One-Tailed Hypothesis Test: From Intuition to Precision

Imagine you're a quality control manager at a factory that produces 500 mL water bottles. The machine is set to fill exactly 500 mL, but you suspect it has started underfilling — cheating customers. You don't care if it overfills (that's fine), only if it underfills. This suspicion is a one-sided question: "Is the mean fill less than 500 mL?" That's the heart of a one-tailed test.

The Intuition: A Directional Question

A hypothesis test is a formal way to decide between two competing claims about a population parameter (like a mean or proportion). A one-tailed (or one-sided) test asks a question with a direction: "Is the parameter less than some value?" or "Is it greater than some value?" It's like asking "Is the new drug better than the old one?" (not just different). The alternative hypothesis points in one direction only.

The Precise Statement: Hypotheses and the Rejection Region

Every hypothesis test has two hypotheses:

• Null hypothesis ($H_0$): The default assumption — no effect, no difference. It always contains an equality ($=$, $\leq$, or $\geq$).
• Alternative hypothesis ($H_1$ or $H_a$): What you want to prove — the research claim. It contains a strict inequality ($<$ or $>$).

For a one-tailed test, the alternative hypothesis is directional. There are two types:

The rejection region (the set of sample outcomes that lead you to reject $H_0$) lies entirely in one tail of the sampling distribution. For a left-tailed test, it's the left tail; for a right-tailed test, it's the right tail.

How It Works: The Decision Rule

You collect a sample, compute a test statistic (like a $z$-score or $t$-score), and compare it to a critical value that marks the boundary of the rejection region. The critical value depends on your chosen significance level $\alpha$ (typically 0.05) and the direction of the test.

• Left-tailed test: Reject $H_0$ if test statistic $< -z_{\alpha}$ (or $< -t_{\alpha}$).
• Right-tailed test: Reject $H_0$ if test statistic $> z_{\alpha}$ (or $> t_{\alpha}$).

Alternatively, you can compute a p-value — the probability of observing a test statistic as extreme as yours, in the direction of $H_1$, assuming $H_0$ is true. For a one-tailed test, the p-value is the area in that single tail. If p-value $< \alpha$, reject $H_0$. …