How One-Sample T-Tests Work in Hypothesis Testing

Hypothesis testing starts with two competing claims about a population parameter — here, a mean μ.

• Null hypothesis (H₀): the cautious default, usually H₀: μ = μ₀ (the mean equals a stated value)
• Alternative hypothesis (H₁ or Hₐ): what you suspect instead

Three common alternatives

• Two-sided: H₁: μ ≠ μ₀ (different in either direction)
• Left-tailed: H₁: μ < μ₀ (smaller than claimed)
• Right-tailed: H₁: μ > μ₀ (larger than claimed)

Write H₀ and H₁ before you look at the p-value. Changing the alternative after seeing the data is a classic way to fool yourself.

Example framing

A machine should fill bottles with 500 ml on average. H₀: μ = 500. If you only care about short fills, use H₁: μ < 500. If any miscalibration matters, use H₁: μ ≠ 500.

Product experiments with two proportions use a different setup — see the A/B testing significance guide for conversion-rate tests.

For a one-sample test of a mean with sample standard deviation s (population σ unknown), the test statistic is:

t = (x̄ − μ₀) / (s / √n)

• x̄ — sample mean
• μ₀ — hypothesized mean under H₀
• s — sample standard deviation
• n — sample size
• s / √n — standard error of the mean

Worked sketch

Suppose x̄ = 512, μ₀ = 500, s = 24, n = 36. Standard error = 24 / √36 = 4. Then t = (512 − 500) / 4 = 3.0. The sample mean sits three standard errors above the claim.

Why t, not z?

When σ is known and n is large, a z-test is common. In practice you usually estimate SD from the sample, so the t distribution (heavier tails) is the default for means. Related standardization ideas appear in the z-score guide — z measures how far a single point sits from a mean; t here measures how far a sample mean sits from μ₀ in SE units.

Plug the same inputs into the free hypothesis testing calculator to confirm the arithmetic.

For a one-sample t-test of the mean, degrees of freedom are usually df = n − 1. That value selects which t curve you use for critical values and p-values.

Why n − 1?

You already used the data once to estimate the mean; the sample SD is slightly biased if you divide by n, so inference uses n − 1 and the t family instead of a single normal curve.

• Small n → heavier tails → need a larger |t| to claim significance
• Large n → t looks nearly normal → results similar to a z-test

Assumptions to respect

• Observations are independent (random sample or randomized experiment)
• Data are approximately normal, or n is large enough for the CLT to help the mean
• You are testing a mean, not a median or a proportion (different tools)

Skewed data with tiny n can make p-values misleading. Plot or summarize the sample before you trust a borderline result.

The p-value is the probability, assuming H₀ is true, of seeing a t-statistic at least as extreme as yours (in the direction(s) allowed by H₁).

Decision rule with α

Pick a significance level before testing — often α = 0.05. If p ≤ α, reject H₀. If p > α, fail to reject H₀ (you do not “prove” H₀ true).

• Small p — data are surprising under H₀; evidence against the null
• Large p — data are compatible with H₀; not proof the claim is correct

What p is not

• Not the probability H₀ is true
• Not the probability you made a mistake in a simple sense without more structure
• Not effect size — a tiny, useless difference can be “significant” with huge n

Always report the mean difference (x̄ − μ₀) and context next to p. Percent-style effect talk sometimes helps stakeholders — see the percentage guide for relative change language after the test.

The alternative hypothesis decides which tail(s) of the t distribution count as “extreme.”

• Two-tailed: both far-left and far-right t values count. Use when any difference matters.
• One-tailed: only one side counts. Use when only “greater than” or only “less than” is scientifically meaningful and you committed to that direction in advance.

Practical warning

One-tailed tests make it easier to reject H₀ in the chosen direction and impossible to claim significance in the other. Do not switch to one-tailed after seeing which way the mean moved — that inflates false positives.

Same |t|, different p

For a symmetric t curve, a two-tailed p-value is roughly twice the one-tailed p for the same |t| (when the result is on the predicted side). Always match the calculator mode to the hypothesis you wrote.

The free hypothesis testing calculator supports one- and two-tailed testing so you can align the p-value with your H₁.

Use a fixed checklist so lab reports and homework stay consistent.

1. State H₀ and H₁ and choose α (for example 0.05).
2. Check assumptions (independence, rough normality / adequate n).
3. Compute x̄, s, and n (or take them from your study).
4. Calculate t = (x̄ − μ₀) / (s / √n) and df = n − 1.
5. Find the p-value for your tail choice (table, software, or calculator).
6. Decide reject or fail to reject H₀; write a sentence in context.
7. Report effect size language: how far is x̄ from μ₀ in original units?

Example conclusion sentence

“With n = 36, x̄ = 512 vs μ₀ = 500 (s = 24), t(35) ≈ 3.0, two-tailed p < 0.05. We reject H₀ and conclude the mean fill differs from 500 ml; the sample runs about 12 ml high.”

For SD steps behind s, revisit the standard deviation guide. For means, use the average calculator.

The tool focuses on one-sample t-tests for means: enter the pieces of the formula and read t and an approximate p-value.

1. Open the free hypothesis testing calculator.
2. Enter the sample mean x̄.
3. Enter the hypothesized population mean μ₀.
4. Enter the standard deviation s (sample SD in the usual one-sample setup).
5. Enter the sample size n.
6. Choose one-tailed or two-tailed to match H₁.
7. Read the t-statistic and approximate p-value; apply your pre-chosen α.

Features to expect

T-statistic generation, p-value approximation, null-style validation framing, and one/two-tailed modes — designed for quick homework checks and lab sanity tests, not as a replacement for full statistical software on messy designs.

Comparing two conversion rates? Switch mental models to proportions and the free A/B testing calculator. Standardizing a single score? Use the free z-score calculator.