Double-blind
Neither the participant nor the experimenter knows who's in which group.
What it means
A double-blind design conceals treatment assignment from both the participants and the researchers who interact with them or assess outcomes, until the data are collected. It guards simultaneously against two distinct threats: participants' expectancy effects and experimenters' conscious or unconscious influence on delivery, measurement, and interpretation. Double-blinding is the standard for rigorous clinical trials precisely because expectations operate on both sides of the encounter. Where even the analyst is kept unaware until the analysis plan is locked, the design approaches triple-blinding.
Why both sides are blinded
Single-blinding hides the assignment from the participant, which quiets expectancy: a patient who believes they got the real drug may report feeling better regardless of biology. But the person delivering and judging the treatment carries expectations too. An enthusiastic clinician can dose more carefully, prompt more warmly, or read an ambiguous symptom as improvement when they know the patient is on the active arm. These effects are mostly unconscious, so goodwill and training do not remove them. Double-blinding closes the second channel by keeping assignment coded until outcomes are recorded. The threats are separable, which is why the design targets both: participant blinding protects reported experience, and staff blinding protects delivery and measurement. Removing one while claiming both is a common and consequential reporting error.
What the evidence shows
The clearest empirical signal is for measurement. Hrobjartsson and colleagues assembled trials that recorded the same outcome with both blinded and non-blinded assessors, and found non-blinded assessors exaggerated odds ratios by about 36 percent on average, with wide variation. The effect concentrates on subjective judgments, where an assessor's expectation has room to move the score. Yet the picture is not uniform. The large MetaBLIND meta-epidemiological study of 142 meta-analyses found no clear average difference in effect estimates between blinded and non-blinded trials overall, though its estimates were imprecise and could not rule out important bias for subjective outcomes. The honest reading: lack of blinding is a genuine and sometimes large risk for soft outcomes, but not a fixed multiplier applied to every trial.
Checking whether the blind held
Blinding is a procedure, not a guarantee. Distinctive side effects, taste, or visible efficacy routinely let participants and staff guess their arm, and once someone guesses correctly the design has quietly reverted to single- or un-blinded. The obvious remedy is to ask participants and staff at the end which arm they think they were on, and to test whether guesses beat chance. In practice this is rarely done: Fergusson and colleagues found that fewer than one in ten placebo-controlled trials in leading journals reported the success of blinding, and among those that did, blinding was often imperfect. Interpreting such tests is itself contested, because correct guesses can reflect a real treatment effect rather than a broken blind, but the near-total silence means most trials cannot say whether their central safeguard actually worked.
Where it breaks down
Some interventions resist blinding entirely. Surgery, psychotherapy, devices, exercise and diet cannot be disguised from the person receiving them, and sometimes not from the person delivering them either; here rigor shifts to blinding the outcome assessor and using objective endpoints. Active placebos, which mimic a drug's side effects, can preserve the blind that an inert sugar pill would betray. Blinding also decays over a long trial as effects accumulate, and can be deliberately broken for safety when a serious adverse event needs the treatment known. None of this makes double-blinding optional where it is feasible; it clarifies that the label certifies a design attempted, not a blind maintained. The strongest trials pre-register how assignment is concealed, keep the analyst blinded until the analysis plan is locked, and report whether the blind survived.
Examples
In a drug trial, identical-looking pills are dispensed by code so that neither the patient nor the examining physician can tell active from placebo.
At a serious wine tasting, bottles are poured from numbered decanters and even the pourer does not know which is the expensive one, so no glance or pause can tip the tasters off.
A food company tests a reformulated recipe against the old one using unlabelled cups handed out by staff who also cannot tell them apart, so nobody's face gives the game away.
In an agricultural field trial, plots are planted from numbered seed packets so the technician scoring yield and disease cannot tell which cultivar or fertilizer each plot received, keeping their ratings honest.
In animal behavior research, observers score coded video clips without knowing which animals received the compound, so their expectations cannot nudge inherently subjective judgments like activity or distress.
First described in Mid-20th-century clinical methodology.
Key references
- Moustgaard, H., Clayton, G. L., Jones, H. E., et al. (2020). Impact of blinding on estimated treatment effects in randomised clinical trials: meta-epidemiological study. BMJ, 368, l6802. doi.org/10.1136/bmj.l6802
- Hrobjartsson, A., Thomsen, A. S. S., Emanuelsson, F., et al. (2012). Observer bias in randomised clinical trials with binary outcomes: systematic review of trials with both blinded and non-blinded outcome assessors. BMJ, 344, e1119. doi.org/10.1136/bmj.e1119
- Fergusson, D., Glass, K. C., Waring, D., & Shapiro, S. (2004). Turning a blind eye: the success of blinding reported in a random sample of randomised, placebo controlled trials. BMJ, 328(7437), 432. doi.org/10.1136/bmj.37952.631667.EE
- Podolsky, S. H. (2023). The (Harry) Gold standard: angina, suggestion and the path to the 'double-blind' test and clinical pharmacology. Part 1: angina relief and suggestion. Journal of the Royal Society of Medicine, 116(7), 246-251. doi.org/10.1177/01410768231183237
- Kaptchuk, T. J. (1998). Intentional ignorance: a history of blind assessment and placebo controls in medicine. Bulletin of the History of Medicine, 72(3), 389-433. doi.org/10.1353/bhm.1998.0159