Two-Tailed Tests

Posted on by user

Two-Tailed Tests

A two-tailed test assesses whether a sample statistic (usually a mean) differs significantly from a hypothesized population value in either direction — that is, whether it is either significantly higher or significantly lower. It is a fundamental tool in null-hypothesis significance testing used across science, manufacturing, finance, and quality control.

Key takeaways

  • A two‑tailed test checks for deviations in both directions from a hypothesized parameter.
  • If the test statistic falls in either tail beyond the chosen significance level, the null hypothesis is rejected.
  • Use two‑tailed tests when you care about any difference (increase or decrease), not a specific direction.
  • For small samples or unknown population variance, use a t‑test; for large samples with known/approximate variance, a z‑test is common.

How a two‑tailed test works (step‑by‑step)

  1. State hypotheses:
  2. Null hypothesis H0: parameter = specified value.
  3. Alternative hypothesis H1: parameter ≠ specified value.
  4. Choose significance level α (commonly 0.05). For a two‑tailed test, split α/2 into each tail.
  5. Compute the test statistic (z or t), using the sample data and appropriate standard error.
  6. Determine the critical values (for z, e.g., ±1.96 when α = 0.05) or compute the p‑value.
  7. Decision:
  8. Reject H0 if the test statistic lies beyond the critical values (or if p‑value < α).
  9. Otherwise, fail to reject H0.

Practical applications

  • Quality control: checking whether average fill weights deviate from a target (too low or too high).
  • Clinical trials: testing whether a new treatment effect differs from a control (better or worse).
  • Finance: comparing average fees, returns, or other metrics between providers or periods.

Example (packaging): If a bag is supposed to contain 50 candies, you might reject bags that average fewer than 45 or more than 55 (these cutoffs reflect your chosen tolerance and α split across both tails).

Worked example (brokerage fees)

Scenario: Population mean fee for broker ABC is $18 with population standard deviation ≈ $6. A sample of n = 100 from a new broker produces a sample mean of $18.75. Test whether the new broker’s mean differs from $18 at α = 0.05.

  1. H0: μ = 18
    H1: μ ≠ 18

  2. Standard error = σ / sqrt(n) = 6 / 10 = 0.6

  3. Z = (sample mean − hypothesized mean) / SE = (18.75 − 18) / 0.6 = 0.75 / 0.6 = 1.25

  4. Critical z for α = 0.05 (two‑tailed) is ±1.96. |1.25| < 1.96, so do not reject H0.

  5. p‑value = 2 * P(Z > 1.25) ≈ 2 * 0.1056 = 0.2112 (≈ 21.1%), which is > 0.05, same conclusion.

Conclusion: There is insufficient evidence to conclude the new broker’s average fee differs from $18.

Two‑tailed vs one‑tailed tests

  • Two‑tailed test: H1 is nondirectional (≠). Use when deviations in either direction matter.
  • One‑tailed test: H1 is directional (>, or <). Use only when you have a clear theoretical or practical reason to ignore deviations in the opposite direction.
    Choosing the wrong tail structure can change conclusions and should be decided before seeing the data.

What is a Z‑score?

A Z‑score expresses how many standard deviations a value is from the mean: Z = (value − mean) / standard deviation. Z = 0 means the value equals the mean; Z = ±1 means one standard deviation away. In large samples or known σ situations, Z is used to compute tail probabilities and p‑values.

Bottom line

Two‑tailed tests detect any significant difference from a hypothesized value, regardless of direction. They are widely applicable and require careful choice of significance level, correct test statistic (z or t), and a preplanned decision about directionality (two‑tailed vs one‑tailed) to ensure valid conclusions.