Conclusions for Proportion Tests

When we conduct a hypothesis test for a population proportion, the final and most important step is not just calculating a p-value or test statistic — it is communicating what those numbers actually mean in the real world. A conclusion for a proportion test must be written in plain language that directly addresses the original research question, references the specific context of the study, and honestly acknowledges the limitations of what the data can tell us. The logic of hypothesis testing is built around two competing claims: the null hypothesis H₀, which typically states that a population proportion equals some baseline value p₀, and the alternative hypothesis Hₐ, which asserts that the true proportion is different, greater than, or less than p₀. After computing a test statistic z and a corresponding p-value, we compare the p-value to a chosen significance level α (commonly 0.05). If the p-value is less than α, we reject H₀; otherwise we fail to reject it. Crucially, these outcomes are not symmetric — rejecting H₀ provides statistical evidence for Hₐ, but failing to reject H₀ does not prove H₀ is true. Beyond the binary reject/fail-to-reject decision, good statistical conclusions address effect size, practical significance, the scope of inference (can results be generalized?), and potential sources of error or bias. A small p-value in a massive survey might detect a trivially tiny difference, while a large p-value in a tiny sample might simply reflect insufficient power. Understanding the distinction between statistical significance and practical importance is a hallmark of sophisticated data reasoning.

Key takeaways
  • State the Decision
  • Interpret in Context
  • Address Practical Significance
  • State the Scope of Inference
  • Acknowledge Potential Errors
  • Check Conditions Retrospectively

State the Decision

Compare your computed p-value to the significance level α. If p-value < α, the decision is to reject H₀. If p-value ≥ α, the decision is to fail to reject H₀. Always state the numerical values: for example, 'Since p-value = 0.032 < α = 0.05, we reject H₀.' Never say 'accept H₀' — that phrasing implies a certainty that the test cannot provide.

Interpret in Context

Translate the statistical decision into the language of the original problem. Name the population, name the variable, and state what the evidence suggests. Example: 'There is sufficient evidence to conclude that the proportion of students at Jefferson High who eat breakfast daily is greater than 0.40.' Always mention the direction of the alternative hypothesis (greater than, less than, or different from) and the claimed value p₀.

Address Practical Significance

A statistically significant result (small p-value) does not automatically mean a practically important one. Examine the sample proportion p̂ relative to p₀. If p̂ = 0.412 vs p₀ = 0.40, the difference is only 1.2 percentage points — possibly trivial in practice, even if statistically detectable with a very large sample. Conversely, a large difference (say p̂ = 0.62 vs p₀ = 0.40) that just misses significance with a tiny sample deserves mention as potentially meaningful.

State the Scope of Inference

Conclusions can only be generalized to the population from which the sample was randomly drawn. If students were surveyed using a voluntary online poll, the results cannot be generalized to all students. Additionally, even a properly randomized observational study cannot establish causation — it can only show association. Only a randomized experiment can support causal language such as 'X causes Y.'

Acknowledge Potential Errors

Every decision carries a risk of error. A Type I Error occurs when we reject a true H₀; its probability equals α. A Type II Error occurs when we fail to reject a false H₀; its probability is β, which depends on sample size, effect size, and α. If you reject H₀, acknowledge the possibility of a Type I Error. If you fail to reject H₀, consider whether the sample was large enough to detect a meaningful difference (i.e., whether the test had sufficient power, defined as 1 − β).

Check Conditions Retrospectively

A valid conclusion requires that the test conditions were met: (1) Random sampling or random assignment; (2) Independence (sample ≤ 10% of population for sampling without replacement); (3) Large sample condition: np₀ ≥ 10 and n(1 − p₀) ≥ 10, ensuring the sampling distribution of p̂ is approximately Normal. If conditions were not met, note that the conclusion should be interpreted with caution.

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 proportion of successes observed in the sample, computed as the number of successes divided by sample size n.
  • p_0 — Null proportion (dimensionless): The hypothesized value of the population proportion stated in H₀.
  • n — Sample size (count): The total number of independent observations in the sample.
  • z — Test statistic (standard deviations): How many standard errors the sample proportion lies from the null proportion; compared to the standard Normal distribution to find the p-value.

Worked Example

Problem. A school administrator claims that exactly 55% of students at a large high school support a new dress code policy (p₀ = 0.55). A student journalist randomly surveys n = 120 students and finds that 75 support the policy. Using α = 0.05, test whether the true proportion differs from 0.55, and write a complete contextual conclusion.

z≈1.651,p-value≈0.099>0.05⇒Fail to reject H0z \approx 1.651,\quad \text{p-value} \approx 0.099 > 0.05 \Rightarrow \text{Fail to reject } H_0

Answer. Fail to reject H₀. Insufficient evidence that the proportion differs from 0.55 at α = 0.05.

Figure
Diagram
Two-Sided Z-Test: Decision Regions (α = 0.05)-440z = -1.96z = 1.96RejectH₀α/2=0.025RejectH₀α/2=0.025Fail to Reject H₀p-value ≥ αz ≈ 1.65(our example)Shaded red regions = rejection regions; our z = 1.65 falls in the fail-to-reject zone.
Origins · 1 minJerzy Neyman and how this was discovered
JN
Jerzy Neyman
1894–1981 · Polish-American

Together with Egon Pearson, Neyman developed the formal framework of hypothesis testing — including the concepts of null and alternative hypotheses, Type I and Type II errors, significance levels, and the power of a test. This framework is the mathematical foundation underlying every conclusion written for a proportion test today. Neyman originally trained as a pure mathematician in Poland and only pivoted to statistics after corresponding with Karl Pearson. He later founded the statistics department at UC Berkeley, which became one of the most influential statistics programs in the world.

  • One-Sample Proportion Z-Test — The conclusion is the final step of the one-sample proportion z-test procedure; understanding the full test procedure is essential context.
  • Confidence Intervals for Proportions — Confidence intervals complement hypothesis tests by estimating the plausible range of the true proportion, adding richer context to any conclusion.
  • Type I and Type II Errors — Every conclusion must acknowledge the possibility of decision errors; understanding Type I and Type II errors is critical to honest interpretation.
  • P-Value Interpretation — The p-value is the primary piece of evidence used to reach a conclusion; correct interpretation of the p-value prevents common misstatements.
  • Statistical vs. Practical Significance — A complete conclusion addresses whether a statistically significant result is also practically meaningful, a distinction central to this concept.