Blinding vs Double-Blind

Blinding is a design feature that keeps knowledge of the treatment assignment away from someone in the experiment. In a single-blind study, either the subjects do not know which treatment they are getting, or the people who interact with them and record the outcome do not. In a double-blind study neither side knows. Two different things are being protected: the response itself, which a subject's expectations can move, and the measurement of the response, which the recorder's expectations can move.

The second one is easy to underrate. Suppose 80 patients rate knee pain from 0 to 10 after an unmarked gel, 40 on the new formula and 40 on an identical inactive one. If the nurse recording the ratings knows who got what and records each treated patient half a point lower than the patient reported, the treated group's mean falls by exactly 0.5. No single rating looks wrong, and a half point on a 0 to 10 scale can be the whole of the effect the trial was built to find.

The misreading is a vocabulary slip: "neither the treatment group nor the control group knew which one they were in, so the study was double-blind." Both groups being kept in the dark is one blind, not two. The second blind covers a different kind of person, the one delivering the treatment or scoring the outcome.

Some treatments cannot be hidden from the subject. Nobody undergoing surgery or a twelve-week exercise program is unaware of it. The response can still be scored by someone who does not know the assignment, so blinding the assessor is the usual fallback, and this is also why a placebo matters: an inactive lookalike is what makes blinding the subject possible at all.

Blinding balances nothing. It cannot stand in for random assignment and it does not remove a confounding variable; what it removes is expectation. Single-blind and double-blind are named in topic 1.13 Experimental Design.

Double-blind names two roles, not two people. The subjects do not know which treatment they received, and neither do the people who interact with them or score the response. The assignment has not vanished: it sits in a code held by someone who never meets a subject, and it is opened once the data are in. Blind exactly one of those two roles and the design is single-blind, whichever one it is. None of this describes what the control group receives.

A cold remedy trial randomly assigns 240 volunteers, 120 to the drug and 120 to an identical placebo, and the response is days until symptoms clear, judged by a nurse working from a checklist. Suppose the nurse knows the assignment and resolves borderline cases a quarter of a day early in the drug group and a quarter of a day late in the placebo group. The estimated difference between the groups then moves by $0.25 + 0.25 = 0.5$ days on its own, against a real drug effect of 0.8 days. Most of what got measured is the nurse. Coding the bottles so she cannot tell removes it and costs nothing.

The sentence to correct is "the study was double-blind because the control group got a placebo." A placebo is something the control group receives; blinding is about who knows. Identical pills from bottles marked A and B, with the nurse holding the key, is single-blind. A second version says the researchers must not know who got what. Somebody has to: the person who ran the randomization knows, and that is fine. The ones who must be kept in the dark are those in contact with subjects or scoring the outcome.

Plenty of experiments cannot be double-blind at all. A subject knows whether she spent twelve weeks lifting weights. Blinding whoever measures the outcome is still available, still worth doing, and still single-blind. Blinding protects a measurement and does nothing to balance the groups, so an unblinded randomized experiment still supports a causal claim while a blinded study without random assignment does not.