I was at the vet a while back; one of my dogs wasn’t well (she’s better now). The vet took her back, and after waiting for a few minutes, the vet came back with her.
Apparently there were two possible diagnosis: let’s call them x and y, as the specifics aren’t important for this anecdote.
The vet specifies that, based on the tests she’s run, she cannot tell which diagnosis is accurate.
So I ask the vet: which diagnosis has the higher base rate among dogs of my dog’s age and breed?
The vet gives me a funny look.
I rephrase: about how many dogs of my dog’s breed and age get diagnosis x versus diagnosis y, without running the tests you did?
The vet gives me another funny look, and eventually replies: that doesn’t matter.
My question for Lesswrong: Is there a better way to put this? Because I was kind of speechless after that.
“Base rate” is statistics jargon. I would ask something like “which disease is more common?” And then if they still don’t understand, you can explain that its probably the disease that is most common, without explaining Bayes rule.
Mightn’t the vet have already factored the base rate in? Suppose x is the more common disease, but y is more strongly indicated by the diagnostics. In such a case it seems like the vet could be justified in saying that she cannot tell which diagnosis is accurate. For you to then infer that the dog most likely has x just because x is the more common disease would be putting undue weight on the Bayesian priors.
So, it seems like there could be 2 things going on here:
1: Maybe, in your Vet’s mind, she is telling you “We can’t tell if this is A or B”, and you are asking “But which is it?”, and by refusing to answer she is doubling down on the whole “We don’t know A or B” situation.
Like, I know what you mean is what you actually said, but normal people don’t say that, and the Vet is trying to reiterate that you do not know which of A or B this is. She is trying to avoid saying “mostly A”, and you saying “ok, treat A”, and then the dog dies of B, and you are like “You said A, you fraud, I’m suing you!”.
2: The vet honestly doesn’t know the answer to your question. She is a person who executes the procedures in his/her manuals, not a person who follows the news about every animal’s frequent ailments. In her world if an animal shows A you do X, if an animal shows B you do Y. Your question is outside of her realm of curiosity.
As far as another way to phrase this, I’d go with “Well, which do you think it is, A or B?”. The vet’s answer ought to be informed by her experience, even if it isn’t explicitly phrased as “well, mostly this is what dogs suffer from”. If she reiterates that there is no way to know, I’d figure this was a first case, CYA situation, and stress that I wouldn’t be mad if she was wrong.
The vet honestly doesn’t know the answer to your question.
I’d suggest this is likely. Assuming both ailments are relatively common and not obviously known to be rare, I’d bet the vet just doesn’t know the data necessary to discuss base rates in a meaningful way that would help determine X or Y.
Side note: My experience is that sometime the tests needed to help narrow down illnesses in animals are prohibitively expensive.
I was at the vet a while back; one of my dogs wasn’t well (she’s better now). The vet took her back, and after waiting for a few minutes, the vet came back with her.
Apparently there were two possible diagnosis: let’s call them x and y, as the specifics aren’t important for this anecdote.
The vet specifies that, based on the tests she’s run, she cannot tell which diagnosis is accurate.
So I ask the vet: which diagnosis has the higher base rate among dogs of my dog’s age and breed?
The vet gives me a funny look.
I rephrase: about how many dogs of my dog’s breed and age get diagnosis x versus diagnosis y, without running the tests you did?
The vet gives me another funny look, and eventually replies: that doesn’t matter.
My question for Lesswrong: Is there a better way to put this? Because I was kind of speechless after that.
“Base rate” is statistics jargon. I would ask something like “which disease is more common?” And then if they still don’t understand, you can explain that its probably the disease that is most common, without explaining Bayes rule.
Mightn’t the vet have already factored the base rate in? Suppose x is the more common disease, but y is more strongly indicated by the diagnostics. In such a case it seems like the vet could be justified in saying that she cannot tell which diagnosis is accurate. For you to then infer that the dog most likely has x just because x is the more common disease would be putting undue weight on the Bayesian priors.
So, it seems like there could be 2 things going on here:
1: Maybe, in your Vet’s mind, she is telling you “We can’t tell if this is A or B”, and you are asking “But which is it?”, and by refusing to answer she is doubling down on the whole “We don’t know A or B” situation.
Like, I know what you mean is what you actually said, but normal people don’t say that, and the Vet is trying to reiterate that you do not know which of A or B this is. She is trying to avoid saying “mostly A”, and you saying “ok, treat A”, and then the dog dies of B, and you are like “You said A, you fraud, I’m suing you!”.
2: The vet honestly doesn’t know the answer to your question. She is a person who executes the procedures in his/her manuals, not a person who follows the news about every animal’s frequent ailments. In her world if an animal shows A you do X, if an animal shows B you do Y. Your question is outside of her realm of curiosity.
As far as another way to phrase this, I’d go with “Well, which do you think it is, A or B?”. The vet’s answer ought to be informed by her experience, even if it isn’t explicitly phrased as “well, mostly this is what dogs suffer from”. If she reiterates that there is no way to know, I’d figure this was a first case, CYA situation, and stress that I wouldn’t be mad if she was wrong.
I’d suggest this is likely. Assuming both ailments are relatively common and not obviously known to be rare, I’d bet the vet just doesn’t know the data necessary to discuss base rates in a meaningful way that would help determine X or Y.
Side note: My experience is that sometime the tests needed to help narrow down illnesses in animals are prohibitively expensive.
A straightforward question would be: “What’s the probability for diagnosis A and what’s the probability for diagnosis B?”.
Unfortunately you are likely out of luck because your vet doesn’t know basic statistics to give you a decent answer.