The very smart Ezra Klein is very confused about how to measure the placebo effect and what to do if it is a large fraction of the treatment effect.

His post here.

Briefly he assumes that changes in people treated with a placebo are the effect of the placebo treatment. This is not at all what the placebo parts of controlled studies are for. Change with placebo is the course of the syndrome with no treatment plus the placebo effect.

Also, even if Klein had data on the placebo effect, the correct comparison is not placebo effect divided by treatment effect, but the difference compared to costs including financial costs and the costs of side effects.

Klein seems to think that the cost of off patent anti-depressants is a noticeable fraction of the cost of treating depression. Of course it isn't. The cost is almost entirely the cost of diagnosis and even the most minimal counselling and follow up which would be needed with placebo treatment. The cost of the pills is almost entirely the cost of checking purity, advertizing and profits for the manufacturers which again would still with placebo treatment (or could be eliminated with the same un-acceptable level of risk for both placebo and anti-depressant treatment).

My full indignant comment below.

I do not believe that it is significantly cheaper to give a placebo than say generic fluoxetine Prozac (off patent). The cost of producing the active ingredient in the pill is tiny. On the other hand, fluoxitine causes an improvement on the Hamilton scale which is strongly statistically significantly greater than the placebo.

There is no reason to look at the ratio of the treatment effect to the placebo effect. The comparison should be between the lowest plausible estimate of the difference in the effects
and the costs of the treatment (not just in dollars but also in side effects such as, gasp, weight loss which are, I'd guess a main reason so many people take Prozac).

Also, the bar labled placebo is *not* an estimate of the placebo effect. It is a measurement of the change in symptoms of people who received the placebo. For all the data which you present show, it could just be the normal course of depression, knee pain or angina. To measure the placebo effect, you have to compare treatment with a placebo to no treatment. That means you have to get people to participate (which means return for the second measurement of their symptoms) even though they know that nothing is being done for them.

Consider the placebo effect on the common cold. If you have a cold and I give you a
sugar pill, in a week you will feel better. This is not an argument for giving people with colds sugar pills. That should be obvious. the argument you make in this post for syndromes other than the common cold is just as obviously invalid.


The evidence that prozac is better than the placebo is statistically significant at standard significance levels. The lower end of the 95% interval of the benefit from prozac over the placebo is definitely positive. Comparing it to the cost of prozac (tiny *including* measured side effects) is the reasonable thing to do.

We are talking about less than a gram per year per patient of a chemical which can be easily synthesized (plus the side effects which are a on average a benefit).

see http://tinyurl.com/5gdytl for more of my thoughts on the topic (it's a google search of this blog).

Update: However this is a very important public policy announcement. I didn't know Fafblog was back ( I mean I know it was back but I didn't know it is back). Fafblog -- more effective that a placebo -- the cure for what Ailes us.