An interesting blog post which argues that in medical studies the great majority of improvement in non-intervention arms that is attributed to the placebo effect actually comes from regression to the mean.
The issue is distinguishing placebo (defined as a psychosomatic effect) from “natural healing” and I suspect it will be not easy—in diseases where psychosomatic placebo “can’t happen”, can natural healing happen?
The question is whether those coherently exist.
If the placebo works for the disease humans in their natural enviroment might do something they believe will cure the disease and thus you have natural healing.
Same objection: do such exist? Can you give any examples?
The problem is that the difference between (psychosomatic) placebo and natural healing is just the involvement of the mind. If no natural healing is possible, what kind of magic is the mind doing?
It’s easier to exclude placebo—e.g. if the patient is in a long-term coma, no placebo effects seem to be possible.
I meant placebo as baseline effect (from all sources, psychosomatic or statistical), and the falsifiable prediction is it should drastically decrease in situations where regression to the mean should not happen.
Not clear why psychosomatic effects happen, may work in coma. Very clear why regression to the mean happens, well understood issue in sampling from a distribution. So: easier to exclude well-understood thing.
Actually, you can view this as a causal issue, the blog post is really about a type of selection bias, or “confounding by health status.”
edit: Lumifer, this is curious. I mentioned chronic disease in my original response. Do you … parse what people write before you respond?
I meant placebo as baseline effect (from all sources, psychosomatic or statistical), and the falsifiable prediction is it should drastically decrease in situations where regression to the mean should not happen.
I think the core point of that article (and one I agree with) is that if we want to attribute the ‘placebo effect’ to medical care, we need to measure not the difference between the patient before and after placebo treatment, but the difference between the after for no treatment and the after for placebo treatment. And so it seems very useful (for determining the social benefit of medicine / homeopathy / etc.) to separate out psychosomatic effects (which are worth paying for) from statistical effects (which aren’t worth paying for).
I think this part is a bit too strong, which corrupts one of the main points of the whole post:
The other contribution comes from the get-better-anyway effect. This is a statistical artefact and it provides no benefit whatsoever to patients.
It’s not called the stay-the-same-anyway effect, it’s called the get-better-anyway effect. The patient who reports lower pain a week later actually is in less pain. Health isn’t repeated draws from an urn: if you crack a rock one day it won’t regress to the mean. It’ll stay cracked. That people heal is not a statistical artefact.
That is, I agree much more with the O’Connell quote (emphasis mine):
If this finding is supported by future studies it might suggest that we can’t even claim victory through the non-specific effects of our interventions such as care, attention and placebo. People enrolled in trials for back pain may improve whatever you do. This is probably explained by the fact that patients enrol in a trial when their pain is at its worst which raises the murky spectre of regression to the mean and the beautiful phenomenon of natural recovery.
Regression to the mean plays a part, especially for chronic variable conditions like lower back pain or depression, but even there natural recovery plays a huge part (otherwise the condition would be a degenerative one).
It’s not called the stay-the-same-anyway effect, it’s called the get-better-anyway effect.
I agree, but here I am (uncharacteristically :-/) inclined to the charitable reading and treat “it” in “it provides no benefit whatsoever” as referencing placebo.
I would also think of regression to the mean (in this context) as an observable manifestation of “natural recovery” and not oppose them.
I think the structure of the paragraph is pretty clear (differentiating sentence, name A, explain A, name B, explain B, compare A and B), and the rest of the article matches my interpretation.
I would also think of regression to the mean (in this context) as an observable manifestation of “natural recovery” and not oppose them.
Yes, one could say that natural recovery is the mechanism by which regression to the mean works.
The chief thing I’m objecting to is the idea that the regression is in some way illusory or nonexistent. In the discussion of the NSLBP, for example, DC claims “none of the treatments work” when I think the result is the opposite, that “all of the treatments work.” Now, DC and I agree on the right course of treatment (do nothing) for the same reason (why spend more to get the same effect as doing nothing?), but we disagree on the presentation. Instead of “treatment” vs “no treatment,” both of which are equally ineffective, cast it as “natural recovery plus treatment” vs. “natural recovery alone,” both of which are equally effective.
Here you might get into an object level vs. meta level debate. I argue that one should talk up doing nothing instead of talking down treatments that are no better than doing nothing, because it will be hard to convince the man on the street reasoning by post hoc ergo propter hoc that his attempts did not actually lead to recovery, but if convinced to try doing nothing then the same fallacy will, when doing nothing turns out to work, cause him to gain trust in doing nothing. One could respond that the important point is not that he get the object level question right, but that he avoid fallacious reasoning.
cast it as “natural recovery plus treatment” vs. “natural recovery alone,” both of which are equally effective
That naturally leads to the effect of treatment being zero which is conventionally called “the treatment does not work”.
When you have some baseline process and some zero-effect interventions on top of it, I think it’s misleading to say that all these interventions work.
I argue that one should talk up doing nothing instead of talking down treatments that are no better than doing nothing
These, of course, are not mutually exclusive. Besides, you need to do something to counteract the proponents of the no-effect treatments—such people exist (typically they are paid for providing these treatments) and if you just ignore them they will dominate the debate.
The placebo group is called such because it receives the placebo treatment, not because medical researchers think all improvement in it is attributable to the placebo effect. Results are reported as improvement in the treatment arm vs. the placebo arm, and never have I seen these differences explicitly reported as treatment effect vs. placebo effect, and I’ve read hundreds of medical papers. The real magnitude of the placebo effect is almost never of interest in these papers. Some professionals in the medical community could have such a misconception because of the usual lack of scientific training, but I’d like to think they are a small minority.
If the placebo effect is of real importance, I think a more significant problem would be the lack of use of active placebos that mimick side effects since most drugs have them and this is a potential source of breaking the blinding of RCTs.
The placebo group is called such because it receives the placebo treatment, not because medical researchers think all improvement in it is attributable to the placebo effect.
Sure. But the question under discussion here is what actually is the placebo effect and how much of it can you attribute to psychosomatic factors and how much to just regression to the mean (aka natural healing).
You are correct in that most intervention studies don’t care about the magnitude of the placebo effect, they just take the placebo arm of the trial as a baseline. But that doesn’t mean that we couldn’t or shouldn’t ask questions about the placebo effect itself.
the question under discussion here is what actually is the placebo effect and how much of it can you attribute to psychosomatic factors and how much to just regression to the mean (aka natural healing).
In that case your opener is slightly polemical :)
But that doesn’t mean that we couldn’t or shouldn’t ask questions about the placebo effect itself.
Agreed. The problem with nonintervention arms for studying the placebo effect is that there aren’t clear incentives for adding them and they cost statistical power.
My n=1 experiment evidences against this. When my son was much younger and complained some part of him was hurting (because, say, he bumped against a wall) I would put lotion on the part and say it was powerful medicine. It usually made him feel better. And I wasn’t even lying because the medicine I had in mind was the placebo effect.
You were lying, because you were making a statement that you knew would be understood as an untruth and with the intention of it being understood as that untruth. The fact that it may be true using a definition that isn’t used by the target doesn’t change that.
Disagree. I believed that my statement would be interpreted as “this will reduce your pain.” Because of my belief in the placebo effect I really thought that the lotion would reduce my son’s pain.
...using a definition that isn’t used by the target...
I suspect you may be overestimating young childrens’ critical thinking abilities. If daddy say X is “powerful medicine”, then “powerful medicine” is defined as X.
An interesting blog post which argues that in medical studies the great majority of improvement in non-intervention arms that is attributed to the placebo effect actually comes from regression to the mean.
To really test this, we should see if placebo is much smaller in studies where this can’t happen (certain chronic diseases for example).
The issue is distinguishing placebo (defined as a psychosomatic effect) from “natural healing” and I suspect it will be not easy—in diseases where psychosomatic placebo “can’t happen”, can natural healing happen?
Pretty sure Ilya suggested the reverse—diseases where natural healing doesn’t happen, but the placebo effect is possible.
The question is whether those coherently exist. If the placebo works for the disease humans in their natural enviroment might do something they believe will cure the disease and thus you have
natural healing.Same objection: do such exist? Can you give any examples?
The problem is that the difference between (psychosomatic) placebo and natural healing is just the involvement of the mind. If no natural healing is possible, what kind of magic is the mind doing?
It’s easier to exclude placebo—e.g. if the patient is in a long-term coma, no placebo effects seem to be possible.
Physical injury, chronic disease.
I meant placebo as baseline effect (from all sources, psychosomatic or statistical), and the falsifiable prediction is it should drastically decrease in situations where regression to the mean should not happen.
Not clear why psychosomatic effects happen, may work in coma. Very clear why regression to the mean happens, well understood issue in sampling from a distribution. So: easier to exclude well-understood thing.
Actually, you can view this as a causal issue, the blog post is really about a type of selection bias, or “confounding by health status.”
edit: Lumifer, this is curious. I mentioned chronic disease in my original response. Do you … parse what people write before you respond?
I think the core point of that article (and one I agree with) is that if we want to attribute the ‘placebo effect’ to medical care, we need to measure not the difference between the patient before and after placebo treatment, but the difference between the after for no treatment and the after for placebo treatment. And so it seems very useful (for determining the social benefit of medicine / homeopathy / etc.) to separate out psychosomatic effects (which are worth paying for) from statistical effects (which aren’t worth paying for).
Sure, I agree. If the article is right.
I think this part is a bit too strong, which corrupts one of the main points of the whole post:
It’s not called the stay-the-same-anyway effect, it’s called the get-better-anyway effect. The patient who reports lower pain a week later actually is in less pain. Health isn’t repeated draws from an urn: if you crack a rock one day it won’t regress to the mean. It’ll stay cracked. That people heal is not a statistical artefact.
That is, I agree much more with the O’Connell quote (emphasis mine):
Regression to the mean plays a part, especially for chronic variable conditions like lower back pain or depression, but even there natural recovery plays a huge part (otherwise the condition would be a degenerative one).
I agree, but here I am (uncharacteristically :-/) inclined to the charitable reading and treat “it” in “it provides no benefit whatsoever” as referencing placebo.
I would also think of regression to the mean (in this context) as an observable manifestation of “natural recovery” and not oppose them.
I think the structure of the paragraph is pretty clear (differentiating sentence, name A, explain A, name B, explain B, compare A and B), and the rest of the article matches my interpretation.
Yes, one could say that natural recovery is the mechanism by which regression to the mean works.
The chief thing I’m objecting to is the idea that the regression is in some way illusory or nonexistent. In the discussion of the NSLBP, for example, DC claims “none of the treatments work” when I think the result is the opposite, that “all of the treatments work.” Now, DC and I agree on the right course of treatment (do nothing) for the same reason (why spend more to get the same effect as doing nothing?), but we disagree on the presentation. Instead of “treatment” vs “no treatment,” both of which are equally ineffective, cast it as “natural recovery plus treatment” vs. “natural recovery alone,” both of which are equally effective.
Here you might get into an object level vs. meta level debate. I argue that one should talk up doing nothing instead of talking down treatments that are no better than doing nothing, because it will be hard to convince the man on the street reasoning by post hoc ergo propter hoc that his attempts did not actually lead to recovery, but if convinced to try doing nothing then the same fallacy will, when doing nothing turns out to work, cause him to gain trust in doing nothing. One could respond that the important point is not that he get the object level question right, but that he avoid fallacious reasoning.
That naturally leads to the effect of treatment being zero which is conventionally called “the treatment does not work”.
When you have some baseline process and some zero-effect interventions on top of it, I think it’s misleading to say that all these interventions work.
These, of course, are not mutually exclusive. Besides, you need to do something to counteract the proponents of the no-effect treatments—such people exist (typically they are paid for providing these treatments) and if you just ignore them they will dominate the debate.
The placebo group is called such because it receives the placebo treatment, not because medical researchers think all improvement in it is attributable to the placebo effect. Results are reported as improvement in the treatment arm vs. the placebo arm, and never have I seen these differences explicitly reported as treatment effect vs. placebo effect, and I’ve read hundreds of medical papers. The real magnitude of the placebo effect is almost never of interest in these papers. Some professionals in the medical community could have such a misconception because of the usual lack of scientific training, but I’d like to think they are a small minority.
If the placebo effect is of real importance, I think a more significant problem would be the lack of use of active placebos that mimick side effects since most drugs have them and this is a potential source of breaking the blinding of RCTs.
Sure. But the question under discussion here is what actually is the placebo effect and how much of it can you attribute to psychosomatic factors and how much to just regression to the mean (aka natural healing).
You are correct in that most intervention studies don’t care about the magnitude of the placebo effect, they just take the placebo arm of the trial as a baseline. But that doesn’t mean that we couldn’t or shouldn’t ask questions about the placebo effect itself.
In that case your opener is slightly polemical :)
Agreed. The problem with nonintervention arms for studying the placebo effect is that there aren’t clear incentives for adding them and they cost statistical power.
My n=1 experiment evidences against this. When my son was much younger and complained some part of him was hurting (because, say, he bumped against a wall) I would put lotion on the part and say it was powerful medicine. It usually made him feel better. And I wasn’t even lying because the medicine I had in mind was the placebo effect.
You were lying, because you were making a statement that you knew would be understood as an untruth and with the intention of it being understood as that untruth. The fact that it may be true using a definition that isn’t used by the target doesn’t change that.
Disagree. I believed that my statement would be interpreted as “this will reduce your pain.” Because of my belief in the placebo effect I really thought that the lotion would reduce my son’s pain.
I suspect you may be overestimating young childrens’ critical thinking abilities. If daddy say X is “powerful medicine”, then “powerful medicine” is defined as X.
You were not measuring actual improvement—you were measuring the amount of whining/complaining.
Which is strongly correlated with pain. A reduction in pain is an actual improvement.
No, not in the sense we are talking about here. Pain is known to be quite psychosomatic, anyway.
Right, which is why the effect in the placebo arm is not called the placebo effect.