Grouping-think
« previous post | next post »
According to a recent press release ("Scientists Have Established a Key Biological Difference Between Psychopaths and Normal People"),
Neuroscientists using MRI scans discovered that psychopathic people have a 10% larger striatum, a cluster of neurons in the subcortical basal ganglia of the forebrain, than regular people. This represents a clear biological distinction between psychopaths and non-psychopathic people.
The journal article (Choy et al., "Larger striatal volume is associated with increased adult psychopathy”) tells us that "Psychopathy was assessed using the PCL-R, which consists of 20 items rated by interviewers on a 3-point scale". (Wikipedia on PCL-R here). And from MRI scans, "segmentation of the caudate, putamen, nucleus accumbens, and globus pallidus was conducted together with the thalamus and cerebellum using standard FreeSurfer parcellation. Total striatal volumes were defined as the sum of the volumes of the four striatal subregions".
The generic plural "psychopaths" suggests a natural kind. And the phrase "a clear biological distinction" suggests well-defined and well-separated clusters of values on both neuro-anatomical and social-psychological dimensions. But what the researchers found was two weakly-correlated variables, each an amalgam of several measurements or evaluations, without any strong indication of clustering. Their Figure 3 (n=108):
So that press release and its mass-media uptake exemplify a cluster of fallacies that needs a name, and (as far as I know) doesn't have one. The key pieces:
- Thinking of distributions as points;
- Inventing convenient but unreal taxonomic categories;
- Forming stereotypes, especially via confirmation bias.
These fallacies are all natural and nearly inevitable aspects of human thought. And of course I'm not the first person to have noticed the elements of this cluster, or to have observed their interconnections. But does the cluster have a name? It should. Perhaps a commenter will inform us about an established term, but meanwhile I'm going to call it "grouping-think".
And putting aside the obvious cases of racial, ethnic, gender, and other sociological stereotypes, this way of thinking has been endemic in medicine for millennia. This starts because boxing the world up into natural kinds is often a useful path to understanding. And without convenient concepts and tools for dealing with continuous multi-variate distributions, what else are you going to do?
Of course some medical conditions belong in well-defined categories. But many behaviorally-defined disorders are "phenotypically diverse" — and a clinician friend explains that "phenotypically diverse is the Greek translation of 'we have no fucking clue'".
For phenotypically diverse disorders without clear etiology, it's increasingly clear that boxology is a mistake — especially now that we actually do have concepts and tools for dealing with complex and diverse multi-dimensional data.
According to The Hierarchical Taxonomy Of Psychopathology (HiTOP),
Objectives of the Hierarchical Taxonomy of Psychopathology (HiTOP) are to advance the classification of psychopathology to maximize its usefulness for research and clinical practice. The HiTOP aims to address limitations of traditional nosologies, such as the DSM-5 and ICD-10, including arbitrary boundaries between psychopathology and normality, often unclear boundaries between disorders, frequent disorder co-occurrence, heterogeneity within disorders, and diagnostic instability.
The HiTOP approaches these problems by conducting an empirical search for psychopathology structures starting from the most basic building blocks and proceeding to the highest level of generality: combining individual signs and symptoms into homogeneous components or traits, assembling them into empirically-derived syndromes, and finally grouping them into psychopathology spectra (e.g., internalizing and externalizing).
This approach reduces within-disorder heterogeneity by grouping related symptoms together and assigning unrelated symptoms to different components. It makes comorbidity an explicit and predictable feature of the model by classifying related syndromes together. Finally, it describes psychiatric phenomena dimensionally, addressing boundary problems and diagnostic instability. The HiTOP is not limited to dimensions, if evidence indicates existence of a natural boundary, this qualitative distinction will be incorporated in the model also.
And since 2009, NIH has been promoting the Research Domain Criteria (RDoC) initiative:
The RDoC framework provides an organizational structure for research that considers mental health and psychopathology in the context of major domains of basic human neurobehavioral functioning, rather than within established diagnostic categories. The framework currently includes six major functional domains (see figure). Different aspects of each domain are represented by three to six psychological/biological dimensions, or constructs, which are studied along the full range of functioning from normal to abnormal. Both behavioral and biological aspects of functioning change and mature throughout childhood/adolescence and across the life span, and so research on development is essential. Equally important is the study of various aspects of the environment, including the physical environment, cultural components, and factors such as social determinants of health. The RDoC framework encourages researchers to measure and integrate many classes of variables (units of analysis, e.g., behavioral, physiological, and self-report data) in order to seek a comprehensive understanding of the construct(s) under study.
But these efforts have a long way to go.
A few relevant past posts:
"Gabby guys: the effect size", 9/23/2006
"The Happiness Gap and the rhetoric of statistics", 9/26/2007
"The Pirahã and us", 10/6/2007
"Pop platonism and unrepresentative samples", 7/26/2008
"David Brooks, Social Psychologist", 8/13/2008
"Misleading pseudo-scientific argument of the week", 9/3/2009
"Mandatory treatment for generic plurals", 9/13/2009
"Physiological politics", 2/15/2010
"Icktheology", 2/18/2010
"Generic comparisons", 11/7/2011
"Generic plural nurses", 9/22/2018
"Group differences", 11/18/2018
"Lombroso and Lavater, reborn as fake AI", 10/22/2019
Cervantes said,
June 9, 2022 @ 6:54 am
Among other possible classifications, this is a reductionist fallacy. In general, psychiatric diagnosis consists of reifying laundry lists of largely subjective observations, with arbitrary cutoffs for ascribing positivity. They're often of the Chinese menu variety — two from column A and three from column B count, regardless of which two or three are taken from a larger list. Often, completely opposite symptoms are acceptable, e.g. agitation or lethargy, somnolence or insomnia. I don't know if there is an exact word for this, but the human tendency to see patterns in what is actually chaos is called pareidolia, though that usually refers to visual images rather than lists of phenomena..
bks said,
June 9, 2022 @ 7:29 am
Who among us is not neurodivergent?
Lucino said,
June 9, 2022 @ 7:48 am
Considering the cut-off point for psychopathy in the US is 30 on the PCL-R, it would appear that the volume is *slightly* higher for US-designated psychopaths in these data than the average, however if the cut-off point were 25 as it is in the UK, it becomes less convincing, not even accounting for how loose the distribution is on the rest of the graph.
I have to agree that drawing conclusions from this study is a misstep, as I would say for any attempt to designate human-defined psychiatric disorders as any sort of "natural kind". Psychopathy is a bit outside my area of knowledge, but during my trip through the mental health system I was told to research into the condition I expecting a diagnosis for. I don't still have the papers I read for that, but from what I remember, there was a good bit of doubt cast on the idea that this disorder was its own standalone thing, with a newer idea being to consider a spectrum of symptoms all stemming from the same core issue (what they often called ACEs — Adverse Childhood Experiences), which would encompass overlapping symptoms of depression, anxiety, personality disorder (Borderline in particular), dissociation, and psychosis.
From this, and the common saying that with mental illness, you rarely stop at just one, I would imagine the future of the field of psychiatry being in this direction of multi-dimensional data analysis and large clusters and/or spectra of disorder rather than traditional boxologies – as the DMS-V. It appears that we really do "have no f***ing clue" about much of human psychiatry for the time being.
Athanassios Protopapas said,
June 9, 2022 @ 9:16 am
It's not just about psychiatric conditions. Most of the literature on dyslexia (which is not a medical or psychiatric issue) is like that. "Dyslexics" have all sorts of "deficits" (cognitive, educational, psychological, biological, you name it), yet the actual empirical picture looks exactly like this graph.
wanda said,
June 9, 2022 @ 9:30 am
I might call that phenomenon "essentialism with boundary intensification." Essentialism is the tendency to believe that certain categories are "natural kinds" defined by some sort of unobservable "essence." "Boundary intensification" is the tendency to believe that things that are grouped as different "natural kinds" are way more different than they actually are. In the case of this press release, the "natural kinds" would be "normal people" and "psychopaths," who are presented as completely different kinds of people, and the paper is presented as a search for this "essence" that separates the two kinds of people. Cognitive psychologists and science educators have found evidence that the human tendency to "essentialize through boundary intensification" impedes people's understanding of evolution, genetically modified organisms, and racial groups.
KeithB said,
June 9, 2022 @ 9:57 am
obligatory xkcd:
https://xkcd.com/2048/
mg said,
June 9, 2022 @ 10:50 am
As far as this statistician is concerned it's mainly an artifact of the desperation to get a publishable finding, any finding, engendered by "publish or perish" combined with "we don't need to spend money on a methodologist/statistician who knows what they're doing – we have easy software". This is compounded by journals not routinely having statistical review for potentially accepted articles (you wouldn't believe some of the things I've seen submitted).
Jerry Packard said,
June 9, 2022 @ 1:03 pm
@Athanassios Protopapas
While I would take with many large grains of salt the validity of the psychopathy scale and its purported relation to the corpora striata, as far as dyslexia is concerned, our work in Chinese dyslexia has shown a direct relation with cognitive deficits – most notably deficits in morphological awareness – with graphs that, even in the aggregate, look nothing like the one presented (Wu, S., Packard, J., Shu, H. 2009. Morphological Deficit and Dyslexia Subtypes in Chinese. In Law, Weekes and Wong, _Language Disorders in Speakers of Chinese_. New York: Multilingual Matters. 112-137).
Brett said,
June 9, 2022 @ 2:00 pm
Quoting myself:
AntC said,
June 9, 2022 @ 4:12 pm
Figure 3 (n=108)
Is the most significant statistic here: no generalisation is meaningful from a sample of 108.
(Also, those were all males.)
Andreas Johansson said,
June 10, 2022 @ 12:29 am
@Jerry Packard:
What's morphological awareness?
(I've never been formally been diagnosed with dyslexia, but as a kid I certainly had far more than my fair share of trouble learning to read and write.)
Kristian said,
June 10, 2022 @ 12:01 pm
I’m not a psychiatrist but things I read written by psychiatrists generally acknowledge that the diagnostic categories are fuzzy. It’s not as though they think there’s some kind of intensified boundary between someone with just enough symptoms of depression to get the diagnosis and someone with not quite enough.
Rosie Redfield said,
June 10, 2022 @ 2:08 pm
The distribution of values along the 'Psychopathy' axis is also a bit suspect. Rating 20 factors, each on a 3-point scale, should give integer values between 0 and 40 (the scale provided), but many of the plotted values appear to not be integers.
David Morris said,
June 10, 2022 @ 5:10 pm
I am intrigued by several things: 1) the line at 19-ish, which has nine individuals with striatal volumes ranging from approx 175000 to approx 28000 – a difference of more than 50%; 2) the two outliers at 25-ish (the highest volume but a moderately high psychopathy score) and at 28-ish (who has a volume half of that of the former, and the lowest overall, but a higher psychopathy score); and 3) the two individuals with the lowest volumes, who have such different scores.
Jerry Packard said,
June 10, 2022 @ 7:12 pm
@Andreas Johansson
Morphological Awareness is a person's ability to analyze and perceive morphemes. This ability can be measured, and people show substantial variation on this ability.
An English speaker with poor morphological awareness might have a hard time perceiving that the meaning of 'bank' in the word 'banker' is different from the meaning of 'bank' in the word 'riverbank', or that the meaning of 'corn' in the word 'cornbread' is different from the meaning of 'corn' in the word 'corner'.
A Chinese speaker with poor morphological awareness might have a hard time perceiving that the meaning of 'xīn1' in the spoken word 'xin1zang4' (心脏) ‘heart’ is different from the meaning of 'xin1' in the spoken word 'xin1wen2' (新闻) ‘news’ or the meaning of 'xin1' in the spoken word 'xin1ku3' (辛苦) ‘work hard.’
In English, dyslexia is phonologically based – most people with dyslexia have deficits in phonological awareness. In Chinese, dyslexia is less phonologically based and more morphologically based – most people with dyslexia in Chinese have deficits in morphological awareness.
Athanassios Protopapas said,
June 13, 2022 @ 2:07 am
@Jerry Packard
You seem to be referring to a non-peer reviewed chapter in a 2008 book that is not publicly accessible. Why don't you just post the relevant graph for everyone to see? Better yet, if there is in fact a true qualitative deficit (in anything), why not point at papers that demonstrate it?
My answer: Because there is none. Dyslexia itself is well understood to be of the "graded"/"continuous" kind, with an arbitrary criterion/cutoff, like high cholesterol or high blood pressure (doi:10.1146/annurev-clinpsy-032814-112842). We are talking about the low end on a continuum of reading skill. And reading skill is associated with a lot of other things, producing graphs like the one posted here in large unselected samples that haven't been artificially pre-grouped into categories based on the arbitrary cutoffs.