(I couldn’t quickly find an ‘official authority’ saying the above, but it must somewhat reflect the mindset of some of them, since otherwise the case for single dose is pretty overwhelming).
Of course from a Bayesian perspective this entire idea of “no evidence to suggest” is almost always meaningless, as pointed out in the full thread of the above tweet: https://twitter.com/robertwiblin/status/1345800480144945152. There was plenty of evidence at the time for everything the WHO dismissed, as evidenced by all the people on this very site who got it right.
However not everyone thinks in terms of Bayesian statistics. Viewing the entire world as a probability distribution and acting accordingly, is not for the average person. Instead the way of deciding between the unsubstantiated and reality is via empirical science. Whilst not perfect, treating something as false until one has done a carefully regulated study is certainly far better than what we had in the past. It sounds at first like the WHO is making the correct decisions here—waiting till we have ‘evidence’ for something before acting on it, and evidence is not anecdotal data (under the empirical view of things), but double blinded placebo controlled studies. How do we articulate what the WHO did wrong here, without using the word Bayesian?
One idea I had was to use another commonly used phrase: “no reason to suggest”. Whilst they sound similar the phrases I think mean very different things to the average person.
To deal in extremes, consider the 2 statements:
1. There is no reason to suggest holding onto the tail of a plane as it takes off is dangerous.
2. There is no evidence to suggest holding onto the tail of a plane as it takes off is dangerous.
The first is obviously false. It takes about 2 seconds to think of reasons why it’s a terribly stupid idea.
The second is less obviously false. By evidence some people mean a certain level of rigorously done study. That has presumably never taken place for this exact question.
In other cases the two are likely to agree. For example there is no reason or evidence to suggest that the vaccine can make you infertile.
So here’s a simple trick: whenever you read a sentence containing the phrase “no evidence to suggest”, try replacing it with the phrase “no reason to suggest”.
If they both seem equally true, then that’s fine. If the latter seems obviously false, then the sentence is likely misleading.
And if, as is usually the case, the modified sentence seems less true, but not obviously false, then the claim is probably not as strong as it makes out, but still may be somewhat valid.
This is basically reinventing Bayesian statistics. However it doesn’t require any thinking about probability, priors, or technical lingo. It’s a simple heuristic to easily tell if in a particular case, a “no evidence” claim is informative. If there is strong reason to suggest something is true, even lacking evidence, it’s worth assuming it’s probably true.
Heuristic: Replace “No Evidence” with “No Reason”
The phrase “no evidence to suggest” has been used as an excuse to avoid inaction in the face of what is in fact ample evidence.
Health authorities continued bumbling response to coronavirus has really served to highlight this, starting right at the beginning:
https://twitter.com/WHO/status/1217043229427761152
And then continuing pretty steadily though the crisis:
https://twitter.com/UNGeneva/status/1244661916535930886
https://twitter.com/robertwiblin/status/1345800492367142917
(I couldn’t quickly find an ‘official authority’ saying the above, but it must somewhat reflect the mindset of some of them, since otherwise the case for single dose is pretty overwhelming).
Of course from a Bayesian perspective this entire idea of “no evidence to suggest” is almost always meaningless, as pointed out in the full thread of the above tweet: https://twitter.com/robertwiblin/status/1345800480144945152. There was plenty of evidence at the time for everything the WHO dismissed, as evidenced by all the people on this very site who got it right.
However not everyone thinks in terms of Bayesian statistics. Viewing the entire world as a probability distribution and acting accordingly, is not for the average person. Instead the way of deciding between the unsubstantiated and reality is via empirical science. Whilst not perfect, treating something as false until one has done a carefully regulated study is certainly far better than what we had in the past. It sounds at first like the WHO is making the correct decisions here—waiting till we have ‘evidence’ for something before acting on it, and evidence is not anecdotal data (under the empirical view of things), but double blinded placebo controlled studies. How do we articulate what the WHO did wrong here, without using the word Bayesian?
One idea I had was to use another commonly used phrase: “no reason to suggest”. Whilst they sound similar the phrases I think mean very different things to the average person.
To deal in extremes, consider the 2 statements:
1. There is no reason to suggest holding onto the tail of a plane as it takes off is dangerous.
2. There is no evidence to suggest holding onto the tail of a plane as it takes off is dangerous.
The first is obviously false. It takes about 2 seconds to think of reasons why it’s a terribly stupid idea.
The second is less obviously false. By evidence some people mean a certain level of rigorously done study. That has presumably never taken place for this exact question.
In other cases the two are likely to agree. For example there is no reason or evidence to suggest that the vaccine can make you infertile.
So here’s a simple trick: whenever you read a sentence containing the phrase “no evidence to suggest”, try replacing it with the phrase “no reason to suggest”.
If they both seem equally true, then that’s fine. If the latter seems obviously false, then the sentence is likely misleading.
And if, as is usually the case, the modified sentence seems less true, but not obviously false, then the claim is probably not as strong as it makes out, but still may be somewhat valid.
This is basically reinventing Bayesian statistics. However it doesn’t require any thinking about probability, priors, or technical lingo. It’s a simple heuristic to easily tell if in a particular case, a “no evidence” claim is informative. If there is strong reason to suggest something is true, even lacking evidence, it’s worth assuming it’s probably true.