One-Sided / Two-Sided Disclosure
A one-sided test halves the p-value compared to the equivalent two-sided test. Using a one-sided test without a pre-specified, scientifically justified directional hypothesis is equivalent to running a two-sided test and dividing the p-value by two — which is not a legitimate statistical practice. Licklider requires an explicit declaration and, for one-sided tests, a stated hypothesis before the result can be reported.
STEP 1 —The Pitfall: One-Sided Tests as a p-Value Reduction Strategy
One-sided (directional) tests are appropriate when the researcher has a strong, pre-specified hypothesis about the direction of the effect —for example, "treatment will increase tumor regression rate" —and when an effect in the opposite direction would be irrelevant or impossible. In practice, these conditions are rarely met, and one-sided tests are often applied retrospectively to borderline results.
- The p-value halving problem. A two-sided t-test produces p = 0.06. The researcher switches to a one-sided test (hypothesizing improvement) and reports p = 0.03. The analysis is presented as significant when the two-sided result was not. This is a form of p-hacking.
- No declared hypothesis. A one-sided test without a pre-specified directional hypothesis is a two-sided test with the unfavorable tail removed. If the hypothesis was not declared before seeing the data, the one-sided test is not justified.
- Missing direction in the Methods. Papers that use one-sided tests rarely state this explicitly in the Methods. A reader seeing "p = 0.03" cannot know whether this is a one-sided or two-sided result —a difference that matters for interpretation and meta-analysis.
STEP 2 —Journal Requirement
JAMA's statistical guidelines state that "one-tailed tests should be used only when there is a clear and convincing scientific rationale that effects in one direction are impossible or uninteresting, and this rationale should be stated explicitly." BMJ states that one-sided tests "require a prior and specific directional hypothesis —if used, the rationale must appear in the Methods."
The FDA's guidance on clinical trial statistical analysis requires that all inferential tests be declared as one-sided or two-sided in the Statistical Analysis Plan, with the rationale for one-sided tests explicitly justified. ICH E9 similarly requires pre-specification of the directionality of all confirmatory tests before data are unblinded.
STEP 3 —Licklider's Solution
Input
- An explicit selection of test directionality: Two-sided / One-sided (greater) / One-sided (less).
- For one-sided tests: a mandatory free-text field for the pre-specified directional hypothesis ("I hypothesize that treatment will increase the outcome, because窶ｦ").
- The analysis intent from the Data Contract (confirmatory / exploratory). One-sided tests in confirmatory analyses receive heightened scrutiny.
Output
- Directionality label in figure annotation. The generated significance annotation always includes the directionality: "p = 0.03 (one-sided, greater)" rather than simply "p = 0.03."
- Hypothesis record in audit log. The declared directional hypothesis is logged with a timestamp and included in the session audit record, providing verifiable evidence that the hypothesis was pre-specified.
- Methods text contribution. The directionality and hypothesis text are automatically included in the auto-generated Methods paragraph (see STEP 4).
Guard
Guard condition —One-sided test without hypothesis (blocking): One-sided test selection cannot proceed without a completed directional hypothesis. The analysis is blocked until the hypothesis text field is filled. The hypothesis must contain at least 20 characters to discourage trivial entries.
Guard condition —One-sided confirmatory test (advisory): When a one-sided test is applied to a confirmatory analysis, an advisory is displayed: "One-sided tests in confirmatory analyses require strong prior justification. Verify that the directional hypothesis was pre-specified before data collection." The advisory is included in the session risk log.
STEP 4 —Draft Output (Draft / Needs review)
Example: Two-sided test (standard)
Group differences were assessed using a two-sample Welch t-test (two-sided).
No directional hypothesis was specified prior to analysis.Example: One-sided test with pre-specified hypothesis
Group differences were assessed using a one-sample Welch t-test (one-sided,
greater). The directional hypothesis —that systemic treatment would increase
tumor regression rate relative to local treatment, based on the systemic
mechanism of the drug —was pre-specified and logged prior to data unblinding
(directional hypothesis ID: hyp-3f2a, timestamp: 2026-04-01T09:15:22Z).
A one-sided test was used because effects in the opposite direction (systemic
worse than local) were considered clinically irrelevant given the known
pharmacokinetics of the treatment.