Concept Architecture
Concept
Theoretically, Two-Tailed Test is a statistical hypothesis testing procedure used to determine whether an observed parameter differs significantly from a hypothesised value in either direction. It evaluates evidence against the null hypothesis without specifying the direction of the effect, making it appropriate when both increases and decreases are scientifically plausible. The concept is founded on classical statistical inference and decision theory and is widely applied in health economics, clinical trials and health technology assessment.
Mathematically, a two-tailed test is based on a test statistic whose sampling distribution is known under the null hypothesis. The significance level is divided equally between the two tails of the distribution, and the null hypothesis is rejected if the observed test statistic falls into either critical region. Depending on the statistical model, test statistics may follow the standard normal, t, ?� or F distribution.
In practice, a two-tailed test is performed by specifying the null and alternative hypotheses, calculating the appropriate test statistic, and comparing the resulting p-value with the chosen significance level. In health economics, two-tailed tests are routinely used to compare mean costs, quality-adjusted life-years, regression coefficients and treatment effects when no justified directional hypothesis exists.
Purpose
Used to determine whether an observed parameter differs significantly from a hypothesised value in either direction while controlling the overall probability of a Type I error.
Mathematical Formulae
Primary Formula
H?: ? = ??
H?: ? ? ??
Supporting Formulae
Z = (x? ? ??) / (� / �n)
t = (x? ? ??) / (s / �n)
Reject H? if |Test Statistic| > Critical Value
or equivalently
Reject H? if p < �
where each tail contains �/2.
Related Mathematical Methods
- Hypothesis Testing
- Z-Test
- Student's t-Test
- Wald Test
- Likelihood Ratio Test
- Score Test
- p-Value
- Confidence Interval
Example
A health economist compares the mean annual healthcare cost of patients receiving a new intervention with a historical mean of �9,500. From a sample of 100 patients, the mean cost is �9,900 with a sample standard deviation of �1,600.
t = (9,900 ? 9,500) / (1,600 / �100)
= 400 / 160
= 2.50
Using a two-tailed significance level of � = 0.05, the p-value is approximately 0.014. Because p < 0.05, the null hypothesis is rejected, indicating that the mean annual healthcare cost differs significantly from �9,500.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| T.TEST | =T.TEST(B2:B101,C2:C101,2,2) | Perform a two-tailed two-sample t-test comparing treatment groups. |
| T.DIST.2T | =T.DIST.2T(ABS(E2),99) | Calculate the two-tailed p-value from a t-statistic. |
| NORM.S.DIST | =2*(1-NORM.S.DIST(ABS(E2),TRUE)) | Calculate a two-tailed p-value using the standard normal distribution. |
| IF | =IF(F2<0.05,"Reject H?","Do not reject H?") | Interpret statistical significance. |
VBA (Optional)
Automate two-tailed hypothesis testing across multiple clinical or economic outcomes and generate summary tables of test statistics and p-values.
Sources
- Casella G, Berger RL. Statistical Inference.
- Lehmann EL, Romano JP. Testing Statistical Hypotheses.
- Agresti A. Foundations of Linear and Generalized Linear Models.
- Drummond MF, et al. Methods for the Economic Evaluation of Health Care Programmes.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
- NICE. Health Technology Evaluation Manual.
Related Concepts (2)
Library
Publications
1
Statistical Analysis of Cost-Effectiveness Data — Willan & Briggs, 1st Edition ed., 2006 (John Wiley & Sons)
A synthesis of statistical methods for analysing cost-effectiveness data, including net-benefit regression, confidence intervals for the ICER, cost-effectiveness acceptability curves, and covariate adjustment. Part of the Wiley Statistics in Practice series.
BookView source →
Frequently Asked Questions (6)
What is a two-tailed test?
A hypothesis test assessing evidence for an effect in either direction, without specifying in advance whether it is expected to be better or worse.
Source: Neyman & Pearson 1933
Why does a two-tailed test consider both directions of an effect?
A two-tailed test looks for evidence of an effect in either direction, whether a treatment turns out better or worse, rather than committing in advance to one. It considers both possibilities because at the outset a treatment could plausibly help or harm, and honesty requires allowing for a surprise in either direction. To keep the overall false-positive risk fixed, it splits the significance level between the two tails, which makes it slightly more demanding than a one-tailed test but far more defensible. Allowing for a surprise either way is its rationale. Kirkwood and Sterne (2003) describe this.
Source: Kirkwood & Sterne 2003
When is a two-tailed test used?
A two-tailed test is used when an effect in either direction is of interest or possible, which is the usual situation, since typically it matters whether a treatment is better or worse, not only whether it is better. So a two-tailed test is the default choice for most hypothesis tests, appropriate whenever the direction of a possible effect is not confidently predicted in advance or an effect either way would be relevant, which is why it is generally preferred over a one-tailed test, being more conservative and able to detect effects in both directions rather than only one.
Source: Neyman & Pearson 1933
How does a two-tailed test differ from a one-tailed test?
A two-tailed test assesses evidence for an effect in either direction, splitting the significance level between both tails, while a one-tailed test assesses evidence in only one specified direction, placing the whole significance level in one tail. The two-tailed test is more conservative and can detect effects either way, whereas the one-tailed test is more powerful in its chosen direction but blind to the other. So the two differ in whether they consider one or both directions, with the two-tailed test generally preferred unless a one-sided alternative is strongly justified, since it guards against missing an effect in the unconsidered direction.
Source: Neyman & Pearson 1933
Why is a two-tailed test usually preferred?
A two-tailed test is usually preferred because it considers effects in both directions, guarding against missing a real effect opposite to what was expected, and because it avoids the potential misuse of a one-tailed test to more easily obtain significance. So a two-tailed test is generally the default for its conservatism and its ability to detect effects either way, which is why it is recommended unless there is strong, prespecified justification for testing only one direction, since assuming a direction and using a one-tailed test risks overlooking important effects the other way and can inflate the false positive rate if not properly justified.
Source: Neyman & Pearson 1933
How does a two-tailed test split the significance level?
A two-tailed test splits the significance level between the two tails of the distribution, so that for a five per cent level, 2.5 per cent is placed in each tail, and the null is rejected if the test statistic falls in either extreme. So a two-tailed test divides the significance level equally between the upper and lower tails, which is why its critical values are further from the centre than those of a one-tailed test at the same overall level, reflecting that it allocates the false positive risk to both directions, allowing it to detect effects either way while keeping the total type I error at the chosen level.
Source: Neyman & Pearson 1933
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 26 Dec 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/two-tailed-test
- Term code
- HE-ES-SA-223
Stable URI · Machine-readable · Resolvable · CC BY 4.0