Some of this terminology seems useful for talking about consciousness (which I suppose is a sort of psychiatric condition); there’s obviously a useful distinction between humans and algae, but trying to pin down an exact dividing line proves difficult, especially when you entertain thought experiments like removing one neuron at a time from a brain. Lots of debates about e.g. animal consciousness seem possible to phrase in terms of the shape of this graph. See also the Sorites Paradox.
"My guess is most professionals, and an overwhelming majority of laymen, are actually confused on this point, and this messes them up in a lot of ways."--how much of this is just reification as a crutch for computational efficiency given limited time/energy/attention?
The thought processes you are moving away from are simpler than the ones you are moving toward, and docs are pressed for time. Likewise, laypeople are mostly probably trying to make decisions about their lives with framings like 'I think that man may be a narcissist so I probably won't go out on a date with him'
The reifications have costs but also benefits and I'm glad you're digging deeper in your practice and inquiry. I worry somewhat about unintended consequences when categorical terms for dimensional characteristics escape the clinical setting and make their way into the ambient construct stew
As a side note, there are common categories in use for wealth like HNW, VHNW, UHNW, but they certainly don't capture the upper-end variation around someone like Bezos (for whom even 'billionaire' is off by orders of magnitude...)
The next time someone asks me about my mental health I’ll feel tempted to tell them that I’m “the Jeff Bezos of absentmindedness”. For a while now I’ve intuitively been moving away from the taxonometric definitions of mental illness so it is nice to read something that really goes into the math of it. Even in the cases where there is a definitive “reason” for the mental illnesses listed in the study, like the flu virus for the flu, it usually seems like it’s either some kind of traumatic brain injury (difficult to reverse for now) or something genetic (impossible to reverse fo now) so holding out for a magic bullet is implausible... It’s all symptom management. But symptom management is often what can help someone get over the arbitrary-feeling “line” between “Person doing poorly” and “Person doing well”!
Thank you for this excellent article. I found the concept of a "resident of Extremistan" to be an excellent phrase for describing someone like myself. I fall into the 2% or less in seven major categories, including autism spectrum and bipolar disorder.
I just signed up here as a founding member after Jonathan V. Last at The Bulwark gave a recommendation.
Statistical question / hypothesis regarding substance use disorder:
Many analytic methods don't deal well with massing at 0 and/or 1 (whatever the boundary condition is). There are large proportion of the population who are teetotalers, and a small proportion of the population of drinkers drive a significant amount of alcohol consumption. I assume (probably incorrectly) that many substances have similar distributions of use per capita/time.
Question - wouldn't this bimodal distribution drive taxonicity in statistical analyses? Despite this, substance use is not cleanly divided in the clinical context where substance use is 1) comorbid, 2) often a coping mechanism, 3) not necessarily problematic.
It seems like language is biased towards categorical thinking - either you choose to use a word or you don't. You need to know from context that it's not what people really mean, that tallness is dimensional. Sometimes it's tricky to hint at dimensionality without being overly vague.
Is “person with autism” really misguiding? I mean, we already have and use terms such as “people of color”, or “people of size” and I don't think anyone really assumes those are binary categories.
1. Why is CCFI of 0.5 chosen as the border between categorical and dimensional? In other words, it seems "being categorical" or "being dimensional" is itself dimensional - is this true?
2. Question about your practice (ignore if inappropriate) - comparing the post here and on your site, I see you've kept the personal tone, which surprised me. Did you get any feedback from patients/non-SSC-readers about the content there? It might be too soon to know, of course.
I wonder if the positive feedback loop of addiction makes it categorical. "Propensity to become addicted to cigarettes" could be dimensional, but "is currently addicted to cigarettes" (or "has ever been addicted to cigarettes") could be more like separate categories, because it means "consumed enough nicotine for the positive feedback loop to kick in." How much that positive feedback loop affects you varies, but that's like variance in the strength of the flu which makes the "has flu" lump wider than the "doesn't have flu" lump.
Maybe also relevant that "how much nicotine would it take to set off the positive feedback loop" and "how strongly will the positive feedback loop affect you if it gets triggered" are strongly correlated (I imagine) as two aspects of "propensity to become addicted to cigarettes." So the people who are more strongly affected by that positive feedback loop are also more likely to have triggered it (since that takes less nicotine for them).
My personal impression of autism is that it's both. Aspergers seems to me to be more dimensional, whereas low-functioning Autism (for lack of a better term) seems more taxonic.
I dont have any particular study to back this up, its just more of a result of all the stuff I have read about Autism. Its probably also influenced by my belief that it might be wrong (for social reasons) to lump the two together into one diagnosis.
One of the things I love about this blog is how you pull meaningful conclusions that have fascinating real-world implications out of math that goes over my head (but is still interesting to read an analysis of). I went into this one taking a gamble that it would be interesting to me, and hit pay dirt at the end. Cheers :) and again, welcome back!!
Has someone does the analysis to determine whether psychiatric disorders themselves split into two taxa of taxonic and non-taxonic, or if they lie on a spectrum between the two?
Psychological constructs in general are just low-dimensional approximations of complex, high-dimensional processes. The relevant question is not "are they real?", but "are they useful?".
For a similar discussion, see section "Realist intuitions impede progress in psychology" on page 4 here: https://psyarxiv.com/xj5uq
This is a conventional take which doesn't even attempt to grapple with the true complexities of the issue. The question is not whether current diagnoses are dimensional or categorical but rather whether they exist as meaningful separate diseases at all. And most data points suggest the answer is no, that the current system is a joke. A nosology useful to clinicians for efficient collegial
communicate but with no benefit - and often harm - to the patient.
Mentioning riches as an example of a dimensional variable is apt because it appears to be as useful a variable as depression in understanding human beings (I.e. not). That's because a) it's unreliable - most people's wealth fluctuates, often quite drastically, over the course of their lives. And of course b) it doesn't actually exist - money is a social construct.
When we talk about wealth what are we really attempting to get at? The variable that scientists - and particularly epidemiologists - have found most approximates that is socioeconomic status, which takes into account education, profession.and economic resources.
Depression more than likely is but one pillar of a larger, actually useful category - perhaps "internalizing disorder". Or a more accurate system might blow up the whole thing and not even use the heterogeneous DSM concept of depression at all.
Any system must deal seriously not only with the current in-vogue biological factors, but also must account for culture and time bound disorders and the drastically different rates of recovery of mental illnesses depending on where one is (i.e. westernized vs non westernized regions). When outcomes for schizophrenia are better in sub saharan africa than the US, something is wrong.
The taxon-dimension distinction seems important in discussions of talent; while talent varies greatly, "genius" as a truly separate category doesn't seem to exist. But many people, including Freddie there, conflate the question of whether genius is real with the question of whether talent is real.
An argument I've encountered that ADHD "really is" a specific medical condition, and not just a fancy name for below-average concentration skills, is that apparently when you give ADHD medicine to a person without ADHD, it actually makes them more jittery and less able to focus -- IOW the medicine has opposite effects on people with and without ADHD. Is that true, and does the argument prove what it sets out to prove?
This is a smart, entertaining post and I learned a lot. I was also reminded of something that I am reminded of a lot, which is that a lot of very smart people don’t really know much about addiction or at least don’t think enough about the power of words in its weird realm.
The exclamation point after “gambling addiction” I found particularly troubling, and a little snarky. Spend some time at a GA meeting and there will be no question in your mind that it’s an extremistanic not dimensional phenomenon. There is no meaningful line that connects my thrice a decade purchase of lottery tickets and gambling addiction.
My own experience and long-term observations of folks in recovery tell the same story about alcohol. There are cats and dogs and to an alcoholic alcohol is catnip, which lots of dogs might have now and again but does not lead them to massively destructive behavior. Also, sadly, while Jeff Bezos could buy enough (I suppose) ivory back scratchers to exit his personal extremistanic state, the addict has a one way ticket. Pickles cannot revert to cucumbers, etc.
Why do I bother to write this? Because, again in my experience, one of the defining features of addiction is that there is a constant internal dialogue that amounts to “abstinence is an overreaction, I can [gamble, drink, have the occasional benzo, etc.] just like everybody else.” It’s a disease that works to convince the afflicted that they are healthy. So in my view it is really dangerous to casually propagate dimensionality theories about addiction. Down that path lies an enormous chasm of pain for the addict and those around them.
"And look what isn’t on here – schizophrenia, which is just a really obvious separate taxon. You know, the condition where sometimes between ages 18 and 25 for men and ages 25 to 35 or so for women, over the course of a few weeks, seemingly normal people start getting extreme hallucinations and eventually devolve into a state where they often can’t live a normal life or even speak meaningful sentences? The taxometricians are saying ah, whatever, it happens to all of us, they’re just the people who it happens to more than average?"
2) "Has the flu" in terms of medical treatment isn't categorical based on "is infected by influenza virus or not" - it's an assessment of the dimensional response to being infected/exposed, which may run the whole way from "shrugged it off without even a sniffle or headache" to "death". As a reminder, this dimensional nature of the response to covid is part of what is making it so hard to manage.
3) If your chart is showing 'gender' as being more likely dimensional than categorial, then I think your chart has serious data issues. I myself would use that as a 'sniff test'.
Overall, I agree that looking at the human condition as a spectrum in most cases is very useful, and that it's probably best to intervene only so much as necessary to, as you say, allow a person to go on with their life in a productive and pursing happiness manner. Whether this is best money, or medical (procedure/chemical) treatment, or coaching into personal growth...there are lots of tools.
> this has another implication: stimulants shouldn’t be thought of as magic bullets that “cure” “ADHD” by fixing the underlying cause, in the same way that Tamiflu cures the flu by blocking flu viruses. They should be thought of as things that affect the underlying stew of variables that cause ADHD in some helpful way.
I don't think this is right. Conceptually, we should be able to say that, among the many dimensions along which we can measure a person, some of them are, by virtue of extreme or unfortunately-combined values, causing ADHD in that person.
And some of them are being modified by the expression of ADHD -- instead of being causes, they are among its effects.
The point of medication is to relieve a problem, such as an inability to see things. You can do that by removing a cause of the problem, and leaving the uncaused effect to disappear naturally (we can call this approach "laser eye surgery"), or you can do it by removing the effect while not doing anything about the cause. (We can call that approach "glasses".)
I had the impression that stimulants are viewed more as altering a symptom of ADHD than as affecting its causes.
I'm curious as to how this might work when there are more than two relevant categories: for example, colds also produce flu-ish symptoms, and the presence of another lump in the middle would move it away from the bimodal setup. I may have to look into this.
Minor nitpick, but you should probably make better figures than the hand-drawn ones on your website, even if the axes are completely 'symbolic' and only serve to prove a qualitative point. It'd look more professional.
Now do the same thing with traditional US denominations for race and you'll understand why it is seen rejected as a biological construct by the overwhelming majority of geneticists and population scientists.
> or you’ll get disillusioned and radicalized and start saying all psychiatry is fake.
This is about where I am.
My own experience with trying to get diagnoses and treatment for depression and ADHD for myself and for my kids has been super-frustrating. I know two young men who were diagnosed early with ADHD who later went through a smorgasbord of different diagnoses; one ended up as bi-polar and the other as autistic. My sense is of psychiatrists looking at clouds and saying "methinks it is like a weasel".
Add in the fact that you have to pay $1000s to each cloud-watcher before they can identify the right weasel and the fact that the next cloud-watcher will almost certainly tell you that the previous one was full of shit and it's hard to have confidence in the industry as a whole.
I wonder if, perhaps, this is why so many of us are eager to sign up with Scott's practice. At least Scott is aware of the uncertainty in the field. I fear the unwarranted certainty that so many doctors speak with most of all.
A lot of people have already identified the seemingly arbitrary deciding line between "dimensional" and "taxon" but it seems to me like there is a little question-begging in the other direction too.
Scott lists the criteria for various conditions, but the very existence of both the conditions and the criteria is, itself, down to some questionable statistical shenanigans.
Before asking whether depression is a categorical thing or a dimensional thing, shouldn't we be asking whether depression is a thing at all? And is it categorically a different thing than ADHD or bipolar? Or did they all just arise from more multi-variate cloud-staring?
I may have to use the "Jeff Bezos of absent-mindedness" line at some point. But beyond that, this is a helpful way of looking at something I've been thinking about on my own account for a while. (Though it would have helped if I hadn't read it on the Lorien website earlier.)
I wonder to what extent these 'objective' categories are just artifacts.
For example, imagine that angriness is a normal distribution, but psychiatrists don't actually test how angry someone is, but they consistently ask the question whether the person beats their partner. People who do are diagnosed with IED, those who do not, aren't.
What if anger is just a normal-distribution, where most people express this in societally acceptable ways, like posting angry screeds on Twitter? What if beating your partner has high societal and relationship costs, which result in high threshold costs. In other words, there is little cost to beating your partner more if you have done so before, but a high cost if you never did it before. This would then tend to result in an artificial dichotomy, that reflects external conditions, rather than the person being diagnosed.
The IED diagnoses would then seem meaningful, because it correctly identifies a 'real category' of people who beat their partner a lot (one of two clear bumps on a graph), but in reality, this is in largely part due to strong disapproval of partner-beating by society and most partners.
If you were to measure anger by number of angry tweets, you might find something close to a normal distribution, so then IED would suddenly no longer seem like a real category.
I've been interested in a related topic recently: suppose there's a condition that's truly taxonic, but we live in a world with diagnostic fads and Szaszian malingering. Would we be able to find which people have the real condition?
Think back to the 1990s when every other young girl with vaguely negative body image was tagged with an "anorexia" label. Might this cause anorexia to be seen as dimensional when it's actually taxonic?
I can sort of see why autism might be taxometric. My wife has autism. She's extremely high-functioning and I'm weird enough that an outsider wouldn't necessarily see a huge difference between us, but from the inside, there definitely is. There's basically a social processing module that she's missing. Stuff that to me is trivially obvious to the point that I have no clue how to explain it because everyone just gets it is completely baffling to her. She does a splendid job dealing with social situations despite this, but it's definitely a difference. Likewise with the sensory sensitivities, which I basically don't have.
(Of course, this is just one anecdote, and may not generalize.)
Beyond the distinction between categorical - continuous distribution, there is also a more fine grained scheme, to which you allude, in which the "continuous" category is split into two sub-types: those with long and those with short tails. Short tailed distributions (e.g. the normal or exponential distribution) have a well defined scale (the standard deviation for the normal distribution) which allows you to say whether an individual is average or exceptional, even though the distribution is continuous. Height is an example. For long tails, i.e. power laws, you don't have such a measuring stick - they are "scale free". For example, wealth is power-law distributed: for any 10 people of net worth $x, there is one person of net worth $10x - this makes it much harder to define "exceptional cases" than in the case of normally distributed attributes. Here's my question: which "continuous" psychiatric conditions are long, and which are short tailed?
One place that categorical thinking (about ADHD) has real-world consequences is in academic accommodations. If a doctor signs off that you that you pass the bar for ADHD, you get 50% extra time on all exams, no questions asked. If you're not quite at the bar, nothing. As a professor, I've responded to this arbitrariness by trying to make exams less time-sensitive, but not with perfect success.
If ADHD is dimensional and prescribing medication is a matter of risks and benefits, why do we only prescribe it either EVERY DAY or not at all? Some people would like to have it available to take once a week, for, say, their most tedious task. What are the very serious risks that outweigh that need, assuming they have no history of addiction or heart problems?
Scott, I'm reading the brief primer, but not seeing addressed why they choose these methods over normal cluster analysis, which is typically what you'd do when trying to decide whether some category is "real" or not statistically. Is it just because they have low-dimensional data? That doesn't seem like a good reason, since that makes cluster analysis even more interpretable if you can actually plot the clusters. "Within group variance is less than between group variance" at least seems more principled and intuitive than "CFFI is greater than 0.5."
Minor nitpick, Scott uses Taleb's terminology incorrectly. Bezos doesn't live in Extremistan. Wealth is in Extremistan, while height is in Mediocristan.
Mediocristan is where you find all the nice normal distributions which look like Bell Curves. Extremistan is where you find distributions with really long tails (power law type stuff).
Another good analogy with height is that, theoretically, there's nothing _wrong_ with a short man; but, practically, he's generally going to struggle in life in our current social milieu. In the same way, there may be nothing inherently wrong with someone who struggles concentrating (maybe they're just not interested in the subjects of the current social milieu), but they'll struggle "fitting in". This might be a more tempered way of looking at the often strange spiritual implications of psychology and psychiatry rather than going full Szasz.
There's a classical logic reason to deny that there is a difference between categorical traits and dimensional ones. Take ADHD for example. If it is dimensional, then there will be cases where you want to say of someone that they are not ADHD, but they are not not ADHD either. They sit in the vague borderlands, with some ADHD symptoms, but not the full suite, or they don't have them very severely. Well, according to the rule of double negation, "being not not ADHD" = "being ADHD". From the preceding reasoning, it follows that the person is both ADHD and not ADHD, which is a contradiction. Since this reasoning will work with all dimensional/non-taxonic traits ("rich", "tall", etc.), it follows that no trait is truly dimensional. They are all categorical/taxonic. Now, it may be a permanent limitation on human knowledge that we can never know the precise boundary between being ADHD and not being ADHD, but there must be such a boundary. If you think, hey, let's just give up classical logic and say there are propositions with truth-value gaps, that leads to problems too. One of these is how to fix the boundary between propositions with truth values and those without without just replicating the whole vagueness problem all over again. Plus it seems, as my Dad would say, like too much sugar for a nickel.
From what I gather the RDoc framework (https://pubmed.ncbi.nlm.nih.gov/26845519/) is an attempt to steer psychiatric/psychological research in the direction of more dimensionality and away from categorization. I would think the advent of digital phenotyping and behavior tracking afforded by apps/bands would enable greater granularity of behavior/symptoms in this regard
A problem here is that many of the drugs that psychiatrists prescribe are controlled substances, so you need clear guidelines to avoid prosecution. Of course there's an obvious solution - stop controlling substances - but in the actual world.
The end reminds me of a conversation I had with Thomas Szasz many years ago. I asked how mental illness could be only a metaphorical "illness" if people who are not functional become functional by taking psychiatric medications. He said, "Do you know anyone who can't function without cigarettes? Would you say they suffer from a nicotine deficiency?" A similarly pragmatic POV.
I don't think you are properly following out the logic of your argument. You start with flu because we have an unambiguous test, infection with the influenza organism, that shows it to be categorical, you either do have it or don't. If the various statistical tests show it instead to be a continuum, that's evidence not that it is a continuum but that they are insufficiently powerful to tell. That's a straightforward Bayesian argument, starting with a very strong prior, itself based on direct evidence, that people either do have flu or don't.
In the case of gender we again have a test, indeed two tests with almost perfect overlap, genetics and structure, that divides the population into two categories that cover almost everyone. If the statistical tests that were used fail to identify it as category rather than continuum, that's evidence not that it's a continuum but that the tests are sometimes insufficiently powerful to tell. It follows from that that the failure of the tests to identify schizophrenia as categorical doesn't mean it isn't categorical. Indeed, if you have other strong evidence that it is, it gives you another strong reason to lower your confidence that the tests give the right answer.
One could, I suppose, try to avoid that conclusion by saying that what they are measuring is gender, not sex, and it isn't defined by genetics or structure. That's like saying that what you are looking for is not infection with the influenza virus but symptoms of flu. But in both cases, that amounts to throwing away strong evidence — the existence of an unambiguous categorical difference that correlates with the symptoms being observed. How, after all, do you decide what the symptoms of male or female gender are, other than by observing their correlation with other symptoms of the same gender? So how can it make sense to throw out XX/XY genetics, which correlate pretty strongly with m/f gender?
Not sure if the figure style has a particular pedogogical purpose, but here are some less pixelated versions you are free to use, Scott (and I can fiddle with them further if they don't match the existing ones properly).
Interesting analysis. In order to make it more comprehensive, I'd add the concept of a threshold effect. We see these different types of systems, where reaching a certain threshold in a continuous variable system has a dramatically different effect than any value below that threshold.
The action potential in a neuron is a classic example, where a slightly lower intracellular voltage doesn't trigger the neuron to fire, but then once the threshold is reached a complex series of events unfolds. Another example is phosphorylation of the rb protein, which requires a certain threshold to be reached to trigger the cell cycle. Below-threshold rb phosphorylation gives you no cell division. Once the threshold is reached, the cell will divide.
The gender thing is cute, but what does it actually mean? Some people are saying that if the test doesn't show gender as taxonic then it's obviously bunk. That's certainly true if they've done the test properly, with at least one factor indicating biological sex. But what if they haven't (as seems to be the case)? Does that mean they're measuring some kind of "gender minus sex"? Should we expect that to be taxonic?
it's interesting to think about a difference between absolute and relative categories, sort of complicates the relationship between the dimensional and the categorical.
rabbits and humans are categorically distinct in relation to one another and according to specific criteria. but if you abstract either term from the relation, the categorical definition says virtually nothing about the thing it defines (ie. a certain weight or what have you).. its Diogenes with the chicken..
relative categories are pragmatic constructions which are highly unstable, misinterpretable because of their contingency, hence the aforementioned example. the problem is that, qua the terms of the definition, the relatively defined category can belong to a larger set of things that share the qualities by which it is defined (again, size, shape, weight, whatever) -- whence comes the possibility of some equivocal identification. the relative category can inadvertently create a new dimensionality consisting of the vague objects falling under it (especially if empirical conditions are not ideal).
an absolute category, by contrast, would need to be defined according to terms which do not and could not make something falling under it a member of some more-general set (of arbitrarily non-rabbitish things, or what have you).
and with an absolute or singular definition it need not be a matter of either/or in terms of diagnosis, or even both/and.. it's a matter of necessarily belonging to an absolute category to a contingent degree (which of course could still be 100%). so i suppose the absolute category diagnoses according to singular intensity rather than a bi- or multi-valent belonging in this or that category.
> My guess is most professionals, and an overwhelming majority of laymen, are actually confused on this point, and this messes them up in a lot of ways.
Thank you for writing this. This had annoyed me a lot, back in the school.
Many professors seemed to confuse the map with the territory, although they might have used confusing languages just for convenience' sake.
Regardless, such confusion have propagated to fellow students, and the vicious cycle continues.
Many physical diseases involving complex diagnostic criteria seemed to share similar proneness for confusion.
This kind of reasoning should be taught at med schools.
> I’ll grant them pedophilia – it really does seem like people either are or aren’t pedophiles and there’s something weird and specific going on there.
Total aside to your main point, but (speaking as one myself) I'm not sure if this is true. Certainly I know people with varying levels of attraction, people who are attracted to both children and adults but more to one or the other, people who only realized later in life that they are attracted to children, etc. I think this one could easily turn out to be dimensional.
In other notes, thank you for not conflating pedophiles with child molesters.
If "taxometrics" gives the wrong answer for schizophrenia and gender, shouldn't we just ignore it?
--
You suggest that for "dimensional" traits like wealth or ADHD-ness, we shouldn't expect a specific underlying cause. But extreme values of a dimensional trait seem like they might well have an underlying cause. If you ask why Jeff Bezos is so wealthy, the answer isn't, "well, you know, who can say, it's just a confluence of a bunch of small factors: he worked a bit harder than normal, his parents were a bit wealthier than normal, he saved a bit more scrupulously than most, etc., etc." No, there's one big reason, namely that he founded Amazon. Similarly, I think it's true that people at the true utter extremes of the height distribution have a single specific underlying disorder that causes them to be really really tall or short. It seems like an empirical question whether the extreme values of a dimensional trait have a specific underlying cause or not.
Machine learning takes a bunch of items with labels ("rich", "not rich") and attributes ("owns more than 5 cars", "reads the New Yorker") and then comes up with a formula ("algorithm") for deciding whether an unknown item fits a label. The formula typically has no explanation (it might have a confidence value for a decision), but there are ways of giving a partial explanation of how it works (e.g., fitting a decision tree to the formula).
The DSM criteria sound like a simplified version of machine learning. And, as we know, machine learning used without care can have all kinds of problems; for example, built-in bias because wrong attributes were chosen or the attributes are correlated with other things that were ignored, such as gender or ethnicity.
Interesting but spoilt by muddle headed thinking about 'schizophrenia'. There is a vast literature stretching over thirty years, most of which has nothing to do with taxometrics, which supports the continuum model. Moreover, something that has been overlooked is that the diagnosis groups together a collection of symptoms which researchers have increasingly recognised may have nothing to do with each other, so its possible that some are taxonic (I have my suspicions about hallucinations) but some are not. I carried out a very large taxometric study of paranoid symptoms and the findings wholey supported a continuum model: Elahi, A., Perez Algorta, G., Varese, F., McIntyre, J. C., & Bentall, R. P. (2017). Do paranoid delusions exist on a continuum with subclinical paranoia? A multi-method taxometric study. Schizophrenia Research, 190, 77-81. doi:10.1016/j.schres.2017.03.022
I actually tried making that argument about schizophrenia to Greg Cochran, and he replied that GWAS studies really do support the notion that it's just the extreme end of broad variation, like mental retardation:
I really do like this piece in general. I am a physician and I also am an individual with high functioning ASD. My problem is with the final paragraph. Language and the meanings assigned to words is a highly variable thing, but as someone with ASD, my "feeling" is that terminology such as "a person with ASD" is much less binary/dualistic than "ASD person." Describing me as "a person with ASD" at least makes me feel like I am on a continuum that includes everybody else. Not that the way I as an individual feel about it is all that important. Just introducing a perspective.
It's not really clear to me that what taxometrics is measuring is well-defined. We could of course *define* categorical properties to be ones that pass MAXCOV or some other test, but if we reduce categoricity to that, it's not clear why we should care about it. So what does it mean to "neatly, objectively separate into different groups"?
One answer would be "conceptually dividing things into these discrete categories is the right way to think about them". Spelled out like that it should be pretty clear that this can depend on a lot of things other than the statistical properties of the dataset, but I suspect that this interpretation is a big part of why taxometrics feel like they're telling us something important.
Another possibility would be "the process which generates the observed data significantly involves a discrete random variable". I see two problems with this one. First is that it's not really testable: we can get similar distributions by means of very different underlying processes. For example, if we have a variable X distributed according to a standard normal distribution, then arctan(100X) will be highly bimodal, and if our observables are three different noisy measures of arctan(100X), they'll pass the MAXCOV test, and I'm pretty sure that they or something very similar would pass any other statistical test of categoricity as well*, even though no discrete variables were involved in the process.
The other problem with characterizing categoricity as "the process which generates the observed data significantly involves a discrete random variable" is that the physical processes which actually produce the data are not the same sort of thing as the mathematical models we use to think about them. With a mathematically defined process for generating a random variable, we can say "look, right there, on line 3, there's a discrete random variable", but the physical process isn't always going to have some specific location of the variable that we could point to. If there's a specific gene or something, that's pretty clear-cut, but often there isn't. And when there isn't, whether or not there's a discrete variable involved is a property of the *model*, not the world, and we might get different answers depending on what model we use and how granular it is. For example, you talked about "having the flu" as a binary categorical property, and certainly under many situations it's appropriate to model it as a discrete random variable. But a more detailed model might look at viral count and antibody count as two real variables which change continuously over time. Or something like that; I know very little about immunology. But I'm pretty sure you can't tie it to something as physically unambiguous as "are there more than zero viruses in your body" because someone who fully recovered from the flu and has antibodies shouldn't be classified as infected if they inhale a few viruses.
Maybe these problems have been resolved somewhere in the taxometric literature, but I'm not seeing them addressed here or in the linked primer, and they seem pretty fundamental.
*because arctan(cX) converges in distribution to a coin flip between pi/2 and -pi/2 as c goes to infinity, so statistical tests shouldn't be able to distinguish it from the discrete case for large enough c
Time for the meta question! Is the distinction between taxonic and non-taxonic traits itself taxonic or non-taxonic? What would the CCFI distribution look like for lots of different traits?
There's a tasty and tempting irony in taxometrics. The field says: "Turns out many of these supposed disorders are actually traits that are non-taxonic." Then one is tempted to say "I'm going to care deeply about which traits end up in the binary bins of taxonic or non-taxonic!"
This kind of side-steps the really important question of "when and why do seemingly smooth distributions give rise to important and distinct taxa?"
The answer to this question is nearly always, "there are self-reinforcing effects going on and the taxa are divided by the point in the distribution where the variable will grow over time to where it will shrink".
So your income graph may look perfectly smooth, but there are nevertheless two distinct taxa in people who are living hand-to-mouth, who will see stagnating or declining incomes over time and people who are investing surplus income in future income generation, who will see income growing over time.
Similarly, disease infectiousness may be a continuous spectrum, but there is a big division at r = 1 between diseases that will die out after a few people have been sick and those that will become pandemic.
It feels like many (if not most) mental illnesses have this kind of flavor. The distinction between a person who is depressed and one who is merely miserable is not really how miserable they feel. The distinction is whether that misery compounds itself over time until they cannot get out of bed (and are then miserable about the fact they have not gotten out of bed) or whether it decays naturally as they switch to happier activities to make themselves feel better. Treatments, too, tend to focus on breaking cycles, not really shifting the set point.
It's ironic that this is missed in the article, because your example of the most clear taxon in Psychology is mostly defined by its instability. If your schizophrenics had been unable to speak meaningful sentences since childhood, would you be quite so convinced that they were not just at the long tail of the language development spectrum?
By this distinction, depression, addiction, anxiety, schizophrenia, bipolar disorder, etc. would likely be distinct taxa categorized by the point where they become unstable. ADHD and autism would probably be somewhat arbitrary lines drawn on a distribution.
In short, taxometrics doesn't seem well-validated as a statistical technique. Unlike most statistical techniques one sees in practice, it neither has backing from strong mathematical theory, nor from the "real-world stress test" of a large number of people independently using it on different problems. That doesn't mean it *couldn't* work, but feels like it has a closer family resemblance to "algorithms I personally invent for personal projects that seem to work OK" than "anything you might see used by eg Andrew Gelman in a paper."
Separately, I'm unsure it makes sense to apply it in psychiatric patient samples, even if the idea makes sense in some other contexts. From the post:
> After all that, you still have the data you have, which in psychiatric contexts will be sampled from the general population in a very non-uniform way.
> The idea makes sense in an idealized world where your research sample is drawn randomly from the population of All Possible Humans. But psychiatric samples are very unlike that. You can try to remedy that by introducing some control people from the general population, but then you’re introducing a two-category structure into the data (controls vs. patients)!
When I was in college I was depressed so I went to the doctor; they had me fill out a questionnaire and based on my answers decided I wasn't really depressed and basically told me to get over it. One of many experiences that really put me off seeking medical help :(
Could someone please do a meta-analysis of Comparison Curve Fit Index studies to determine if 'taxonicity' and 'dimensionality' are separate categories or just different points on a continuous spectrum?
First reaction; YES! So much what I’ve been thinking and teaching and explaining to patients for years.
Then so no no no; eating disorders and substance abuse as categorical? This mistake arises because psychiatrists (and most MDs and most hospital-based psychologists, the ones who do most of the research) only meet the most severe cases, the ones who haven’t managed to stop by themselves or with the help/push of concerned friends, families, employers, AA etc. (BTW, for those most severe cases, there’s very high co-morbidity with early trauma and/or personality disorders. Both also dimensional.)
I’m betting the researchers doing the studies that led to those results in the meta-analysis only included people severe enough to seek/ be brought to professional treatment. Plus we often don’t ask about these issues, unless it’s part of the presenting problem.
Scott, go find the research on how many people have ‘sub-clinical’ ED, alcohol abuse, etc. How many people are ‘overly careful’ about what they eat or how trim they stay, or who binge, but not to the point their teeth are rotting. How many have a year or two or three of drinking or drugging enough to start wreaking some havoc. How many manage to pull themselves back from the brink. To ‘fix themselves’, alone or with the help of a school counsellor, an EAP provider..... to start a virtuous cycle that leads them to a much better place. Without ever talking to their GP about it or seeing a psychiatrist.
> I’ll grant them pedophilia – it really does seem like people either are or aren’t pedophiles and there’s something weird and specific going on there.
As a carrier of this trait, I'm conflicted about this myself.
On one hand, I'm definitely able to fall in love, have sex, and form emotional attachment with adults. I've had a reasonably full love life but none of my lovers ever were under the age of consent. Right now I am happily married with kids (and yes, my wife knows all about me).
On the other hand, I feel definitely binary about my sexual attractions. There are many adult females that are about as attractive to me as a crocodile. For an attraction to be possible, I need to see some hard-to-define childishness in demeanor that doesn't depend much on age but that most adult women (and even some children) totally lack.
So, if I define myself as someone sexually attracted to children (i.e. a binary category with a legal cutoff), then I am on a spectrum: I can experience attraction on both sides of the cutoff. But if I define myself as one attracted a certain personality/appearance trait, which is itself a spectrum, then I am binary: if you don't have this trait I have zero interest.
I believe when people talk about autism spectrum, they are talking about a different thing than taxonometrics. What they are saying is that there are multiple ways to be autistic, and all autistic vary a lot on their presentation of autism. They are trying to dispel the popular image of autism as Rain Man, or Sheldon Cooper. This is entirely compatible with autism being its own taxa, with some fuzzyness in the border, just like some people may have the flu, but no or few simptons, or not have the flu but have some flu-like simptons.
Anedoctly, I've seen activists saying, precisely this, that autism is a spectrum, but some people are clearly in the spectrum and some are clearly out.
It sort of seems like the psychiatric profession should move away from the notion of 'diagnosis'. Diagnosis is fundamentally a verb for categorizing. Perhaps it'd be more appropriate to think of things in terms of scores. Patients take a depression test, which gives them a 0-100% depression score, and then you can define ranges of that score that indicate the value of clinical intervention. Functionally that's sort of already what's happening, but re-framing it in this way seems like it might be a better epistemic map projection.
Reading this as a physician, I feel suddenly struck by how much of medicine may actually be comprised of things-more-like-height.
Hypertension, as mentioned, is a classic example, but I feel like actually the more closely you look at various health conditions, the less they independently exist as discrete things in the world.
A few examples;
Diabetes (type 2) - basically like high blood pressure. we use arbitrary cut-offs of certain Hba1c measures. The fact that ‘pre-diabetes’ is a diagnosis in itself is telling.
Dementia - very much a spectrum, we use scoring systems, ‘mild cognitive impairment’ being a diagnosis ‘before’ you reach a score “bad enough” to have dementia.
Heart attack - on a spectrum with stable and unstable angina. Yes we have various measures to determine how likely it is that part of the heart muscle is damaged (ECG changes, blood test markers, again arbitrary cut-offs) - but part of your heart muscle dying is not really something that does or doesn’t happen. Suppose 5 myocytes died because your heart was under strain temporarily? - not a heart attack.
Even infections - it’s not really as simple as ‘is this virus or bacteria present in X body system?’ - if you culture somebody’s nasal cavity, or poo, or skin, you will often grow bacteria, and these are not infections. One of the key defining criteria of a pneumonia is whether or not you can see an area of infection on a chest x-ray - again, this is obviously not yes or no.
I think extremely few things really exist in a discrete way (aside from genetic disorders controlled by a single gene, which seem to be an unusual case). We make up categories to describe continuous variables in a way which is most helpful for deciding when to use different treatments. Seems equally true to me for most mental and physical health conditions.
I feel like this missed a good opportunity to make another supporting point: Jeff Bezos only makes $81k/yr salary. (Everything else he has was likely fundraised by sales of AMZN stock to people looking to own small fractions of Amazon... probably 9-figures some years, but probably near 0 some other years. And it also would include a lot of people who are technically broke. Point being, categories are difficult, and easy to manipulate.)
Wonderful text, thanks a lot. There's even a new section in the DSM in which they want to make numbers out of all these complicated diagnoses, I think it's called "emerging theories and models". My colleagues like this a lot. Nice terminology for something that is probably mostly driven by convenience.
In the old days, there were complex psychiatric diagnoses, categorical, and the psychiatrist spent quite some time finding out the problem. The problem: you cannot run the nice parametric statistics, you need proper numbers.
So, rating scales were developed, Hamilton rating scale for depression, anxiety, and so on. The ratings were made by psychiatrists, and they needed to take classes to learn how to do it properly. The problem: it takes so much time. Imagine a randomized study with 200 participants, nobody has time to perform 200 interviews.
Then, more and more questionnaires were invented (the nice term for this is: "patient-reported outcomes"). Beck depression inventory, GAD-7 is also a questionnaire. When I ask my students in class, how do you want to measure X, they almost surely respond: "With a questionnaire". Using questionnaires, the researcher can save a lot of time because the patient is diagnosing himself or herself. More than 20 on BDI -> moderate depression (I don't remember the exact figure). No psychiatrist needed for this diagnosis. And since clinical studies are often collaborative, and every collaborater has a pet scale, many questionnaires were put on the table of the patient. The problem: piles of paper forms that no one wants to type into the computer.
Then, the internet came and everyone bought a smartphone. Nowadays, the patient receive a link to an ever-growing pile of online questionnaires, which you cannot escape until you have clicked on the last of, say, 500 questions. I have seen studies in which, according to my calculations, patients spent hours of with filling out online forms. One of the phd students involved in such projects told me that many patients just put a cross in the middle category because the online forms are way too long. There's also a nice orwellian word for this nightmare: "measurement-based care".
One more funny thing is that there are scientific guidelines from the regulatory authorities in medicine (EMA, FDA) that actually limit the use of all this. E.g., the need for a "primary outcome" in a clinical study, which prevents you from collecting 20 variables and choosing the one that shows most pretty results. E.g., the EMA guideline for patient-reported outcomes in oncology that says, at most 20 min for questionnaire. E.g., the EMA guideline for depression that tells the researchers to do blinded assessment using rating scales. I am digressing, sorry.
There is actually a quite sophisticated "representational theory of measurement" by Krantz, Suppes, Tversky, Luce. The theory was developed to enable measurement even in the absence of concatenation operations such as putting rods together so that the length adds up, or putting two objects in the same pane so that the weight is the sum of the two. In the social sciences, such concatenations don't exist, and Krantz et al.'s theory can be very helpful here to establish an interval scale. But the theory is complicated, so no one makes use of it, neither the proponents nor the opponents of the dimensional approach. Instead, the meaning of a scale is mostly established by its name. If the name of a scale is Novaco Anger Scale, it obviously measures anger. Add a few correlations with other "instruments", and claim that your scale is "valid".
One more funny thing: I think the whole measurement hype in psychology started with intelligence; and early intelligence test were made for diagnosing mental abilities in children. More recently (1980ies onwards), Falmagne and Doignon developed so-called knowledge structures for qualitative assessment of skills, so dancing plus singing is not equal to 2. This stuff is more and more used in e-learning systems like ALEKS that some of our children have to do homework with. In education, they noticed that one cannot make numbers out of everything.
Once I asked my psychotherapist about my diagnosis. His response: "Why would you want to know? I only use it for the health insurance."
I am fervently interested in whether or not you (and others here) have opinions on so-called "positive psychology". For the uninitiated, positive psych is a comparatively recent (1960s or so) effort to use science to understand mental wellness instead of mental illness. In the nutshell explanation I usually use, it's the difference between fixing a blown knee versus training to break the 4-minute mile.
Most of the pos psych stuff promoted (and I do mean promoted) to layman is suspiciously indistinguishable from self-help snake-oil (keep a gratitude journal! don't let failures get you down! develop a growth mindset!), with the caveat that it is, by and large, snake-oil that appears to have, on average, replicable results - per its own (questionable IMO) measures - when people buy into it.
A lot of it is sold (and I do mean sold) as preventative medicine, but I think it fits great with my 10-minutes-old understanding of this taxometric distinction. If most people are not so much afflicted by a mental disease as they are uncomfortable or disabled due to their position out on the tail vis-à-vis anxiety levels, it makes sense that sometimes mental health-oriented boosters /could/ help just as much as, or more than, mental disease-oriented corrections.
The evidence is chaotic soup served with statistical smoke-and-mirrors, but with my methodological background, I've blamed it mostly on early commitment to suboptimal (proxy) measures - not so much where you draw the line between tall/short, but whether or not the yardstick is remotely objective or validly measuring the phenomenon you say it is. Now I'm going to be turning over in my head also whether or not it's a case of: "My guess is most professionals, and an overwhelming majority of laymen, are actually confused on this point, and this messes them up in a lot of ways."
Homelessness is probably categorical enough that "person experiencing homelessness" may be more appropriate than "homeless person" if you care just about sending the right signal.
(Since it takes more bits to explain disagreement than to indicate agreement, and positive reinforcement is useful when true, please take my disagreement here to indicate that any portion I _haven't_ disagreed with is a portion that I agree with.)
You say that "I think a lot of people still want psychiatry to deliver the single [cause of a given psychiatric phenomenon]. It’s not going to be able to do that." I don't think the fact that a given phenomenon is dimensional means that this is impossible; I would read it as meaning that the root cause must also be dimensional. To massively oversimplify the brain, let's say that all other things being equal an increase of X% in the average activity of one's Anxiety Gland produces an X% increase in one's reported anxiety levels, and that all medications which affect anxiety are ultimately things that have an impact of reducing Anxiety Gland activity by some amount. (The stuff that can impact the activity levels of the Anxiety Gland may be complex and multivariate, but I don't think this is different from categorical issues - if there are developmental failsafes so that N out of M mostly independent processes have to fail in order to produce a Categorically Abnormal Anxiety Gland, otherwise the development of a Normal Anxiety Gland is an attractor state, I'd still call Categorical Abnormality of the Anxiety Gland the "single cause", especially if it was what could be affected by interventions.) This would still be a single cause; it would just be a single dimensional cause.
You also say that "If most mental disorders are dimensional variation rather than taxa, that kind of makes the DSM look pretty silly, doesn’t it?...All of this is predicated on the idea that [there are] specific thing[s] called [disorders] that you either do or don’t have." I think another way to regard the DSM, or at least what an idealized DSM could be, is a manual of best practices for tradeoffs which are codified as acceptable at arbitrary cutoffs, much like your hypothetical Sanders tax plan is an arbitrary cutoff made for practical purposes. If a patient has more than X% excess Quality, then the profession regards it as reasonable to experiment with certain interventions which fall under the banner of "Interventions for Excess Quality Disorder", but below that cutoff they think it's not worth the risk.
In this case, it would be good and necessary to renegotiate the definitions of Excess Quality Disorder as new information came up, even if "Excess Quality Disorder" is an arbitrary cutoff. Let's say that new research comes in indicating that Intervention A for mild Excess Quality Disorder is more harmful than previously expected; in that case we might move the boundaries of Excess Quality Disorder to reflect that we want to be more careful with it. Or maybe we discover that while our cutoff was X%, people with (0.8 * X)% excess Quality have a reduction in quality of life greater than we previously thought, so we move the boundaries in the other direction. In this case, "Excess Quality Disorder" is perhaps an inferior category to "Eligible for Intervention {A,B...Z}", but may be worth using for convenience, communication, or research.
Autism, like height and other dimensional traits are influenced by 100s or more genes, each with a small influence up or down. So traits that are influenced by relatively few genes a larger amount each should be more categorical, e.g. schizophrenia (if my impression of schizophrenia is correct).
One challenge I've personally experienced with definitions for psychiatric diagnoses that are based on multiple dimensions is that people tend to attribute all the (categorical) symptoms to people with only enough of them to fit the definition.
My experience is with autism, or at least autism-spectrum disorders. People are diagnosed with autism based on possessing characteristics such as problems with social interaction and communication, and repetitive behaviour or restricted interests (think: Rain Man). My son was diagnosed as "on the spectrum" at a young age. This meant that teachers and other professionals would assume he had all the most common characteristics, whereas in fact he had only a few, and this meant their approach was often not appropriate at all.
This happens all over the place, of course: it's the problem of "natural kinds". All instances of Democrats, ice creams, and churches are not alike in their categories on all dimensions. Politics is often a difficult thing not least because it forces hard demarcations where in reality they're fuzzy (I vote for party X because it's X or Y, not because I believe everything that X profess to believe).
We cannot carve the world at its joints, because they're in our mind rather than in the world. The same is, perhaps unsurprisingly, true of the mind itself.
Thanks for the fine exegesis. I have practiced psychiatry for many a year. DSM III came out when I was in medical school. It sought to make psychiatric diagnosis, which was vague and overly influence by psychoanalytic theory, more objective, but using description and taking away theory. It was felt essential to make psychiatry a legitimately medical specialty, and at the time it was hailed as a triumph. Alas, they threw out the baby with the bathwater. All of the useful tools of personal history, development, personality and temperament were junked. But whatever has been gained from categories makes it harder to treat patients as individuals. (Just ask the insurance company flunkies who rule over the prior authorization domain, and decide how many sessions and which medications are "medically necessary.") DSM IV and V made matters even worse. Not only are the debates about where the cutoffs are for a disorder inane, political considerations have tainted and even overwhelmed whatever medical legitimacy the descriptive categories might have had.
I have long been reconciled that my profession will never be as objective as brain surgery or cardiology. No matter how you dress it up, psychiatry is as much art as science, and I like it that way. As much I look forward to advances of neuroscience and pharmacology, the DSM ain't science, and Big Data tells a lot of Big Fibs.
I appreciated the argument on person-centred language. I've not seen this argument brought up in this way vs "recovery oriented" language. From a medical model it probably makes more sense, particularly when charting, to not use person-centered language.
I’m surprised to hear that narcissistic personality disorder is generally considered by psychiatrists to be categorical rather than dimensional. I’d love to learn more here.
It intuitively seems like people vary on a spectrum in terms of how selfish, egotistical, etc. they are. A lot of people certainly seem to have subclinical levels of narcissism.
Is it that narcissistic personality disorder is something radically different from the personality trait normally referred to as narcissism?
I liked the mention of extremistan. It made me wonder—
*Are* the distributions of psychological variables (e.g. anxiety) fat-tailed (e.g. power law distributed instead of bell curves)? If so, that would maybe be relevant to whether and why people underestimate how anxious GAD people can be, etc.
This is really interesting, but as someone who has ADHD, the idea that ADHD specifically is dimensional seems to directly contradict my experiences. The "symptoms" that get ADHD noticed and diagnosed tend to look like just having an unusual amount of trouble paying attention, the way Jeff Bezos has an unusual amount of money, but there's a slew of other things that have no reason to correlate with ADHD the way they do unless we consider them side effects of an underlying categorical difference.
If I just have an unusual amount of trouble concentrating, why does Adderall make me sleepy? Why does fidgeting help me think? Why are boredom and the perception of failure almost physically painful, in ways that confuses my non-ADHD friends when I describe them? Why do I share these traits with my roommate, who doesn't have difficulty concentrating at all, but has enough symptoms of hyperactive-type ADHD to also get a diagnosis? Why is there enough of a correlation between hyperactivity and inattention that the combined type is even recognized?
There's no reason inattention would correlate with these other traits without an underlying cause. A non-ADHD person who had a short attention span probably wouldn't also fall asleep after three cups of coffee or cry because they got mildly criticized at work.
I suspect some of the confounding findings have to do with the selection of variables that are included in the analysis and how they distribute through the general population in realms unrelated to the ‘condition’ being investigated. I don’t have the time to think through the implications, but further study of the efficacy of the statistical method for separating behavioral disorders is definitely warranted.
tl;dr: A secret variable does exist, but you can't read it. [we might be in the future, though]
You can't avoid false negatives, but in many cases, you can definitely be certain that you don't have a false positive.
[according to Dr. Russel Barkley, who is a famous ADHD researcher]
That is not quite right. There actually is such a variable and it can be 1.
ADHD is brain damage in five specific affected brain regions, that predictably results in a 30% delay of executive functions for your age. In 2/3rds of the cases, this is genetically determined.
And there are also acquired cases, where it's known that strep-bacteria causes an autoimmune overreaction that causes that brain damage.
And yes, ADHD is highly comorbid with other things. And yes, SCT might be another attention disorder, that's separate from it, is not as treatable (yet) and less well understood.
[and having SCT implies 50% chance of having ADHD as well]
And yes, other kinds of brain damage also can lead you to present with ADHD symptoms.
And all of the above is not universally known or accepted in psychiatric practice.
But ADHD is actually a taxon.
The variable exists and you might not be able to reliably say that someone is a 0.
[unless you find some other disease or environmental cause...... like a kid playing Fortnite all day in school, which is an epidemic in its own right]
After all, the space between the variable=1 population bump and the majority of the bell to its left is not empty.
BUT...... you can and should be quite certain sometimes, that you do have "the real deal" in front of you, by looking at the parents or grandparents and seeing them have it as well.
Also seeing the "paradoxical stimulant effect" (stimulants usually don't make you calmer) would provides evidence for a "1", as well.
source:
excellent playlist "30 essential ideas you should know about ADHD" by Dr. Barkley on recent research into ADHD:
Concern about the height example: what if there is a height gene H which can be 0/1. Suppose both H=0 and H=1 lead to unimodal (Gaussian?) phenotype height distributions. If you have a mixture model of this reality then the overall population's height distribution could be unimodal (Gaussian) as well? How can you distinguish this reality from the reality where gene H doesn't exist?
Some of this terminology seems useful for talking about consciousness (which I suppose is a sort of psychiatric condition); there’s obviously a useful distinction between humans and algae, but trying to pin down an exact dividing line proves difficult, especially when you entertain thought experiments like removing one neuron at a time from a brain. Lots of debates about e.g. animal consciousness seem possible to phrase in terms of the shape of this graph. See also the Sorites Paradox.
Very minor nitpick: hypertension guidelines have changed, and AHA defines stage 1 hypertension as SBP ≥130‐139 mm Hg or DBP ≥80‐89 mm Hg
Scott would it be oversimplifying to say mental illness is on a spectrum and diagnoses are at best a way to communicate the cluster of symptoms?
This was such a well-written article. I wish all doctors were this good at math. It would really help them make better decisions for their patients.
"My guess is most professionals, and an overwhelming majority of laymen, are actually confused on this point, and this messes them up in a lot of ways."--how much of this is just reification as a crutch for computational efficiency given limited time/energy/attention?
The thought processes you are moving away from are simpler than the ones you are moving toward, and docs are pressed for time. Likewise, laypeople are mostly probably trying to make decisions about their lives with framings like 'I think that man may be a narcissist so I probably won't go out on a date with him'
The reifications have costs but also benefits and I'm glad you're digging deeper in your practice and inquiry. I worry somewhat about unintended consequences when categorical terms for dimensional characteristics escape the clinical setting and make their way into the ambient construct stew
As a side note, there are common categories in use for wealth like HNW, VHNW, UHNW, but they certainly don't capture the upper-end variation around someone like Bezos (for whom even 'billionaire' is off by orders of magnitude...)
The next time someone asks me about my mental health I’ll feel tempted to tell them that I’m “the Jeff Bezos of absentmindedness”. For a while now I’ve intuitively been moving away from the taxonometric definitions of mental illness so it is nice to read something that really goes into the math of it. Even in the cases where there is a definitive “reason” for the mental illnesses listed in the study, like the flu virus for the flu, it usually seems like it’s either some kind of traumatic brain injury (difficult to reverse for now) or something genetic (impossible to reverse fo now) so holding out for a magic bullet is implausible... It’s all symptom management. But symptom management is often what can help someone get over the arbitrary-feeling “line” between “Person doing poorly” and “Person doing well”!
Thank you for this excellent article. I found the concept of a "resident of Extremistan" to be an excellent phrase for describing someone like myself. I fall into the 2% or less in seven major categories, including autism spectrum and bipolar disorder.
I just signed up here as a founding member after Jonathan V. Last at The Bulwark gave a recommendation.
Statistical question / hypothesis regarding substance use disorder:
Many analytic methods don't deal well with massing at 0 and/or 1 (whatever the boundary condition is). There are large proportion of the population who are teetotalers, and a small proportion of the population of drinkers drive a significant amount of alcohol consumption. I assume (probably incorrectly) that many substances have similar distributions of use per capita/time.
Question - wouldn't this bimodal distribution drive taxonicity in statistical analyses? Despite this, substance use is not cleanly divided in the clinical context where substance use is 1) comorbid, 2) often a coping mechanism, 3) not necessarily problematic.
It seems like language is biased towards categorical thinking - either you choose to use a word or you don't. You need to know from context that it's not what people really mean, that tallness is dimensional. Sometimes it's tricky to hint at dimensionality without being overly vague.
And yes, there is some human-rabbit overlap in sizes.
https://www.catersnews.com/stories/animals/meet-the-worlds-biggest-easter-bunny-and-his-brood/
my takeaway is that if I catch a flu then I will be telling everyone to stay away from me, for I am a Flu Person
Is “person with autism” really misguiding? I mean, we already have and use terms such as “people of color”, or “people of size” and I don't think anyone really assumes those are binary categories.
Babies are basically bunnies. Small, young, dumb, cute.
1. Why is CCFI of 0.5 chosen as the border between categorical and dimensional? In other words, it seems "being categorical" or "being dimensional" is itself dimensional - is this true?
2. Question about your practice (ignore if inappropriate) - comparing the post here and on your site, I see you've kept the personal tone, which surprised me. Did you get any feedback from patients/non-SSC-readers about the content there? It might be too soon to know, of course.
I wonder if the positive feedback loop of addiction makes it categorical. "Propensity to become addicted to cigarettes" could be dimensional, but "is currently addicted to cigarettes" (or "has ever been addicted to cigarettes") could be more like separate categories, because it means "consumed enough nicotine for the positive feedback loop to kick in." How much that positive feedback loop affects you varies, but that's like variance in the strength of the flu which makes the "has flu" lump wider than the "doesn't have flu" lump.
Maybe also relevant that "how much nicotine would it take to set off the positive feedback loop" and "how strongly will the positive feedback loop affect you if it gets triggered" are strongly correlated (I imagine) as two aspects of "propensity to become addicted to cigarettes." So the people who are more strongly affected by that positive feedback loop are also more likely to have triggered it (since that takes less nicotine for them).
My personal impression of autism is that it's both. Aspergers seems to me to be more dimensional, whereas low-functioning Autism (for lack of a better term) seems more taxonic.
I dont have any particular study to back this up, its just more of a result of all the stuff I have read about Autism. Its probably also influenced by my belief that it might be wrong (for social reasons) to lump the two together into one diagnosis.
One of the things I love about this blog is how you pull meaningful conclusions that have fascinating real-world implications out of math that goes over my head (but is still interesting to read an analysis of). I went into this one taking a gamble that it would be interesting to me, and hit pay dirt at the end. Cheers :) and again, welcome back!!
Has someone does the analysis to determine whether psychiatric disorders themselves split into two taxa of taxonic and non-taxonic, or if they lie on a spectrum between the two?
For those like me who read the version of this post on Lorien: the only major addition for ACX is section III.
Psychological constructs in general are just low-dimensional approximations of complex, high-dimensional processes. The relevant question is not "are they real?", but "are they useful?".
For a similar discussion, see section "Realist intuitions impede progress in psychology" on page 4 here: https://psyarxiv.com/xj5uq
This is a conventional take which doesn't even attempt to grapple with the true complexities of the issue. The question is not whether current diagnoses are dimensional or categorical but rather whether they exist as meaningful separate diseases at all. And most data points suggest the answer is no, that the current system is a joke. A nosology useful to clinicians for efficient collegial
communicate but with no benefit - and often harm - to the patient.
Mentioning riches as an example of a dimensional variable is apt because it appears to be as useful a variable as depression in understanding human beings (I.e. not). That's because a) it's unreliable - most people's wealth fluctuates, often quite drastically, over the course of their lives. And of course b) it doesn't actually exist - money is a social construct.
When we talk about wealth what are we really attempting to get at? The variable that scientists - and particularly epidemiologists - have found most approximates that is socioeconomic status, which takes into account education, profession.and economic resources.
Depression more than likely is but one pillar of a larger, actually useful category - perhaps "internalizing disorder". Or a more accurate system might blow up the whole thing and not even use the heterogeneous DSM concept of depression at all.
Any system must deal seriously not only with the current in-vogue biological factors, but also must account for culture and time bound disorders and the drastically different rates of recovery of mental illnesses depending on where one is (i.e. westernized vs non westernized regions). When outcomes for schizophrenia are better in sub saharan africa than the US, something is wrong.
See:
https://www.nature.com/articles/d41586-020-00922-8
https://www.researchgate.net/publication/267383152_Counterflows_for_mental_well-being_What_high-income_countries_can_learn_from_Low_and_middle-income_countries
https://wchh.onlinelibrary.wiley.com/doi/pdf/10.1002/pnp.461
This reminds me of a semi-recent post from Freddie DeBoer, on genius.
https://fredrikdeboer.com/2020/11/29/2049/
The taxon-dimension distinction seems important in discussions of talent; while talent varies greatly, "genius" as a truly separate category doesn't seem to exist. But many people, including Freddie there, conflate the question of whether genius is real with the question of whether talent is real.
You might find these interesting.
"Complexity perspectives on behaviour change interventions"
https://mattiheino.com/2020/10/19/besp/
Youtube Channel: "Complex systems in behavioural sciences"
https://www.youtube.com/channel/UCR9nYEjzOCzQLjxDgKo0EZA
An argument I've encountered that ADHD "really is" a specific medical condition, and not just a fancy name for below-average concentration skills, is that apparently when you give ADHD medicine to a person without ADHD, it actually makes them more jittery and less able to focus -- IOW the medicine has opposite effects on people with and without ADHD. Is that true, and does the argument prove what it sets out to prove?
Clearly we need a meta-meta-analysis to determine whether taxonicity is a categorical or a dimensional concept.
If you use the same methodology to determine if amputation is dimensional, how does it fare?
This is a smart, entertaining post and I learned a lot. I was also reminded of something that I am reminded of a lot, which is that a lot of very smart people don’t really know much about addiction or at least don’t think enough about the power of words in its weird realm.
The exclamation point after “gambling addiction” I found particularly troubling, and a little snarky. Spend some time at a GA meeting and there will be no question in your mind that it’s an extremistanic not dimensional phenomenon. There is no meaningful line that connects my thrice a decade purchase of lottery tickets and gambling addiction.
My own experience and long-term observations of folks in recovery tell the same story about alcohol. There are cats and dogs and to an alcoholic alcohol is catnip, which lots of dogs might have now and again but does not lead them to massively destructive behavior. Also, sadly, while Jeff Bezos could buy enough (I suppose) ivory back scratchers to exit his personal extremistanic state, the addict has a one way ticket. Pickles cannot revert to cucumbers, etc.
Why do I bother to write this? Because, again in my experience, one of the defining features of addiction is that there is a constant internal dialogue that amounts to “abstinence is an overreaction, I can [gamble, drink, have the occasional benzo, etc.] just like everybody else.” It’s a disease that works to convince the afflicted that they are healthy. So in my view it is really dangerous to casually propagate dimensionality theories about addiction. Down that path lies an enormous chasm of pain for the addict and those around them.
"And look what isn’t on here – schizophrenia, which is just a really obvious separate taxon. You know, the condition where sometimes between ages 18 and 25 for men and ages 25 to 35 or so for women, over the course of a few weeks, seemingly normal people start getting extreme hallucinations and eventually devolve into a state where they often can’t live a normal life or even speak meaningful sentences? The taxometricians are saying ah, whatever, it happens to all of us, they’re just the people who it happens to more than average?"
["yes" chad image goes here]
1) Median household income in the article is off by nearly 100% ($63k in 2019) (https://www.census.gov/library/publications/2020/demo/p60-270.html)
2) "Has the flu" in terms of medical treatment isn't categorical based on "is infected by influenza virus or not" - it's an assessment of the dimensional response to being infected/exposed, which may run the whole way from "shrugged it off without even a sniffle or headache" to "death". As a reminder, this dimensional nature of the response to covid is part of what is making it so hard to manage.
3) If your chart is showing 'gender' as being more likely dimensional than categorial, then I think your chart has serious data issues. I myself would use that as a 'sniff test'.
Overall, I agree that looking at the human condition as a spectrum in most cases is very useful, and that it's probably best to intervene only so much as necessary to, as you say, allow a person to go on with their life in a productive and pursing happiness manner. Whether this is best money, or medical (procedure/chemical) treatment, or coaching into personal growth...there are lots of tools.
> this has another implication: stimulants shouldn’t be thought of as magic bullets that “cure” “ADHD” by fixing the underlying cause, in the same way that Tamiflu cures the flu by blocking flu viruses. They should be thought of as things that affect the underlying stew of variables that cause ADHD in some helpful way.
I don't think this is right. Conceptually, we should be able to say that, among the many dimensions along which we can measure a person, some of them are, by virtue of extreme or unfortunately-combined values, causing ADHD in that person.
And some of them are being modified by the expression of ADHD -- instead of being causes, they are among its effects.
The point of medication is to relieve a problem, such as an inability to see things. You can do that by removing a cause of the problem, and leaving the uncaused effect to disappear naturally (we can call this approach "laser eye surgery"), or you can do it by removing the effect while not doing anything about the cause. (We can call that approach "glasses".)
I had the impression that stimulants are viewed more as altering a symptom of ADHD than as affecting its causes.
I'm curious as to how this might work when there are more than two relevant categories: for example, colds also produce flu-ish symptoms, and the presence of another lump in the middle would move it away from the bimodal setup. I may have to look into this.
Minor nitpick, but you should probably make better figures than the hand-drawn ones on your website, even if the axes are completely 'symbolic' and only serve to prove a qualitative point. It'd look more professional.
Now do the same thing with traditional US denominations for race and you'll understand why it is seen rejected as a biological construct by the overwhelming majority of geneticists and population scientists.
> or you’ll get disillusioned and radicalized and start saying all psychiatry is fake.
This is about where I am.
My own experience with trying to get diagnoses and treatment for depression and ADHD for myself and for my kids has been super-frustrating. I know two young men who were diagnosed early with ADHD who later went through a smorgasbord of different diagnoses; one ended up as bi-polar and the other as autistic. My sense is of psychiatrists looking at clouds and saying "methinks it is like a weasel".
Add in the fact that you have to pay $1000s to each cloud-watcher before they can identify the right weasel and the fact that the next cloud-watcher will almost certainly tell you that the previous one was full of shit and it's hard to have confidence in the industry as a whole.
I wonder if, perhaps, this is why so many of us are eager to sign up with Scott's practice. At least Scott is aware of the uncertainty in the field. I fear the unwarranted certainty that so many doctors speak with most of all.
A lot of people have already identified the seemingly arbitrary deciding line between "dimensional" and "taxon" but it seems to me like there is a little question-begging in the other direction too.
Scott lists the criteria for various conditions, but the very existence of both the conditions and the criteria is, itself, down to some questionable statistical shenanigans.
Before asking whether depression is a categorical thing or a dimensional thing, shouldn't we be asking whether depression is a thing at all? And is it categorically a different thing than ADHD or bipolar? Or did they all just arise from more multi-variate cloud-staring?
I may have to use the "Jeff Bezos of absent-mindedness" line at some point. But beyond that, this is a helpful way of looking at something I've been thinking about on my own account for a while. (Though it would have helped if I hadn't read it on the Lorien website earlier.)
I wonder to what extent these 'objective' categories are just artifacts.
For example, imagine that angriness is a normal distribution, but psychiatrists don't actually test how angry someone is, but they consistently ask the question whether the person beats their partner. People who do are diagnosed with IED, those who do not, aren't.
What if anger is just a normal-distribution, where most people express this in societally acceptable ways, like posting angry screeds on Twitter? What if beating your partner has high societal and relationship costs, which result in high threshold costs. In other words, there is little cost to beating your partner more if you have done so before, but a high cost if you never did it before. This would then tend to result in an artificial dichotomy, that reflects external conditions, rather than the person being diagnosed.
The IED diagnoses would then seem meaningful, because it correctly identifies a 'real category' of people who beat their partner a lot (one of two clear bumps on a graph), but in reality, this is in largely part due to strong disapproval of partner-beating by society and most partners.
If you were to measure anger by number of angry tweets, you might find something close to a normal distribution, so then IED would suddenly no longer seem like a real category.
I've been interested in a related topic recently: suppose there's a condition that's truly taxonic, but we live in a world with diagnostic fads and Szaszian malingering. Would we be able to find which people have the real condition?
Think back to the 1990s when every other young girl with vaguely negative body image was tagged with an "anorexia" label. Might this cause anorexia to be seen as dimensional when it's actually taxonic?
I can sort of see why autism might be taxometric. My wife has autism. She's extremely high-functioning and I'm weird enough that an outsider wouldn't necessarily see a huge difference between us, but from the inside, there definitely is. There's basically a social processing module that she's missing. Stuff that to me is trivially obvious to the point that I have no clue how to explain it because everyone just gets it is completely baffling to her. She does a splendid job dealing with social situations despite this, but it's definitely a difference. Likewise with the sensory sensitivities, which I basically don't have.
(Of course, this is just one anecdote, and may not generalize.)
Beyond the distinction between categorical - continuous distribution, there is also a more fine grained scheme, to which you allude, in which the "continuous" category is split into two sub-types: those with long and those with short tails. Short tailed distributions (e.g. the normal or exponential distribution) have a well defined scale (the standard deviation for the normal distribution) which allows you to say whether an individual is average or exceptional, even though the distribution is continuous. Height is an example. For long tails, i.e. power laws, you don't have such a measuring stick - they are "scale free". For example, wealth is power-law distributed: for any 10 people of net worth $x, there is one person of net worth $10x - this makes it much harder to define "exceptional cases" than in the case of normally distributed attributes. Here's my question: which "continuous" psychiatric conditions are long, and which are short tailed?
One place that categorical thinking (about ADHD) has real-world consequences is in academic accommodations. If a doctor signs off that you that you pass the bar for ADHD, you get 50% extra time on all exams, no questions asked. If you're not quite at the bar, nothing. As a professor, I've responded to this arbitrariness by trying to make exams less time-sensitive, but not with perfect success.
If ADHD is dimensional and prescribing medication is a matter of risks and benefits, why do we only prescribe it either EVERY DAY or not at all? Some people would like to have it available to take once a week, for, say, their most tedious task. What are the very serious risks that outweigh that need, assuming they have no history of addiction or heart problems?
Scott, I'm reading the brief primer, but not seeing addressed why they choose these methods over normal cluster analysis, which is typically what you'd do when trying to decide whether some category is "real" or not statistically. Is it just because they have low-dimensional data? That doesn't seem like a good reason, since that makes cluster analysis even more interpretable if you can actually plot the clusters. "Within group variance is less than between group variance" at least seems more principled and intuitive than "CFFI is greater than 0.5."
Minor nitpick, Scott uses Taleb's terminology incorrectly. Bezos doesn't live in Extremistan. Wealth is in Extremistan, while height is in Mediocristan.
Mediocristan is where you find all the nice normal distributions which look like Bell Curves. Extremistan is where you find distributions with really long tails (power law type stuff).
Another good analogy with height is that, theoretically, there's nothing _wrong_ with a short man; but, practically, he's generally going to struggle in life in our current social milieu. In the same way, there may be nothing inherently wrong with someone who struggles concentrating (maybe they're just not interested in the subjects of the current social milieu), but they'll struggle "fitting in". This might be a more tempered way of looking at the often strange spiritual implications of psychology and psychiatry rather than going full Szasz.
There's a classical logic reason to deny that there is a difference between categorical traits and dimensional ones. Take ADHD for example. If it is dimensional, then there will be cases where you want to say of someone that they are not ADHD, but they are not not ADHD either. They sit in the vague borderlands, with some ADHD symptoms, but not the full suite, or they don't have them very severely. Well, according to the rule of double negation, "being not not ADHD" = "being ADHD". From the preceding reasoning, it follows that the person is both ADHD and not ADHD, which is a contradiction. Since this reasoning will work with all dimensional/non-taxonic traits ("rich", "tall", etc.), it follows that no trait is truly dimensional. They are all categorical/taxonic. Now, it may be a permanent limitation on human knowledge that we can never know the precise boundary between being ADHD and not being ADHD, but there must be such a boundary. If you think, hey, let's just give up classical logic and say there are propositions with truth-value gaps, that leads to problems too. One of these is how to fix the boundary between propositions with truth values and those without without just replicating the whole vagueness problem all over again. Plus it seems, as my Dad would say, like too much sugar for a nickel.
From what I gather the RDoc framework (https://pubmed.ncbi.nlm.nih.gov/26845519/) is an attempt to steer psychiatric/psychological research in the direction of more dimensionality and away from categorization. I would think the advent of digital phenotyping and behavior tracking afforded by apps/bands would enable greater granularity of behavior/symptoms in this regard
A problem here is that many of the drugs that psychiatrists prescribe are controlled substances, so you need clear guidelines to avoid prosecution. Of course there's an obvious solution - stop controlling substances - but in the actual world.
The end reminds me of a conversation I had with Thomas Szasz many years ago. I asked how mental illness could be only a metaphorical "illness" if people who are not functional become functional by taking psychiatric medications. He said, "Do you know anyone who can't function without cigarettes? Would you say they suffer from a nicotine deficiency?" A similarly pragmatic POV.
I don't think you are properly following out the logic of your argument. You start with flu because we have an unambiguous test, infection with the influenza organism, that shows it to be categorical, you either do have it or don't. If the various statistical tests show it instead to be a continuum, that's evidence not that it is a continuum but that they are insufficiently powerful to tell. That's a straightforward Bayesian argument, starting with a very strong prior, itself based on direct evidence, that people either do have flu or don't.
In the case of gender we again have a test, indeed two tests with almost perfect overlap, genetics and structure, that divides the population into two categories that cover almost everyone. If the statistical tests that were used fail to identify it as category rather than continuum, that's evidence not that it's a continuum but that the tests are sometimes insufficiently powerful to tell. It follows from that that the failure of the tests to identify schizophrenia as categorical doesn't mean it isn't categorical. Indeed, if you have other strong evidence that it is, it gives you another strong reason to lower your confidence that the tests give the right answer.
One could, I suppose, try to avoid that conclusion by saying that what they are measuring is gender, not sex, and it isn't defined by genetics or structure. That's like saying that what you are looking for is not infection with the influenza virus but symptoms of flu. But in both cases, that amounts to throwing away strong evidence — the existence of an unambiguous categorical difference that correlates with the symptoms being observed. How, after all, do you decide what the symptoms of male or female gender are, other than by observing their correlation with other symptoms of the same gender? So how can it make sense to throw out XX/XY genetics, which correlate pretty strongly with m/f gender?
Not sure if the figure style has a particular pedogogical purpose, but here are some less pixelated versions you are free to use, Scott (and I can fiddle with them further if they don't match the existing ones properly).
https://drive.google.com/drive/folders/1a6yBkVez10cnSMMLgw-DW-ksY5Dj94BB?usp=sharing
Good to have you back.
Interesting analysis. In order to make it more comprehensive, I'd add the concept of a threshold effect. We see these different types of systems, where reaching a certain threshold in a continuous variable system has a dramatically different effect than any value below that threshold.
The action potential in a neuron is a classic example, where a slightly lower intracellular voltage doesn't trigger the neuron to fire, but then once the threshold is reached a complex series of events unfolds. Another example is phosphorylation of the rb protein, which requires a certain threshold to be reached to trigger the cell cycle. Below-threshold rb phosphorylation gives you no cell division. Once the threshold is reached, the cell will divide.
Regarding autism, the plots that nostalgebraist quotes to argue agains Baron-Cohen's theory at https://nostalgebraist.tumblr.com/post/164069138209/in-lieu-of-a-longer-post-ive-been-planning-to look to me like there is something taxonic going on here (the green ones in Act 2).
The gender thing is cute, but what does it actually mean? Some people are saying that if the test doesn't show gender as taxonic then it's obviously bunk. That's certainly true if they've done the test properly, with at least one factor indicating biological sex. But what if they haven't (as seems to be the case)? Does that mean they're measuring some kind of "gender minus sex"? Should we expect that to be taxonic?
it's interesting to think about a difference between absolute and relative categories, sort of complicates the relationship between the dimensional and the categorical.
rabbits and humans are categorically distinct in relation to one another and according to specific criteria. but if you abstract either term from the relation, the categorical definition says virtually nothing about the thing it defines (ie. a certain weight or what have you).. its Diogenes with the chicken..
relative categories are pragmatic constructions which are highly unstable, misinterpretable because of their contingency, hence the aforementioned example. the problem is that, qua the terms of the definition, the relatively defined category can belong to a larger set of things that share the qualities by which it is defined (again, size, shape, weight, whatever) -- whence comes the possibility of some equivocal identification. the relative category can inadvertently create a new dimensionality consisting of the vague objects falling under it (especially if empirical conditions are not ideal).
an absolute category, by contrast, would need to be defined according to terms which do not and could not make something falling under it a member of some more-general set (of arbitrarily non-rabbitish things, or what have you).
and with an absolute or singular definition it need not be a matter of either/or in terms of diagnosis, or even both/and.. it's a matter of necessarily belonging to an absolute category to a contingent degree (which of course could still be 100%). so i suppose the absolute category diagnoses according to singular intensity rather than a bi- or multi-valent belonging in this or that category.
> My guess is most professionals, and an overwhelming majority of laymen, are actually confused on this point, and this messes them up in a lot of ways.
Thank you for writing this. This had annoyed me a lot, back in the school.
Many professors seemed to confuse the map with the territory, although they might have used confusing languages just for convenience' sake.
Regardless, such confusion have propagated to fellow students, and the vicious cycle continues.
Many physical diseases involving complex diagnostic criteria seemed to share similar proneness for confusion.
This kind of reasoning should be taught at med schools.
> I’ll grant them pedophilia – it really does seem like people either are or aren’t pedophiles and there’s something weird and specific going on there.
Total aside to your main point, but (speaking as one myself) I'm not sure if this is true. Certainly I know people with varying levels of attraction, people who are attracted to both children and adults but more to one or the other, people who only realized later in life that they are attracted to children, etc. I think this one could easily turn out to be dimensional.
In other notes, thank you for not conflating pedophiles with child molesters.
If "taxometrics" gives the wrong answer for schizophrenia and gender, shouldn't we just ignore it?
--
You suggest that for "dimensional" traits like wealth or ADHD-ness, we shouldn't expect a specific underlying cause. But extreme values of a dimensional trait seem like they might well have an underlying cause. If you ask why Jeff Bezos is so wealthy, the answer isn't, "well, you know, who can say, it's just a confluence of a bunch of small factors: he worked a bit harder than normal, his parents were a bit wealthier than normal, he saved a bit more scrupulously than most, etc., etc." No, there's one big reason, namely that he founded Amazon. Similarly, I think it's true that people at the true utter extremes of the height distribution have a single specific underlying disorder that causes them to be really really tall or short. It seems like an empirical question whether the extreme values of a dimensional trait have a specific underlying cause or not.
Machine learning takes a bunch of items with labels ("rich", "not rich") and attributes ("owns more than 5 cars", "reads the New Yorker") and then comes up with a formula ("algorithm") for deciding whether an unknown item fits a label. The formula typically has no explanation (it might have a confidence value for a decision), but there are ways of giving a partial explanation of how it works (e.g., fitting a decision tree to the formula).
The DSM criteria sound like a simplified version of machine learning. And, as we know, machine learning used without care can have all kinds of problems; for example, built-in bias because wrong attributes were chosen or the attributes are correlated with other things that were ignored, such as gender or ethnicity.
Just an FYI, but tamiflu (ostelamivir) barely works at all: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4904189
Irrelevant nitpicking of the highest order but I think your figure for average rabbit weight is low by 2-3lbs.
Interesting but spoilt by muddle headed thinking about 'schizophrenia'. There is a vast literature stretching over thirty years, most of which has nothing to do with taxometrics, which supports the continuum model. Moreover, something that has been overlooked is that the diagnosis groups together a collection of symptoms which researchers have increasingly recognised may have nothing to do with each other, so its possible that some are taxonic (I have my suspicions about hallucinations) but some are not. I carried out a very large taxometric study of paranoid symptoms and the findings wholey supported a continuum model: Elahi, A., Perez Algorta, G., Varese, F., McIntyre, J. C., & Bentall, R. P. (2017). Do paranoid delusions exist on a continuum with subclinical paranoia? A multi-method taxometric study. Schizophrenia Research, 190, 77-81. doi:10.1016/j.schres.2017.03.022
Scott, on the topic of autism, have you read this paper and what do you think of it? https://link.springer.com/article/10.1007/s40489-016-0085-x
We published a letter related to this topic:
Schizophrenia does not represent the extreme of a normally distributed trait
https://www.cambridge.org/core/journals/psychological-medicine/article/letter-to-the-editor-schizophrenia-does-not-represent-the-extreme-of-a-normally-distributed-trait/263DD068942ABE41AD84A500417B8C3E
I actually tried making that argument about schizophrenia to Greg Cochran, and he replied that GWAS studies really do support the notion that it's just the extreme end of broad variation, like mental retardation:
https://twitter.com/TeaGeeGeePea/status/1328943170957275139
I really do like this piece in general. I am a physician and I also am an individual with high functioning ASD. My problem is with the final paragraph. Language and the meanings assigned to words is a highly variable thing, but as someone with ASD, my "feeling" is that terminology such as "a person with ASD" is much less binary/dualistic than "ASD person." Describing me as "a person with ASD" at least makes me feel like I am on a continuum that includes everybody else. Not that the way I as an individual feel about it is all that important. Just introducing a perspective.
It's not really clear to me that what taxometrics is measuring is well-defined. We could of course *define* categorical properties to be ones that pass MAXCOV or some other test, but if we reduce categoricity to that, it's not clear why we should care about it. So what does it mean to "neatly, objectively separate into different groups"?
One answer would be "conceptually dividing things into these discrete categories is the right way to think about them". Spelled out like that it should be pretty clear that this can depend on a lot of things other than the statistical properties of the dataset, but I suspect that this interpretation is a big part of why taxometrics feel like they're telling us something important.
Another possibility would be "the process which generates the observed data significantly involves a discrete random variable". I see two problems with this one. First is that it's not really testable: we can get similar distributions by means of very different underlying processes. For example, if we have a variable X distributed according to a standard normal distribution, then arctan(100X) will be highly bimodal, and if our observables are three different noisy measures of arctan(100X), they'll pass the MAXCOV test, and I'm pretty sure that they or something very similar would pass any other statistical test of categoricity as well*, even though no discrete variables were involved in the process.
The other problem with characterizing categoricity as "the process which generates the observed data significantly involves a discrete random variable" is that the physical processes which actually produce the data are not the same sort of thing as the mathematical models we use to think about them. With a mathematically defined process for generating a random variable, we can say "look, right there, on line 3, there's a discrete random variable", but the physical process isn't always going to have some specific location of the variable that we could point to. If there's a specific gene or something, that's pretty clear-cut, but often there isn't. And when there isn't, whether or not there's a discrete variable involved is a property of the *model*, not the world, and we might get different answers depending on what model we use and how granular it is. For example, you talked about "having the flu" as a binary categorical property, and certainly under many situations it's appropriate to model it as a discrete random variable. But a more detailed model might look at viral count and antibody count as two real variables which change continuously over time. Or something like that; I know very little about immunology. But I'm pretty sure you can't tie it to something as physically unambiguous as "are there more than zero viruses in your body" because someone who fully recovered from the flu and has antibodies shouldn't be classified as infected if they inhale a few viruses.
Maybe these problems have been resolved somewhere in the taxometric literature, but I'm not seeing them addressed here or in the linked primer, and they seem pretty fundamental.
*because arctan(cX) converges in distribution to a coin flip between pi/2 and -pi/2 as c goes to infinity, so statistical tests shouldn't be able to distinguish it from the discrete case for large enough c
Time for the meta question! Is the distinction between taxonic and non-taxonic traits itself taxonic or non-taxonic? What would the CCFI distribution look like for lots of different traits?
There's a tasty and tempting irony in taxometrics. The field says: "Turns out many of these supposed disorders are actually traits that are non-taxonic." Then one is tempted to say "I'm going to care deeply about which traits end up in the binary bins of taxonic or non-taxonic!"
This kind of side-steps the really important question of "when and why do seemingly smooth distributions give rise to important and distinct taxa?"
The answer to this question is nearly always, "there are self-reinforcing effects going on and the taxa are divided by the point in the distribution where the variable will grow over time to where it will shrink".
So your income graph may look perfectly smooth, but there are nevertheless two distinct taxa in people who are living hand-to-mouth, who will see stagnating or declining incomes over time and people who are investing surplus income in future income generation, who will see income growing over time.
Similarly, disease infectiousness may be a continuous spectrum, but there is a big division at r = 1 between diseases that will die out after a few people have been sick and those that will become pandemic.
It feels like many (if not most) mental illnesses have this kind of flavor. The distinction between a person who is depressed and one who is merely miserable is not really how miserable they feel. The distinction is whether that misery compounds itself over time until they cannot get out of bed (and are then miserable about the fact they have not gotten out of bed) or whether it decays naturally as they switch to happier activities to make themselves feel better. Treatments, too, tend to focus on breaking cycles, not really shifting the set point.
It's ironic that this is missed in the article, because your example of the most clear taxon in Psychology is mostly defined by its instability. If your schizophrenics had been unable to speak meaningful sentences since childhood, would you be quite so convinced that they were not just at the long tail of the language development spectrum?
By this distinction, depression, addiction, anxiety, schizophrenia, bipolar disorder, etc. would likely be distinct taxa categorized by the point where they become unstable. ADHD and autism would probably be somewhat arbitrary lines drawn on a distribution.
I'm curious what you think of the quasi-critique I made here: https://nostalgebraist.tumblr.com/post/630848028609249280/slatestarscratchpad-i-think-ive-been-looking
In short, taxometrics doesn't seem well-validated as a statistical technique. Unlike most statistical techniques one sees in practice, it neither has backing from strong mathematical theory, nor from the "real-world stress test" of a large number of people independently using it on different problems. That doesn't mean it *couldn't* work, but feels like it has a closer family resemblance to "algorithms I personally invent for personal projects that seem to work OK" than "anything you might see used by eg Andrew Gelman in a paper."
Separately, I'm unsure it makes sense to apply it in psychiatric patient samples, even if the idea makes sense in some other contexts. From the post:
> After all that, you still have the data you have, which in psychiatric contexts will be sampled from the general population in a very non-uniform way.
> The idea makes sense in an idealized world where your research sample is drawn randomly from the population of All Possible Humans. But psychiatric samples are very unlike that. You can try to remedy that by introducing some control people from the general population, but then you’re introducing a two-category structure into the data (controls vs. patients)!
When I was in college I was depressed so I went to the doctor; they had me fill out a questionnaire and based on my answers decided I wasn't really depressed and basically told me to get over it. One of many experiences that really put me off seeking medical help :(
Could someone please do a meta-analysis of Comparison Curve Fit Index studies to determine if 'taxonicity' and 'dimensionality' are separate categories or just different points on a continuous spectrum?
I'd be curious to what extent the legal social worker professions have driven practical psychiatry toward distinctive categories.
First reaction; YES! So much what I’ve been thinking and teaching and explaining to patients for years.
Then so no no no; eating disorders and substance abuse as categorical? This mistake arises because psychiatrists (and most MDs and most hospital-based psychologists, the ones who do most of the research) only meet the most severe cases, the ones who haven’t managed to stop by themselves or with the help/push of concerned friends, families, employers, AA etc. (BTW, for those most severe cases, there’s very high co-morbidity with early trauma and/or personality disorders. Both also dimensional.)
I’m betting the researchers doing the studies that led to those results in the meta-analysis only included people severe enough to seek/ be brought to professional treatment. Plus we often don’t ask about these issues, unless it’s part of the presenting problem.
Scott, go find the research on how many people have ‘sub-clinical’ ED, alcohol abuse, etc. How many people are ‘overly careful’ about what they eat or how trim they stay, or who binge, but not to the point their teeth are rotting. How many have a year or two or three of drinking or drugging enough to start wreaking some havoc. How many manage to pull themselves back from the brink. To ‘fix themselves’, alone or with the help of a school counsellor, an EAP provider..... to start a virtuous cycle that leads them to a much better place. Without ever talking to their GP about it or seeing a psychiatrist.
> I’ll grant them pedophilia – it really does seem like people either are or aren’t pedophiles and there’s something weird and specific going on there.
As a carrier of this trait, I'm conflicted about this myself.
On one hand, I'm definitely able to fall in love, have sex, and form emotional attachment with adults. I've had a reasonably full love life but none of my lovers ever were under the age of consent. Right now I am happily married with kids (and yes, my wife knows all about me).
On the other hand, I feel definitely binary about my sexual attractions. There are many adult females that are about as attractive to me as a crocodile. For an attraction to be possible, I need to see some hard-to-define childishness in demeanor that doesn't depend much on age but that most adult women (and even some children) totally lack.
So, if I define myself as someone sexually attracted to children (i.e. a binary category with a legal cutoff), then I am on a spectrum: I can experience attraction on both sides of the cutoff. But if I define myself as one attracted a certain personality/appearance trait, which is itself a spectrum, then I am binary: if you don't have this trait I have zero interest.
About autism:
I believe when people talk about autism spectrum, they are talking about a different thing than taxonometrics. What they are saying is that there are multiple ways to be autistic, and all autistic vary a lot on their presentation of autism. They are trying to dispel the popular image of autism as Rain Man, or Sheldon Cooper. This is entirely compatible with autism being its own taxa, with some fuzzyness in the border, just like some people may have the flu, but no or few simptons, or not have the flu but have some flu-like simptons.
Anedoctly, I've seen activists saying, precisely this, that autism is a spectrum, but some people are clearly in the spectrum and some are clearly out.
It sort of seems like the psychiatric profession should move away from the notion of 'diagnosis'. Diagnosis is fundamentally a verb for categorizing. Perhaps it'd be more appropriate to think of things in terms of scores. Patients take a depression test, which gives them a 0-100% depression score, and then you can define ranges of that score that indicate the value of clinical intervention. Functionally that's sort of already what's happening, but re-framing it in this way seems like it might be a better epistemic map projection.
It totally makes sense, that autism would look like the third graph in your article, the one with the flu. Autism is probably a mix of different conditions with various reasons. Some versions of autism are strongly genetically determined. There is a sperm donor, whose all ofsprings are autistic, like 100% https://www.google.com/amp/s/www.washingtonpost.com/health/the-children-of-donor-h898/2019/09/14/dcc191d8-86da-11e9-a491-25df61c78dc4_story.html%3foutputType=amp
Reading this as a physician, I feel suddenly struck by how much of medicine may actually be comprised of things-more-like-height.
Hypertension, as mentioned, is a classic example, but I feel like actually the more closely you look at various health conditions, the less they independently exist as discrete things in the world.
A few examples;
Diabetes (type 2) - basically like high blood pressure. we use arbitrary cut-offs of certain Hba1c measures. The fact that ‘pre-diabetes’ is a diagnosis in itself is telling.
Dementia - very much a spectrum, we use scoring systems, ‘mild cognitive impairment’ being a diagnosis ‘before’ you reach a score “bad enough” to have dementia.
Heart attack - on a spectrum with stable and unstable angina. Yes we have various measures to determine how likely it is that part of the heart muscle is damaged (ECG changes, blood test markers, again arbitrary cut-offs) - but part of your heart muscle dying is not really something that does or doesn’t happen. Suppose 5 myocytes died because your heart was under strain temporarily? - not a heart attack.
Even infections - it’s not really as simple as ‘is this virus or bacteria present in X body system?’ - if you culture somebody’s nasal cavity, or poo, or skin, you will often grow bacteria, and these are not infections. One of the key defining criteria of a pneumonia is whether or not you can see an area of infection on a chest x-ray - again, this is obviously not yes or no.
I think extremely few things really exist in a discrete way (aside from genetic disorders controlled by a single gene, which seem to be an unusual case). We make up categories to describe continuous variables in a way which is most helpful for deciding when to use different treatments. Seems equally true to me for most mental and physical health conditions.
"Is X dimensional or taxonic?"
"Ah, well, it's kind of in between."
ಠ_ಠ
I feel like this missed a good opportunity to make another supporting point: Jeff Bezos only makes $81k/yr salary. (Everything else he has was likely fundraised by sales of AMZN stock to people looking to own small fractions of Amazon... probably 9-figures some years, but probably near 0 some other years. And it also would include a lot of people who are technically broke. Point being, categories are difficult, and easy to manipulate.)
Duuuuuude: taxons themselves fall on a spectrum 🤯🤯🤯
Wonderful text, thanks a lot. There's even a new section in the DSM in which they want to make numbers out of all these complicated diagnoses, I think it's called "emerging theories and models". My colleagues like this a lot. Nice terminology for something that is probably mostly driven by convenience.
In the old days, there were complex psychiatric diagnoses, categorical, and the psychiatrist spent quite some time finding out the problem. The problem: you cannot run the nice parametric statistics, you need proper numbers.
So, rating scales were developed, Hamilton rating scale for depression, anxiety, and so on. The ratings were made by psychiatrists, and they needed to take classes to learn how to do it properly. The problem: it takes so much time. Imagine a randomized study with 200 participants, nobody has time to perform 200 interviews.
Then, more and more questionnaires were invented (the nice term for this is: "patient-reported outcomes"). Beck depression inventory, GAD-7 is also a questionnaire. When I ask my students in class, how do you want to measure X, they almost surely respond: "With a questionnaire". Using questionnaires, the researcher can save a lot of time because the patient is diagnosing himself or herself. More than 20 on BDI -> moderate depression (I don't remember the exact figure). No psychiatrist needed for this diagnosis. And since clinical studies are often collaborative, and every collaborater has a pet scale, many questionnaires were put on the table of the patient. The problem: piles of paper forms that no one wants to type into the computer.
Then, the internet came and everyone bought a smartphone. Nowadays, the patient receive a link to an ever-growing pile of online questionnaires, which you cannot escape until you have clicked on the last of, say, 500 questions. I have seen studies in which, according to my calculations, patients spent hours of with filling out online forms. One of the phd students involved in such projects told me that many patients just put a cross in the middle category because the online forms are way too long. There's also a nice orwellian word for this nightmare: "measurement-based care".
One more funny thing is that there are scientific guidelines from the regulatory authorities in medicine (EMA, FDA) that actually limit the use of all this. E.g., the need for a "primary outcome" in a clinical study, which prevents you from collecting 20 variables and choosing the one that shows most pretty results. E.g., the EMA guideline for patient-reported outcomes in oncology that says, at most 20 min for questionnaire. E.g., the EMA guideline for depression that tells the researchers to do blinded assessment using rating scales. I am digressing, sorry.
There is actually a quite sophisticated "representational theory of measurement" by Krantz, Suppes, Tversky, Luce. The theory was developed to enable measurement even in the absence of concatenation operations such as putting rods together so that the length adds up, or putting two objects in the same pane so that the weight is the sum of the two. In the social sciences, such concatenations don't exist, and Krantz et al.'s theory can be very helpful here to establish an interval scale. But the theory is complicated, so no one makes use of it, neither the proponents nor the opponents of the dimensional approach. Instead, the meaning of a scale is mostly established by its name. If the name of a scale is Novaco Anger Scale, it obviously measures anger. Add a few correlations with other "instruments", and claim that your scale is "valid".
One more funny thing: I think the whole measurement hype in psychology started with intelligence; and early intelligence test were made for diagnosing mental abilities in children. More recently (1980ies onwards), Falmagne and Doignon developed so-called knowledge structures for qualitative assessment of skills, so dancing plus singing is not equal to 2. This stuff is more and more used in e-learning systems like ALEKS that some of our children have to do homework with. In education, they noticed that one cannot make numbers out of everything.
Once I asked my psychotherapist about my diagnosis. His response: "Why would you want to know? I only use it for the health insurance."
I am fervently interested in whether or not you (and others here) have opinions on so-called "positive psychology". For the uninitiated, positive psych is a comparatively recent (1960s or so) effort to use science to understand mental wellness instead of mental illness. In the nutshell explanation I usually use, it's the difference between fixing a blown knee versus training to break the 4-minute mile.
Most of the pos psych stuff promoted (and I do mean promoted) to layman is suspiciously indistinguishable from self-help snake-oil (keep a gratitude journal! don't let failures get you down! develop a growth mindset!), with the caveat that it is, by and large, snake-oil that appears to have, on average, replicable results - per its own (questionable IMO) measures - when people buy into it.
A lot of it is sold (and I do mean sold) as preventative medicine, but I think it fits great with my 10-minutes-old understanding of this taxometric distinction. If most people are not so much afflicted by a mental disease as they are uncomfortable or disabled due to their position out on the tail vis-à-vis anxiety levels, it makes sense that sometimes mental health-oriented boosters /could/ help just as much as, or more than, mental disease-oriented corrections.
The evidence is chaotic soup served with statistical smoke-and-mirrors, but with my methodological background, I've blamed it mostly on early commitment to suboptimal (proxy) measures - not so much where you draw the line between tall/short, but whether or not the yardstick is remotely objective or validly measuring the phenomenon you say it is. Now I'm going to be turning over in my head also whether or not it's a case of: "My guess is most professionals, and an overwhelming majority of laymen, are actually confused on this point, and this messes them up in a lot of ways."
Or am I overly skeptical?
Homelessness is probably categorical enough that "person experiencing homelessness" may be more appropriate than "homeless person" if you care just about sending the right signal.
(Since it takes more bits to explain disagreement than to indicate agreement, and positive reinforcement is useful when true, please take my disagreement here to indicate that any portion I _haven't_ disagreed with is a portion that I agree with.)
You say that "I think a lot of people still want psychiatry to deliver the single [cause of a given psychiatric phenomenon]. It’s not going to be able to do that." I don't think the fact that a given phenomenon is dimensional means that this is impossible; I would read it as meaning that the root cause must also be dimensional. To massively oversimplify the brain, let's say that all other things being equal an increase of X% in the average activity of one's Anxiety Gland produces an X% increase in one's reported anxiety levels, and that all medications which affect anxiety are ultimately things that have an impact of reducing Anxiety Gland activity by some amount. (The stuff that can impact the activity levels of the Anxiety Gland may be complex and multivariate, but I don't think this is different from categorical issues - if there are developmental failsafes so that N out of M mostly independent processes have to fail in order to produce a Categorically Abnormal Anxiety Gland, otherwise the development of a Normal Anxiety Gland is an attractor state, I'd still call Categorical Abnormality of the Anxiety Gland the "single cause", especially if it was what could be affected by interventions.) This would still be a single cause; it would just be a single dimensional cause.
You also say that "If most mental disorders are dimensional variation rather than taxa, that kind of makes the DSM look pretty silly, doesn’t it?...All of this is predicated on the idea that [there are] specific thing[s] called [disorders] that you either do or don’t have." I think another way to regard the DSM, or at least what an idealized DSM could be, is a manual of best practices for tradeoffs which are codified as acceptable at arbitrary cutoffs, much like your hypothetical Sanders tax plan is an arbitrary cutoff made for practical purposes. If a patient has more than X% excess Quality, then the profession regards it as reasonable to experiment with certain interventions which fall under the banner of "Interventions for Excess Quality Disorder", but below that cutoff they think it's not worth the risk.
In this case, it would be good and necessary to renegotiate the definitions of Excess Quality Disorder as new information came up, even if "Excess Quality Disorder" is an arbitrary cutoff. Let's say that new research comes in indicating that Intervention A for mild Excess Quality Disorder is more harmful than previously expected; in that case we might move the boundaries of Excess Quality Disorder to reflect that we want to be more careful with it. Or maybe we discover that while our cutoff was X%, people with (0.8 * X)% excess Quality have a reduction in quality of life greater than we previously thought, so we move the boundaries in the other direction. In this case, "Excess Quality Disorder" is perhaps an inferior category to "Eligible for Intervention {A,B...Z}", but may be worth using for convenience, communication, or research.
Autism, like height and other dimensional traits are influenced by 100s or more genes, each with a small influence up or down. So traits that are influenced by relatively few genes a larger amount each should be more categorical, e.g. schizophrenia (if my impression of schizophrenia is correct).
One challenge I've personally experienced with definitions for psychiatric diagnoses that are based on multiple dimensions is that people tend to attribute all the (categorical) symptoms to people with only enough of them to fit the definition.
My experience is with autism, or at least autism-spectrum disorders. People are diagnosed with autism based on possessing characteristics such as problems with social interaction and communication, and repetitive behaviour or restricted interests (think: Rain Man). My son was diagnosed as "on the spectrum" at a young age. This meant that teachers and other professionals would assume he had all the most common characteristics, whereas in fact he had only a few, and this meant their approach was often not appropriate at all.
This happens all over the place, of course: it's the problem of "natural kinds". All instances of Democrats, ice creams, and churches are not alike in their categories on all dimensions. Politics is often a difficult thing not least because it forces hard demarcations where in reality they're fuzzy (I vote for party X because it's X or Y, not because I believe everything that X profess to believe).
We cannot carve the world at its joints, because they're in our mind rather than in the world. The same is, perhaps unsurprisingly, true of the mind itself.
Thanks for the fine exegesis. I have practiced psychiatry for many a year. DSM III came out when I was in medical school. It sought to make psychiatric diagnosis, which was vague and overly influence by psychoanalytic theory, more objective, but using description and taking away theory. It was felt essential to make psychiatry a legitimately medical specialty, and at the time it was hailed as a triumph. Alas, they threw out the baby with the bathwater. All of the useful tools of personal history, development, personality and temperament were junked. But whatever has been gained from categories makes it harder to treat patients as individuals. (Just ask the insurance company flunkies who rule over the prior authorization domain, and decide how many sessions and which medications are "medically necessary.") DSM IV and V made matters even worse. Not only are the debates about where the cutoffs are for a disorder inane, political considerations have tainted and even overwhelmed whatever medical legitimacy the descriptive categories might have had.
I have long been reconciled that my profession will never be as objective as brain surgery or cardiology. No matter how you dress it up, psychiatry is as much art as science, and I like it that way. As much I look forward to advances of neuroscience and pharmacology, the DSM ain't science, and Big Data tells a lot of Big Fibs.
I appreciated the argument on person-centred language. I've not seen this argument brought up in this way vs "recovery oriented" language. From a medical model it probably makes more sense, particularly when charting, to not use person-centered language.
I’m surprised to hear that narcissistic personality disorder is generally considered by psychiatrists to be categorical rather than dimensional. I’d love to learn more here.
It intuitively seems like people vary on a spectrum in terms of how selfish, egotistical, etc. they are. A lot of people certainly seem to have subclinical levels of narcissism.
Is it that narcissistic personality disorder is something radically different from the personality trait normally referred to as narcissism?
I liked the mention of extremistan. It made me wonder—
*Are* the distributions of psychological variables (e.g. anxiety) fat-tailed (e.g. power law distributed instead of bell curves)? If so, that would maybe be relevant to whether and why people underestimate how anxious GAD people can be, etc.
A bizarre, very minor thing: when I load this page, it says "Jan 28" as the publish date, then a second later, it switches to "Jan 27" .
This is really interesting, but as someone who has ADHD, the idea that ADHD specifically is dimensional seems to directly contradict my experiences. The "symptoms" that get ADHD noticed and diagnosed tend to look like just having an unusual amount of trouble paying attention, the way Jeff Bezos has an unusual amount of money, but there's a slew of other things that have no reason to correlate with ADHD the way they do unless we consider them side effects of an underlying categorical difference.
If I just have an unusual amount of trouble concentrating, why does Adderall make me sleepy? Why does fidgeting help me think? Why are boredom and the perception of failure almost physically painful, in ways that confuses my non-ADHD friends when I describe them? Why do I share these traits with my roommate, who doesn't have difficulty concentrating at all, but has enough symptoms of hyperactive-type ADHD to also get a diagnosis? Why is there enough of a correlation between hyperactivity and inattention that the combined type is even recognized?
There's no reason inattention would correlate with these other traits without an underlying cause. A non-ADHD person who had a short attention span probably wouldn't also fall asleep after three cups of coffee or cry because they got mildly criticized at work.
I suspect some of the confounding findings have to do with the selection of variables that are included in the analysis and how they distribute through the general population in realms unrelated to the ‘condition’ being investigated. I don’t have the time to think through the implications, but further study of the efficacy of the statistical method for separating behavioral disorders is definitely warranted.
tl;dr: A secret variable does exist, but you can't read it. [we might be in the future, though]
You can't avoid false negatives, but in many cases, you can definitely be certain that you don't have a false positive.
[according to Dr. Russel Barkley, who is a famous ADHD researcher]
That is not quite right. There actually is such a variable and it can be 1.
ADHD is brain damage in five specific affected brain regions, that predictably results in a 30% delay of executive functions for your age. In 2/3rds of the cases, this is genetically determined.
And there are also acquired cases, where it's known that strep-bacteria causes an autoimmune overreaction that causes that brain damage.
And yes, ADHD is highly comorbid with other things. And yes, SCT might be another attention disorder, that's separate from it, is not as treatable (yet) and less well understood.
[and having SCT implies 50% chance of having ADHD as well]
And yes, other kinds of brain damage also can lead you to present with ADHD symptoms.
And all of the above is not universally known or accepted in psychiatric practice.
But ADHD is actually a taxon.
The variable exists and you might not be able to reliably say that someone is a 0.
[unless you find some other disease or environmental cause...... like a kid playing Fortnite all day in school, which is an epidemic in its own right]
After all, the space between the variable=1 population bump and the majority of the bell to its left is not empty.
BUT...... you can and should be quite certain sometimes, that you do have "the real deal" in front of you, by looking at the parents or grandparents and seeing them have it as well.
Also seeing the "paradoxical stimulant effect" (stimulants usually don't make you calmer) would provides evidence for a "1", as well.
source:
excellent playlist "30 essential ideas you should know about ADHD" by Dr. Barkley on recent research into ADHD:
https://www.youtube.com/watch?v=G2u8E5UqEHU&list=PLzBixSjmbc8eFl6UX5_wWGP8i0mAs-cvY&index=7&ab_channel=AdhdVideos
Concern about the height example: what if there is a height gene H which can be 0/1. Suppose both H=0 and H=1 lead to unimodal (Gaussian?) phenotype height distributions. If you have a mixture model of this reality then the overall population's height distribution could be unimodal (Gaussian) as well? How can you distinguish this reality from the reality where gene H doesn't exist?