A sham control is designed to resemble an intervention without delivering its proposed active component. It can help separate a specific treatment effect from attention, ritual, expectation and measurement behaviour.
The Gold Standard is not merely to label a trial double-blind. It is to specify who is blinded, how the interventions can be distinguished, when guesses are collected and which biases remain if blinding fails.
Expectancy is not an accusation that an outcome is imaginary. Expectations can change reporting, effort, co-interventions, adherence and even physiological responses. Those pathways belong in the design.
The practical question is whether the observed contrast still answers the intended causal question when participants or staff can infer allocation.
eduKateSG · The Gold Standard · Experimental Design
Find your next learning step
Design the causal contrast first, then test which roles could learn the assignment and how that knowledge could change outcomes.
Full chapter index · Worked example · Transfer check · Experimental Design subject root
Continue through The Gold Standard of Education, the Gold Standard collection and the How X Works hub.
Chapter contents
Define the contrast · Chapters 1–3
Measure integrity · Chapters 4–6
Reduce bias · Chapters 7–9
Transfer and audit · Chapters 10–12
CHAPTER 1 OF 12 · Define the contrast
1. A sham defines the comparison, not just the procedure
The treatment effect in a trial is always relative to a control condition. A sham may reproduce preparation, contact, sensory cues and follow-up while withholding the proposed active mechanism.
Changing the sham changes the estimand. A minimal sham may estimate the whole package effect; a highly matched sham may isolate a narrower component.
| Control | What it holds constant | Typical contrast |
|---|---|---|
| No intervention | Natural course and measurement | Whole offered intervention |
| Attention control | Contact time | Treatment beyond attention |
| Sham procedure | Ritual and sensory experience | Proposed active component |
| Active comparator | Alternative treatment | Relative effectiveness |
Participants, interventionists, clinicians, outcome assessors, adjudicators, laboratory staff, data managers and analysts may have different access to allocation.
Use role-specific language rather than the vague labels single-blind or double-blind. Explain who was blinded, when and by what procedure.
CHAPTER 3 OF 12 · Define the contrast
3. Map how the active and sham conditions can be distinguished
Side effects, device sensations, visible settings, treatment duration, staff behaviour and post-procedure instructions can reveal allocation. Even packaging or scheduling may provide clues.
Before launch, rehearse both pathways and record distinguishable cues. Some cues can be matched; others require blinded assessment, standardised scripts or more objective endpoints.
| Cue | Who may notice | Design response |
|---|---|---|
| Distinct sensation | Participant | Active placebo or matched simulation |
| Different procedure time | Staff and participant | Standardised visit structure |
| Characteristic adverse effect | Participant and clinician | Measure guesses and expectations |
| Visible device setting | Assessor | Independent blinded evaluator |
A trial randomises 120 participants equally to an active procedure or sham. At follow-up, 45 of 60 active participants guess active. In the sham group, 20 guess active and 40 correctly guess sham.
The mean outcome improves by 12 points in the active arm and 7 in sham, a five-point difference. The guesses show that allocation may have been partly distinguishable, but they do not tell how much of the five points is specific treatment, expectation or an effect-driven guess.
| Assigned arm | Guess active | Guess sham | Correct-guess rate |
|---|---|---|---|
| Active | 45 | 15 | 75.0% |
| Sham | 20 | 40 | 66.7% |
| Total | 65 | 55 | 70.8% |
Practical check: collect the reason and confidence for each guess. A participant may infer assignment from benefit, harm, a sensory cue or something a staff member said.
Guesses collected only after outcomes can be consequences of treatment response. Someone who improves may guess active because they improved, not because the procedure was distinguishable at baseline.
When feasible, measure expectations before allocation, anticipated assignment after the procedure and guesses at several planned times. Interpret each measure according to its timing.
CHAPTER 6 OF 12 · Measure integrity
6. Do not condition casually on post-randomisation guesses
Comparing outcomes among people who guessed correctly and incorrectly can destroy the balance created by randomisation. Guessing is influenced by treatment, side effects and outcome.
Use such analyses descriptively or within a prespecified causal framework. The primary intention-to-treat contrast remains anchored to random assignment.
A credible sham can require sensory cues or minor procedures, but it should not expose participants to avoidable harm merely to improve resemblance. Ethics, feasibility and scientific value must be weighed together.
When a full sham is not justified, use blinded independent outcome assessment, objective measurements and transparent limits rather than pretending complete blinding.
| Design goal | Possible tool | Limit |
|---|---|---|
| Match sensation | Active placebo or simulation | May introduce its own effects or risks |
| Match contact | Scripted visits | Cannot hide obvious procedures |
| Protect outcome rating | Blinded assessor | Participant-reported outcomes remain exposed |
| Protect analysis | Masked group labels | Does not repair earlier behaviour changes |
Knowledge of assignment can change encouragement, follow-up intensity, rescue treatment and adherence advice. Use scripts, visit checklists and prespecified co-intervention rules where they fit the clinical context.
Audit deviations by arm. An equal visit count does not guarantee equal communication.
CHAPTER 9 OF 12 · Reduce bias
9. Separate expectancy from reporting and performance pathways
Participants may report symptoms differently, practise more, seek other care or persist longer when they expect benefit. Clinicians may interpret ambiguous findings differently when they infer treatment.
Measure baseline expectations, treatment credibility, adherence, co-interventions and assessor blinding. These variables illuminate pathways but do not automatically convert an imperfect trial into a fully adjusted causal estimate.
CHAPTER 10 OF 12 · Transfer and audit
10. Transfer check: change the intervention to psychotherapy
A psychotherapy trial cannot blind the therapist to the technique being delivered. Participants may also recognise the approach. The design can still blind independent assessors, standardise contact time and measure treatment credibility and expectations.
If the control receives supportive conversation, the contrast is the specific therapy package beyond that support. If the control is a waitlist, the contrast includes attention, scheduling and expectancy.
Transfer question: in an exercise trial, which cues reveal assignment, which outcomes can be assessed blindly and which co-interventions could differ because participants know their programme?
A sham is not automatically inert. It can create physiological, behavioural or experiential effects. Conversely, a credible sham is not proof that blinding succeeded.
Correct guesses do not always mean broken blinding: a real treatment effect can reveal assignment. Incorrect guesses do not prove absence of bias. Blinding evidence must be interpreted with timing and reasons.
Report the control rationale, matched and unmatched features, blinded roles, allocation concealment, guess questions, timing, reasons, expectancy measures, deviations and any arm differences in co-interventions.
Connect these observations to the estimand and limits. A useful conclusion states what the randomised contrast supports and which specific mechanism remains uncertain.
Evidence check: a reader should be able to identify every role, every plausible unblinding cue, the timing of integrity checks and the analysis choices made before outcome data were seen.
Frequently asked questions
Does failed blinding invalidate a randomised trial?
Not automatically. Randomisation can remain intact, but behaviour, reporting and assessment biases may alter the observed contrast. The design and interpretation should address those pathways.
Should allocation guesses be tested against 50%?
A simple test may be descriptive, but unequal guess options, “do not know” responses and outcome-driven guesses complicate interpretation. Report timing, reasons and confidence.
Is an active placebo always better than an inert sham?
No. It may improve resemblance but can introduce effects or risks of its own. The choice must fit the causal question and ethical constraints.
Sources and further reading
- U.S. FDA — Design Considerations for Pivotal Clinical Investigations for Medical Devices
- U.S. FDA — Guidance on blinding and sham effects in device trials
- Orsini and colleagues — Systematic review of blinding integrity in psychedelic trials
- SPIRIT 2025 — Explanation and elaboration for trial protocols
- Butler and colleagues — Expectancy in placebo-controlled psychedelic trials
Connected eduKate reading: treatment fidelity and estimands · noncompliance and complier effects · attrition and post-randomisation selection · interim monitoring and early stopping
Return to reading routes · Return to the Experimental Design subject root