Conclusions for Mean Tests
When we conduct a hypothesis test for a population mean, the final and most important step is translating our statistical result into a meaningful conclusion about the real-world situation. A p-value or a test statistic alone is meaningless without context — the whole point of statistical inference is to make decisions about populations based on sample data, and those decisions must be communicated in plain language that answers the original research question. Drawing a conclusion for a mean test involves three connected parts: a decision about the null hypothesis (reject or fail to reject), a statement of statistical significance, and a contextual interpretation that references the actual variable, the population, and the direction of any observed effect. A conclusion that simply says 'we reject H₀' without explaining what that means for sleep hours, reaction times, or exam scores is incomplete and unhelpful. This concept ties together everything in a one-sample or two-sample t-test: the hypotheses, the significance level α, the t-statistic, the p-value, and the confidence interval. Mastering contextual conclusions is essential for any statistical analysis you will ever perform, from high school projects to professional research reports.
- State Your Decision Rule
- Compare p-value to α and State the Decision
- Write the Contextual Conclusion
- Connect to a Confidence Interval (When Appropriate)
State Your Decision Rule
Before interpreting results, recall the decision rule: if p-value ≤ α, reject H₀; if p-value > α, fail to reject H₀. The significance level α is chosen before the test (commonly α = 0.05) and represents the maximum probability of a Type I error you are willing to accept. Never change α after seeing the data.
Compare p-value to α and State the Decision
Compute or read off the p-value from your t-distribution with the appropriate degrees of freedom. For a one-sample t-test, df = n − 1. For a two-sample t-test, df is approximated by the Welch formula. Explicitly compare: 'Since p = 0.023 < α = 0.05, we reject H₀.' This single sentence is your statistical decision.
Write the Contextual Conclusion
Now translate the decision into the language of the problem. A complete contextual conclusion contains four elements: (a) the decision (reject or fail to reject), (b) a reference to the null hypothesis value μ₀, (c) the name of the variable and population, and (d) the direction of the effect if H₀ is rejected. Example: 'We have sufficient evidence at the 5% significance level to conclude that the mean daily screen time of teenagers in the city is greater than 4 hours.' Notice this sentence never says 'prove' — we are making a probabilistic inference, not a mathematical proof.
Connect to a Confidence Interval (When Appropriate)
A confidence interval provides additional information beyond a binary reject/fail-to-reject decision. If you also constructed a 95% confidence interval for μ, you can state: 'We are 95% confident that the true mean screen time is between 4.3 and 5.1 hours.' When the hypothesized value μ₀ falls outside the confidence interval, this is consistent with rejecting H₀ at the corresponding α level. Always check this consistency and report the interval to communicate the magnitude and precision of your estimate.
The Key Equation
- \bar{x} — Sample mean (same as variable): The arithmetic average of the n observed data values.
- \mu_0 — Hypothesized mean (same as variable): The specific value of the population mean stated in the null hypothesis H₀: μ = μ₀.
- s — Sample standard deviation (same as variable): Measures spread in the sample; used to estimate the unknown population standard deviation σ.
- n — Sample size (count): Number of independent observations. Larger n reduces the standard error and increases power.
- \text{df} — Degrees of freedom (dimensionless): Controls the shape of the t-distribution used to find the p-value; equals n − 1 for a one-sample test.
Worked Example
Problem. A sports nutritionist claims that elite cyclists consume a mean of μ = 3500 kcal per day. A researcher suspects the true mean is higher and collects a random sample of n = 25 elite cyclists, finding x̄ = 3680 kcal and s = 310 kcal. Conduct a one-sided hypothesis test at α = 0.05 and write a complete contextual conclusion.
Answer. Reject H₀. There is sufficient evidence at α = 0.05 that elite cyclists' mean daily intake exceeds 3500 kcal.
Origins · 1 minWilliam Sealy Gosset and how this was discovered
Gosset developed the t-distribution in 1908 while working as a chemist at the Guinness brewery in Dublin, Ireland. Because Guinness prohibited employees from publishing research, he published under the pseudonym 'Student' — which is why the distribution is officially called Student's t-distribution. His work allowed valid inferences to be drawn from small samples when the population standard deviation is unknown, which is exactly the situation in virtually every real-world mean test. Gosset's famous paper 'The Probable Error of a Mean' was published in the journal Biometrika while he was analysing the quality of barley and hops used to brew Guinness beer — making one of statistics' most important distributions a product of beer science.
Related Concepts
- One-Sample t-Test — The one-sample t-test generates the t-statistic and p-value that feed directly into the conclusions covered in this concept.
- Confidence Intervals for Means — Confidence intervals complement hypothesis test conclusions by providing a range of plausible values for the true population mean μ.
- Type I and Type II Errors — Understanding what can go wrong with a conclusion — false rejections (Type I) and missed effects (Type II) — is essential context for interpreting any hypothesis test decision.
- Two-Sample t-Test — Conclusions for two-sample mean tests follow the same four-step framework but compare two population means, requiring contextual language about the difference μ₁ − μ₂.
- P-Value Interpretation — Correctly interpreting the p-value as a conditional probability under H₀ is a prerequisite for writing accurate contextual conclusions.