Blinding
Also known as: Masking, Single-blind
Hiding who got the treatment so expectations don't contaminate results.
What it means
Blinding is the practice of withholding knowledge of treatment assignment from those who could otherwise bias a study's conduct or outcomes. Single-blinding keeps participants unaware; further blinding can extend to those delivering the intervention, assessing outcomes, and analyzing data. Its purpose is to neutralize placebo responses, demand characteristics, differential care, and biased measurement that arise when people know who received what. Blinding is most critical for subjective outcomes and least feasible for interventions that announce themselves, such as surgery or exercise.
How knowledge leaks in
Blinding guards the stretch of a trial after randomization, where knowledge of assignment leaks into the data through several separate channels. A participant who knows she received the active drug may report more improvement, try harder, or stay enrolled when a placebo patient would quit, so the arms are no longer compared on equal terms. A clinician who knows tends to watch treated patients more closely and add small co-interventions. An assessor who knows nudges an ambiguous scan or rating toward the hoped-for result. Even the statistician, deciding how to handle outliers or which model to run, can be swayed once the groups are labeled. Each is a distinct leak, which is why blinding is layered rather than a single switch.
What the evidence shows
The rationale for blinding is strong, but the empirical size of the bias it prevents is genuinely contested. Within-trial comparisons, where the same outcome is scored by both a blinded and a non-blinded assessor, show large effects: across binary outcomes non-blinded assessors exaggerated odds ratios by about 36 percent, and on rating scales by more than half a standard deviation. Between-trial evidence tells a quieter story. Schulz's 1995 survey found non-double-blind trials overstated effects by roughly 17 percent, but a much larger 2020 meta-epidemiological study of over 1,100 trials found no average difference between blinded and unblinded trials at all. The gap is real: between-trial comparisons are noisy and confound blinding with everything else that differs, while within-trial designs isolate the assessor's judgment.
Blinding versus allocation concealment
Blinding is routinely confused with allocation concealment, and the two protect against different biases at different moments. Allocation concealment happens at the point of randomization: it hides the upcoming assignment from whoever enrolls patients, so a recruiter cannot steer a sicker patient toward the control arm or wait for a better prognosis before revealing the next slot. It prevents selection bias in who ends up where. Blinding happens afterward: it hides the assignment already made, so expectations cannot color treatment, reporting, or measurement. A trial can conceal allocation yet be impossible to blind, as with surgery. Empirically, poor concealment inflates effects more consistently than poor blinding, which is one reason the two should never be collapsed into a single quality score.
When blinding fails, and how to tell
Blinding can be broken by the intervention itself. A drug with a distinctive taste, a flushing side effect, or an unmistakable benefit can reveal to participants and staff which arm they are in, quietly unravelling the design partway through. Researchers sometimes ask participants to guess their assignment to check whether the blind held, but the test is contested: correct guesses may reflect that the treatment worked rather than that blinding failed, so a broken blind and a real effect are hard to tell apart. For interventions that announce themselves, the fallback is to blind the outcome assessor instead, or to use sham procedures and objective endpoints. Reporting which of these was done matters more than the label 'double-blind,' which clinicians and authors define inconsistently.
Examples
Coding pills as 'A' and 'B' so neither patient nor nurse knows which is the active drug.
Orchestras that audition players behind a screen judge the sound alone, because knowing who is about to play quietly colours what the panel thinks it hears.
A supermarket testing its own-brand coffee pours every sample into unmarked cups: shoppers told which cup came from the premium bag rate it higher before they have even tasted it.
A patient in a pain study is not told whether her pill is the real drug or a sugar pill, so hope for relief cannot inflate how much improvement she reports.
Particle physicists sometimes add a secret numerical offset to their data and remove it only after the analysis is fixed, so no one can nudge the method toward the result they hope to find.
First described in Eighteenth-century origins; standardized in 20th-century trials.
Key references
- Salazar, J., Moustgaard, H., Bracchiglione, J., & Hróbjartsson, A. (2025). Empirical evidence of observer bias in randomized clinical trials: updated and expanded analysis of trials with both blinded and non-blinded outcome assessors. Journal of Clinical Epidemiology, 183, 111787. doi.org/10.1016/j.jclinepi.2025.111787
- 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
- Hróbjartsson, A., Thomsen, A. S. S., Emanuelsson, F., et al. (2013). Observer bias in randomized clinical trials with measurement scale outcomes: a systematic review of trials with both blinded and nonblinded assessors. CMAJ, 185(4), E201-E211. doi.org/10.1503/cmaj.120744
- Hróbjartsson, 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
- Schulz, K. F., Chalmers, I., Hayes, R. J., & Altman, D. G. (1995). Empirical evidence of bias: dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA, 273(5), 408-412. doi.org/10.1001/jama.1995.03520290060030