influencing others.

Check the Large-Sample Normality Condition

For a confidence interval, verify np̂ ≥ 10 and n(1 − p̂) ≥ 10. For a hypothesis test, use the claimed proportion p₀ instead: np₀ ≥ 10 and n(1 − p₀) ≥ 10. These conditions ensure the sampling distribution of p̂ is approximately normal, which justifies using z-procedures. If the conditions fail, you may need an exact binomial method.

Select the Specific Procedure

Once conditions are verified: (a) One-proportion z-interval: Use when estimating p. Formula is p̂ ± z* √(p̂(1−p̂)/n). (b) One-proportion z-test: Use when testing H₀: p = p₀. The test statistic is z = (p̂ − p₀) / √(p₀(1−p₀)/n). (c) Two-proportion z-procedures: Use when comparing proportions from two independent groups. Conditions must be checked separately for each group.

Determine the Alternative Hypothesis Direction (For Tests Only)

If testing, decide whether the alternative hypothesis is one-sided (Hₐ: p < p₀ or Hₐ: p > p₀) or two-sided (Hₐ: p ≠ p₀) based on the research question — not on what the data show. A two-sided test is appropriate when you have no prior reason to expect a direction; one-sided is used when theory or context dictates a specific direction before data collection.

The Key Equation

z=p^−p0p0(1−p0)nz = \frac{\hat{p} - p_0}{\sqrt{\dfrac{p_0(1-p_0)}{n}}}
  • \hat{p} — Sample proportion (dimensionless): The observed proportion of successes in the sample, calculated as x/n.
  • p_0 — Null hypothesis proportion (dimensionless): The claimed population proportion stated in the null hypothesis H₀.
  • n — Sample size (count): The total number of observations in the sample.
  • \sqrt{p_0(1-p_0)/n} — Standard error (under H₀) (dimensionless): The standard deviation of the sampling distribution of p̂ when H₀ is true. Using p₀ here (not p̂) is what distinguishes tests from confidence intervals.

Worked Example

Problem. A school claims that 65% of its students complete all homework assignments. A random sample of 120 students finds that 66 completed all assignments. At the α = 0.05 significance level, is there convincing evidence that the true proportion differs from 65%? Show full procedure selection and carry out the test.

z≈−2.30,p-value≈0.021<0.05z \approx -2.30, \quad p\text{-value} \approx 0.021 < 0.05

Answer. Reject H₀; the data provide convincing evidence that the true proportion differs from 0.65 (z ≈ −2.30, p-value ≈ 0.021).

Figure
Diagram
Selecting a Procedure for ProportionsResearch QuestionAbout one proportion?Estimate p→ Confidence IntervalTest a Claim→ Hypothesis TestEstimateTestConditions for CI:Random, Independence (10%),np̂ ≥ 10 and n(1−p̂) ≥ 10Conditions for Test:Random, Independence (10%),np₀ ≥ 10 and n(1−p₀) ≥ 10p̂ ± z* √(p̂(1−p̂)/n)One-proportion z-intervalz = (p̂−p₀)/√(p₀(1−p₀)/n)One-proportion z-testBoth procedures require all conditions to be verified before use.
Origins · 1 minJerzy Neyman and how this was discovered
JN
Jerzy Neyman
1894–1981 · Polish-American

Jerzy Neyman, working with Egon Pearson, developed the formal framework of hypothesis testing and the theory of confidence intervals in the 1930s. His concept of a confidence interval — a random interval that contains the true parameter with a specified probability — is the direct theoretical basis for the one-proportion z-interval used today. Neyman also formalized the notion of Type I and Type II errors, giving statisticians a rigorous language for evaluating test procedures. Neyman originally trained as a pure mathematician in Poland and only turned to statistics when he was asked to consult on agricultural experiments — he reportedly had to teach himself the subject quickly. He later founded the statistics department at UC Berkeley, which became one of the world's premier statistics programs.

  • Confidence Intervals for Proportions — The one-proportion z-interval is one of the two main procedures selected in this framework; understanding its construction and interpretation is essential.
  • Hypothesis Testing for Proportions — The one-proportion z-test is the other main procedure; this concept covers computing the test statistic, p-value, and drawing conclusions.
  • Sampling Distribution of the Sample Proportion — Both z-procedures are built on the approximately normal sampling distribution of p̂; verifying the normality condition is a key step in procedure selection.
  • Comparing Two Proportions — When the research question involves two independent groups, the selection process extends to two-proportion z-intervals and z-tests.