Perhaps the mental health diagnoses should be given in percentiles.
Some people complain that the definitions keep expanding, so that these days too many kids are diagnosed with ADHD or autism. The underlying reason is that these things seem to be on a scale, so it is arbitrary where you draw the line, and I guess people keep looking at those slightly below the line and noticing that they are not too different from those slightly above the line, and then they insist on moving the line.
But the same thing does not happen with IQ, despite the great pressure against politically incorrect results, despite the grade inflation at schools. That is because IQ is ultimately measured in percentiles. No matter how much pressure there is to say that everyone is above the average, the math only allows 50% of people to be smarter than the average, only 2% to be smarter than 98%, etc.
Perhaps we should do the same with ADHD and autism, too. Provide the diagnosis in form of: “You are more hyperactive than 85% of the population”, controlled for age, maybe also for sex if the differences are significant. So you would e.g. know that yes, your child is more hyperactive than average, but not like super exceptionally hyperactive, because there are two or three kids with a comparable diagnosis in every classroom. That would provide more useful information than simply being told “no” in 1980, or “yes” in 2020.
Objection: Some things are not linear! People are already making this objection about autism. It is also popular to make it against intelligence, and although the IQ researchers are not impressed, this meme refuses to die. So it is very likely that the moment we start measuring something on a linear scale, someone will make this objection. My response is that I do not see how a linear scale is worse than a binary choice (a degenerate case of linear scale) that we have now.
A better objection is that some things are… uhm, let me give you an example: You hurt your hand very painfully, so you ask a doctor whether it is broken. The doctor looks at an x-ray and says: “well, it is more broken than hands of 98% percent of the population”. WTF, was that supposed to be a yes or no?
So, the percentiles can also hide an important information, especially when the underlying data are bimodal or something like that. Perhaps in such cases it would help to provide a histogram with the data and a mark saying “you are here”, with the percentile.
It seems that the broken hand example is similar to situations where we have a deep understanding of the mechanics of how something works. In those situations, it makes more sense to say “this leg is broken; it cannot do 99% of the normal activities of daily living.” And the doctor can probably fix the leg with pins and a cast without much debate over exactly how disabled the patient is.
Yeah, having or not having a gears model makes a big difference. If you have the model, you can observe each gear separately, for example look at a hurting hand and say how damaged are bones, ligaments, muscles, skin. If you don’t have a gears model, then there is just something that made you pay attention to the entire thing, so in effect you kinda evaluate “how much this matches the thing I have in my mind”.
For example, speaking of intelligence, I have heard a theory that it is a combination of neuron speed and short term memory size. No idea whether this is correct or not, but using it as a thought experiment, suppose that it is true and one day we find out exactly how it works… maybe that day we will stop measuring IQ and start measuring neuron speed and short term memory size separately. Perhaps instead of giving people a test, we will measure the neuron speed directly using some device. We will find people who are exceptionally high at one of these things and low at the other, and observing them will allow us to even better understand how this all works. (Why haven’t we found such people already, e.g. using factor analysis? Maybe they are rare in nature, because the two things strongly correlate. Or maybe it is very difficult to distinguish them by looking at the outputs.)
Similarly, a gears model might split the diagnosis of ADHD into three separate numbers, and autism into seven. (Numbers completely made up.) Until then, we only have one number representing the “general weirdness in this direction”. Or a boolean representing “this person seems weird”.
I don’t think we can measure most of these closely enough, and I think the symptom clustering is imperfect enough that this doesn’t provide enough information to be useful. And really, neither does IQ—I mean it’s nice to know that one is smart, or not, and have an estimate of how different from the average one is, but it’s simply wrong to take any test result at face value.
In fact, you do ask the doctor if your hand is broken, but the important information is not binary. It’s “what do I do to ensure it heals fully”. Does it require surgery, a cast, or just light duty and ice? These activities may be the same whether it’s a break, a soft-tissue tear, or some other injury.
Likewise for mental health—the important part of a diagnosis isn’t “how severe is it on this dimension”, but “what interventions should we try to improve the patient’s experience”? The actual binary in the diagnosis is “will insurance pay for it”, not “what percent of the population suffers this way”.
If you want to know whether someone would benefit from a drug or other mental treatment the percentage is irrelevant.
Diagnoses are used to determine whether insurance companies have to pay for treatment. The percentage shouldn’t matter as much as whether the treatment is helpful for the patient.
Perhaps the mental health diagnoses should be given in percentiles.
Some people complain that the definitions keep expanding, so that these days too many kids are diagnosed with ADHD or autism. The underlying reason is that these things seem to be on a scale, so it is arbitrary where you draw the line, and I guess people keep looking at those slightly below the line and noticing that they are not too different from those slightly above the line, and then they insist on moving the line.
But the same thing does not happen with IQ, despite the great pressure against politically incorrect results, despite the grade inflation at schools. That is because IQ is ultimately measured in percentiles. No matter how much pressure there is to say that everyone is above the average, the math only allows 50% of people to be smarter than the average, only 2% to be smarter than 98%, etc.
Perhaps we should do the same with ADHD and autism, too. Provide the diagnosis in form of: “You are more hyperactive than 85% of the population”, controlled for age, maybe also for sex if the differences are significant. So you would e.g. know that yes, your child is more hyperactive than average, but not like super exceptionally hyperactive, because there are two or three kids with a comparable diagnosis in every classroom. That would provide more useful information than simply being told “no” in 1980, or “yes” in 2020.
Objection: Some things are not linear! People are already making this objection about autism. It is also popular to make it against intelligence, and although the IQ researchers are not impressed, this meme refuses to die. So it is very likely that the moment we start measuring something on a linear scale, someone will make this objection. My response is that I do not see how a linear scale is worse than a binary choice (a degenerate case of linear scale) that we have now.
A better objection is that some things are… uhm, let me give you an example: You hurt your hand very painfully, so you ask a doctor whether it is broken. The doctor looks at an x-ray and says: “well, it is more broken than hands of 98% percent of the population”. WTF, was that supposed to be a yes or no?
So, the percentiles can also hide an important information, especially when the underlying data are bimodal or something like that. Perhaps in such cases it would help to provide a histogram with the data and a mark saying “you are here”, with the percentile.
It seems that the broken hand example is similar to situations where we have a deep understanding of the mechanics of how something works. In those situations, it makes more sense to say “this leg is broken; it cannot do 99% of the normal activities of daily living.” And the doctor can probably fix the leg with pins and a cast without much debate over exactly how disabled the patient is.
Yeah, having or not having a gears model makes a big difference. If you have the model, you can observe each gear separately, for example look at a hurting hand and say how damaged are bones, ligaments, muscles, skin. If you don’t have a gears model, then there is just something that made you pay attention to the entire thing, so in effect you kinda evaluate “how much this matches the thing I have in my mind”.
For example, speaking of intelligence, I have heard a theory that it is a combination of neuron speed and short term memory size. No idea whether this is correct or not, but using it as a thought experiment, suppose that it is true and one day we find out exactly how it works… maybe that day we will stop measuring IQ and start measuring neuron speed and short term memory size separately. Perhaps instead of giving people a test, we will measure the neuron speed directly using some device. We will find people who are exceptionally high at one of these things and low at the other, and observing them will allow us to even better understand how this all works. (Why haven’t we found such people already, e.g. using factor analysis? Maybe they are rare in nature, because the two things strongly correlate. Or maybe it is very difficult to distinguish them by looking at the outputs.)
Similarly, a gears model might split the diagnosis of ADHD into three separate numbers, and autism into seven. (Numbers completely made up.) Until then, we only have one number representing the “general weirdness in this direction”. Or a boolean representing “this person seems weird”.
I don’t think we can measure most of these closely enough, and I think the symptom clustering is imperfect enough that this doesn’t provide enough information to be useful. And really, neither does IQ—I mean it’s nice to know that one is smart, or not, and have an estimate of how different from the average one is, but it’s simply wrong to take any test result at face value.
In fact, you do ask the doctor if your hand is broken, but the important information is not binary. It’s “what do I do to ensure it heals fully”. Does it require surgery, a cast, or just light duty and ice? These activities may be the same whether it’s a break, a soft-tissue tear, or some other injury.
Likewise for mental health—the important part of a diagnosis isn’t “how severe is it on this dimension”, but “what interventions should we try to improve the patient’s experience”? The actual binary in the diagnosis is “will insurance pay for it”, not “what percent of the population suffers this way”.
If you want to know whether someone would benefit from a drug or other mental treatment the percentage is irrelevant.
Diagnoses are used to determine whether insurance companies have to pay for treatment. The percentage shouldn’t matter as much as whether the treatment is helpful for the patient.