One-Tailed z-Tests
What you'll learn: How to perform hypothesis tests when you care about deviations in only one direction—either greater than or less than a hypothesized value.
The Core Idea
In a two-tailed test (previous lesson), you rejected H₀ when the test statistic fell in either extreme tail. But sometimes you only care about one direction. For example:
- A manufacturer claims their new process increases mean battery life (μ > μ₀)
- A doctor wants to know if a drug reduces mean blood pressure (μ < μ₀)
A one-tailed z-test focuses all your "rejection budget" (significance level α) into one tail of the normal distribution.
Right-Tailed Test (μ > μ₀)
Alternative hypothesis: H₁: μ > μ₀
You reject H₀ only if your test statistic z is too large (positive).
Critical value approach: Reject H₀ if z > z_α (the value where P(Z > z_α) = α)
- For α = 0.05: reject if z > 1.645
- For α = 0.01: reject if z > 2.326
P-value approach: Compute P(Z > z_observed) and reject if this probability < α.
Left-Tailed Test (μ < μ₀)
Alternative hypothesis: H₁: μ < μ₀
You reject H₀ only if your test statistic z is too small (negative).
Critical value approach: Reject H₀ if z < -z_α
- For α = 0.05: reject if z < -1.645
- For α = 0.01: reject if z < -2.326
P-value approach: Compute P(Z < z_observed) and reject if this probability < α.
Choosing the Direction
Your alternative hypothesis should match your research question before seeing the data. Think of it like aiming a spotlight: you illuminate only the region you care about, making it easier to detect effects in that direction.
Key Takeaway: One-tailed z-tests place all α in one tail, using critical values ±z_α instead of ±z_{α/2}, making them more sensitive to effects in the specified direction but blind to the opposite direction.