Two results can fail to differ significantly because they are close, because the measurements are imprecise or both. Equivalence and non-inferiority claims ask more specific questions. They require a justified margin defining which differences would matter and evidence precise enough to assess that margin.
Identify which comparison was intended
AHRQ’s guidance distinguishes equivalence within a predefined margin from non-inferiority relative to a one-directional limit. It cautions against treating the absence of a statistically significant difference as evidence of equivalence.AHRQ — Assessing Equivalence and Noninferiority: Guidance (opens in a new tab)
The distinction matters in research summaries describing two conditions as the same. That phrase may mean no difference was detected, that the estimates were similar or that a formal equivalence criterion was met. These are different statements.
| Claim | Question |
|---|---|
| Superiority | Is one outcome better under the specified comparison? |
| Equivalence | Are differences small enough in both directions? |
| Non-inferiority | Can an unacceptable disadvantage be excluded? |
| No detected difference | Did the stated difference test fail to reject its null? |
This article explains interpretation of study claims. It does not establish that any research peptide, formulation or supplier product is therapeutically interchangeable with another.
Compare interval limits with a stated margin
For an original nonclinical illustration, suppose a study defines differences within −5 to +5 response units as small enough for its specific equivalence question. Assume its planned assessment uses an appropriate two-sided interval criterion and that the reported intervals below are the required intervals.
| Interval for the difference | Relation to the ±5-unit limits |
|---|---|
| −2 to +3 | Entirely within the limits |
| −9 to +8 | Extends beyond both limits |
| +1 to +3 | Within the limits despite excluding zero |
The first pattern meets the stated interval criterion; the second is too imprecise to do so. The third shows why a detectable nonzero difference can still be small enough to meet an equivalence margin. The questions about exact zero and about acceptable size are different.
Do not turn non-inferiority into equality
Suppose larger values are favourable and the difference is new condition minus reference. A non-inferiority claim might require excluding a disadvantage worse than −5 units. A result well above zero can meet that lower-bound requirement without being equivalent within ±5.
Reverse the subtraction order or use an outcome where smaller is favourable and the relevant direction changes. Read the defined contrast before interpreting which interval boundary matters.
AHRQ emphasises a justified, preferably prespecified margin and sufficient precision relative to it. The interval method, confidence level and hypothesis should be consistent with the study’s chosen inferential framework.AHRQ — Assessing Equivalence and Noninferiority: Guidance (opens in a new tab)
Do not widen the acceptable margin after seeing an inconvenient interval without acknowledging the change. A post-result choice can make similarity easier to claim while weakening the link to the original scientific question.
Read similarity alongside the study’s ability to compare
Precision alone is not enough if the measurement cannot distinguish the relevant outcomes or if both conditions were implemented poorly. Similar observed results can arise because a study was insensitive to the contrast it intended to evaluate.
Missing outcomes, exclusions and departures from the intended comparison can also affect a similarity claim. They deserve appraisal in relation to the specific analysis rather than being ignored because the estimates look close.
Keep the conclusion tied to the outcome and setting. Equivalence on one assay readout does not establish identical chemistry, identical behaviour on other outcomes or comparable clinical benefit and risk.
A clear summary names the comparison, margin, interval and criterion, then states whether that criterion was met and what limitations remain. It avoids replacing a bounded statistical conclusion with the much broader word identical.
Sources and further detail
- AHRQ — Assessing Equivalence and Noninferiority: Guidance (opens in a new tab)
2012 methods guidance, sections on margins, interval precision and language read. Used for interpretation, not current regulatory approval criteria. All ±5-unit intervals are original hypothetical cases; the chosen interval criterion is assumed rather than prescribing a universal confidence level.
Sources checked 19 September 2026. Worked examples are illustrative unless a supplied report is explicitly identified. This article has not undergone independent scientific peer review.