[mortehu]

If a randomized controlled trial includes just two groups, one which gets a treatment and another which gets placebo, it cannot attribute any results to the placebo, because it's missing a control for the placebo, i.e. a group which gets no intervention at all. Many studies have made this error, but that doesn't mean there's never a placebo effect.

http://www.dcscience.net/2015/12/11/placebo-effects-are-weak...

[perl4ever]

I don't disagree exactly about missing a control for the placebo, but I think a control for a placebo is logically impossible, and therefore nobody has ever, or can ever, measured a placebo effect. Does that mean "there's never a placebo effect"? That's a philosophical question.

[dragonwriter]

> I don't disagree exactly about missing a control for the placebo, but I think a control for a placebo is logically impossible, and therefore nobody has ever, or can ever, measured a placebo effect.

The control for a placebo is no intervention, and not only has placebo effect been measured, but differential placebo effects by administration method have been measured (a placebo treatment that involves piercing the skin, for instance, has a bigger effect than a pill.) This is why a valid placebo needs to match administration method of the intervention being tested.

[perl4ever]

There is no way to coherently define the "placebo effect".

When you test a treatment vs. placebo, and speculate that there is a "placebo effect", it means that you are supposing that lying about treatment produces a therapeutic effect.

But you can't test a placebo vs. nothing, because a placebo is not a treatment, and nothing is not a placebo.

Giving a sugar pill and disclosing you are giving a sugar pill is not the same thing as lying about treatment (unless the recipient believes it is efficacious). Lying about it being efficacious doesn't help your experiment unless it's a credible lie.

Giving "nothing" cannot be a control because it is obviously different from a sugar pill. So you cannot do a blind test.

The idea of testing a placebo effect is an epistemological morass.

[collyw]

Couldn't you,

-Give people an effective drug tell them so.

-Give people a placebo tell them its a real drug.

-Give people a real drug tell them it may be placebo.

-Give them a placebo tell them it may be placebo.

It would certainly tell us something about the placebo.

[bcook]

You mention the word "tell" often, which makes me feel... uncertain.

I get your point but GP seems to understand uncertainty better.

You can't force me to believe.

[xapata]

That study wouldn't be checking placebo but whether the intervention beats placebo. Not sure why that's a mistake if measuring placebo is not a research goal.

[mortehu]

The mistake would be attributing any outcome to the placebo intervention. The mistake isn't in the construction of the experiment, but in the interpretation of its results.

[JamesBarney]

But a lot of reported placebo effects are more likely reversion to the mean.