Abstract
In 1690 William Petty founded political arithmetic by keeping for his calculations only causes with visible foundations in nature and leaving the mutable minds, opinions, appetites, and passions of particular people to others. For almost three centuries, statistics of mental illness respected that boundary, counting admissions, discharges, and deaths while the minds themselves remained objects of interpretation. This article argues that the third edition of the Diagnostic and Statistical Manual of Mental Disorders, published in 1980, reversed Petty’s exclusion, and it specifies how. Leaving mental distress as responsive to its engagement as ever, DSM-III performed five operations that made the record of distress behave as if it were nonrecursive: it turned diagnosis into a count, turned trajectories into bounded episodes, removed the context that made distress intelligible, separated entangled predicaments into discrete units, and redesigned the clinical encounter so that its output would not depend on who conducted it. Structured interviews then removed the clinician altogether, and the resulting counts became the foundation of global mental health. The reversal went unnoticed because it presented itself as a solution to a problem of reliability rather than as a decision about what kind of thing a mental disorder is.
I. Introduction: Petty’s Exclusion and Its Reversal
In the preface to Political Arithmetick, William Petty explained that he would express himself “in Terms of Number, Weight, or Measure; to use only Arguments of Sense, and to consider only such Causes, as have visible Foundations in Nature; leaving those that depend upon the mutable Minds, Opinions, Appetites, and Passions of particular Men, to the Consideration of others” (Petty 1690). The sentence is usually quoted for its first half, as a manifesto of quantitative method. Its second half is more revealing. Petty set minds and passions aside without declaring them unimportant, because they were mutable, because they changed in response to how they were engaged, and he kept for his arithmetic the land, the harvests, and the deaths that did not. The founder of political arithmetic drew, at the very beginning of quantitative social science, the line between what answers back and what does not, and he placed his discipline on one side of it (Ecks 2026a).
For nearly three hundred years the statistics of mental illness kept to Petty’s side of that line, even when they appeared to cross it. Asylums counted admissions, discharges, and deaths. Censuses counted the insane and the idiotic. In 1918 American asylum physicians published a Statistical Manual so that institutions could report comparable figures (American Medico-Psychological Association and National Committee for Mental Hygiene 1918). What these figures counted were institutional events: a person entering or leaving a building, a name on a register, a death in a ward. Each was a bounded, datable mark that no longer responded to anything once it had been entered. The minds of the people concerned, their moods, appetites, fears, and passions, were the business of physicians who interpreted them, and in the middle decades of the twentieth century that interpretation was dominated in the United States by psychoanalysis, for which a symptom was a sign whose meaning lay in a personal history.
In 1980 this arrangement changed. The third edition of the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders replaced brief descriptive paragraphs with explicit criteria for each disorder, adopted what its makers called an atheoretical, descriptive approach to aetiology, and listed 265 diagnoses (American Psychiatric Association 1980). Within a few years, structured interviews based on its criteria allowed interviewers without clinical training to establish, in households across the United States and then across the world, whether a respondent met the criteria for a mental disorder. Minds, opinions, appetites, and passions, the very items Petty had excluded, became countable in whole populations. Appetite itself became a diagnostic criterion of major depression.
The history of DSM-III is well documented. Hannah Decker has reconstructed how the manual was made, Mitchell Wilson and Rick Mayes and Allan Horwitz have traced its professional and economic background, Stuart Kirk and Herb Kutchins have analysed the rhetoric of its reliability claims, and Horwitz and Jerome Wakefield have shown how its decontextualized criteria turned normal sorrow into depressive disorder (Decker 2013; Wilson 1993; Mayes and Horwitz 2005; Kirk and Kutchins 1992; Horwitz and Wakefield 2007). Rather than adding new archival findings to that literature, this article asks a different question: what kind of operation was performed on mental distress in 1980, considered as a living process? The answer proposed here is that DSM-III reversed Petty’s exclusion by making the record of distress behave as if it were nonrecursive, while the distress itself remained as responsive as it had ever been. The article specifies five operations through which this was achieved, follows the criteria from the clinic into the population survey, and explains why so consequential a decision about what kind of thing a mental disorder is passed almost unnoticed.
II. Recursive, Nonrecursive, and Mixed
A process is recursive when its subsequent course is altered by how it is engaged, so that it answers back. A kidney stone is indifferent to its name. It can be called a calcium oxalate deposit, photographed, described in a poem, or counted in a national register, and it will not reorganize itself in response. A person who is told that she is depressed is not indifferent in this way. The diagnosis enters how she understands her own tiredness, how she presents in the next consultation, what she expects of treatment, how her family responds to her, and how she reads her own good and bad days. What is engaged changes because it has been engaged.
Mathematics is a nonrecursive symbolic system. A function returns the same value for the same argument however often it is evaluated, and a rule, once written, does not revise itself because someone has read it. A recursive process as such cannot be modelled mathematically, because every model is itself a nonrecursive deposit. Complex living processes, however, are rarely recursive in every respect. They have mixed recursivity: they combine aspects that answer back to their engagement with aspects that do not. The hypothesis on which this article builds is that every mathematical treatment of a complex living process, including models of feedback and learning, works by isolating the nonrecursive aspects (Ecks 2026a). The recursion is always located in the modellers and the modelled, never in the model.
Two further distinctions matter for what follows. The first is between a process and its record. A death is part of a living history; the entry “consumption” in a seventeenth-century bill of mortality, once written, no longer changes in response to anything the family later learns. Statistics begins with such records. Their stability belongs to the record and implies nothing about the process that left it. The second distinction concerns recursivity levels. Living Value Theory distinguishes coordinated living that proceeds without explicit attention (L1), felt misalignment that has not yet been articulated (L2), articulation in words, gestures, and other symbolic forms (L3), the stabilization of articulations into classifications that can be used for institutional action (L4), and reflection on how all of these are handled (L5). A diagnosis is a paradigmatic L4 form. It takes articulations of felt misalignment and stabilizes them into a category that allows referral, prescription, reimbursement, and communication between strangers (Ecks 2026b).
Mental distress is a domain of unusually pronounced mixed recursivity. Some of its aspects do not answer back to their engagement: a night without sleep has happened once it has happened, and the weight a person has lost can be read from a scale. Most of what matters in it does. Distress is recognized, named, and responded to, by the person and by everyone around her, and each recognition, naming, and response becomes part of what the distress is next. The question this article asks of DSM-III is therefore precise. Which aspects of this mixed process did the manual’s criteria isolate, by what operations, and what happened to the aspects that answer back?
III. Before 1980: Counting Admissions, Interpreting Minds
The first two editions of the Diagnostic and Statistical Manual, published in 1952 and 1968, were slim documents. Their definitions were brief paragraphs that described what a typical patient might be like, and many of their categories carried the language of reaction and neurosis, which located a disorder in a person’s attempt to cope with circumstances or with conflicts whose meaning had to be worked out case by case (Mayes and Horwitz 2005). Whatever one thinks of the psychodynamic theories that informed them, these manuals assumed that a symptom was intelligible only within a life. Two patients who both reported sleeplessness and despair might have quite different disorders, or none, depending on what the sleeplessness and despair were about.
This assumption had a statistical consequence. Outside the institution, there was nothing for the manuals to count. Psychiatric epidemiology before 1980 relied either on treated cases, which counted the people who reached a hospital or clinic, or on community surveys that measured general distress and impairment on rating scales without assigning specific diagnoses. Psychiatry had numbers, and in the asylum era it had many. But they were numbers of institutional events, and they remained on Petty’s side of his own line. The mutable minds were interpreted, one at a time, by clinicians.
By the early 1970s this arrangement was in crisis. Comparative studies showed that psychiatrists in New York and London diagnosed the same patients very differently, with American psychiatrists applying the label of schizophrenia far more widely than their British colleagues (Cooper et al. 1972). Robert Spitzer and Joseph Fleiss reanalysed the existing reliability studies and concluded that agreement between clinicians on most diagnostic categories was poor (Spitzer and Fleiss 1974). In 1973 David Rosenhan published an account of healthy volunteers who had been admitted to psychiatric hospitals after reporting that they heard voices, and who were kept there, diagnosed and medicated, for weeks after they stopped reporting anything unusual (Rosenhan 1973). The same year, the American Psychiatric Association voted to remove homosexuality from its list of disorders, a decision in which Spitzer played a central part and which showed, to admirers and critics alike, that psychiatric categories could be changed by vote (Bayer 1981). Insurers, government agencies, and a growing pharmaceutical research sector wanted to know what exactly they were paying for, and psychiatry’s standing within medicine was under pressure from all sides (Mayes and Horwitz 2005; Wilson 1993).
An alternative was already available. At Washington University in St. Louis, a group of psychiatrists who had kept their distance from psychoanalysis had spent the late 1960s developing explicit criteria for a small number of disorders they believed to have been validated by research. Their paper, published in 1972 with John Feighner as first author, set out operational criteria for fifteen conditions (Feighner et al. 1972). Spitzer, working with Jean Endicott and Eli Robins of the St. Louis group, expanded these into the Research Diagnostic Criteria for a study funded by the National Institute of Mental Health (Spitzer, Endicott, and Robins 1978). The historical record shows that this small group instigated a change of direction that has had profound effects on American and, ultimately, world psychiatry (Kendler, Muñoz, and Murphy 2010). When the American Psychiatric Association appointed Spitzer to lead the task force for the third edition, the method of the research criteria became the method of the official manual.
IV. The Unreliability Was Data
The official story of DSM-III begins with the reliability crisis and presents the manual as its solution. Two clinicians who examined the same patient arrived at different diagnoses; explicit criteria made them arrive at the same one. Reliability was a precondition of validity, since a category that clinicians could not apply consistently could not be studied, and a category that could not be studied could never be shown to correspond to anything. On this account DSM-III was a methodological reform, modest in its ambitions and neutral with respect to theory.
The story is accurate as far as it goes, and the reform had real benefits. But it rests on an interpretation of what the unreliability meant, and that interpretation was never argued for. The crisis studies treated disagreement between clinicians as error: noise around a true diagnosis that a better procedure would reveal. Yet the disagreement had a structure. It varied with the clinician’s training, theoretical orientation, and national tradition, with the setting in which the interview took place, and with what the patient said in response to what the clinician asked. Rosenhan’s pseudopatients were diagnosed in part because they presented themselves in a particular way to particular people in a particular institution, and they were kept in hospital in part because the label, once given, reorganized how everything they did was read. These are exactly the features one would expect if the diagnostic encounter were interrecursive, a meeting of two parties who each respond to the other’s responses, and if what was being diagnosed were itself changing in response to being engaged.
From this perspective, the unreliability of pre-1980 diagnosis was evidence about the domain as well as a defect of procedure. It showed that psychiatric diagnosis, as then practised, was a recursive engagement with a recursive process, and that its outcome depended on who engaged whom, where, and how. That evidence could have prompted a question about what kind of object a mental disorder is, and how a classification ought to treat an object that answers back. It prompted instead a redesign of the engagement so that its output would no longer depend on who conducted it. The recursivity that the crisis had made visible was classified as noise and engineered out of the record.
This is the first sense in which DSM-III reversed Petty’s exclusion. Petty had recognized that minds were mutable and declined to count them. The reliability crisis demonstrated, in its own statistical vocabulary, that diagnostic judgements about minds were mutable too. DSM-III responded by building a procedure in which the mutability would no longer appear, and then counted.
V. Five Operations
What exactly did that procedure do? It is possible to distinguish five operations, each of which can be read off the manual itself. None of them made mental distress less recursive, since no classification can alter the composition of the process it classifies. Each of them made some feature of the diagnostic record behave as if it belonged to a nonrecursive object.
A. From interpretation to count
The first operation turned diagnosis into an arithmetic act. In DSM-III, a major depressive episode required a dysphoric mood or a loss of interest or pleasure in almost all usual activities, together with at least four of eight further symptoms: poor appetite or weight loss, or increased appetite or weight gain; insomnia or hypersomnia; psychomotor agitation or retardation; loss of interest or decreased sexual drive; loss of energy or fatigue; feelings of worthlessness or excessive guilt; diminished ability to think or concentrate; and recurrent thoughts of death or suicide (American Psychiatric Association 1980). The criteria were polythetic. No single symptom beyond the mood criterion was required, and two people could meet the criteria while sharing only one feature. What made a case was a number reaching a threshold.
The significance of this lies less in the particular list than in the form. Before 1980, a clinician deciding whether a patient was depressed asked what the patient’s state meant, in this life, at this point. After 1980, the clinician asked how many of the listed features were present. The first question requires an interpretation that answers back to everything the clinician learns, including the patient’s response to the question. The second requires a tally. A tally is a nonrecursive operation: four is four however often it is counted and whoever counts it. The diagnostic act itself had been given the form of a function from a set of present or absent features to a category.
It is worth noticing what the list contains. Appetite is its first item. Petty excluded “Appetites” from political arithmetic because they were mutable. DSM-III made a change of appetite countable as one feature among eight, to be recorded as present or absent. The appetite did not become less mutable. Its mutability ceased to matter for the record, because the record asked only whether a change had occurred.
B. From trajectory to episode
The second operation turned a trajectory into a bounded state. The criteria required that the symptoms be present nearly every day for at least two weeks. Distress has no natural boundaries of this kind. It rises and falls, recedes and returns, varies with season and circumstance, and is continuous with the ordinary sorrows and exhaustions of living. A duration threshold cuts this continuity into episodes: intervals with a beginning and, in principle, an end, which can be dated, counted, and compared.
Episodes are what epidemiology needs. Prevalence is the proportion of a population in a given state during a given period; incidence is the rate at which new episodes begin. Both require states with edges. The two-week threshold supplied the edge. It did so by fiat, since there is no evidence that distress lasting fifteen days differs in kind from distress lasting twelve, and later research has found that people who have shorter episodes, or fewer symptoms than the criteria require, are at elevated risk of later depression (Kendler and Gardner 1998). The threshold made the episode a nonrecursive unit of record. A person either met the duration criterion during the period in question or did not, and once recorded, the fact did not change, whatever happened next. The trajectory, which is where recovery, relapse, and adaptation occur, became the space between records.
C. From context to symptom
The third operation removed the context in which distress is intelligible. Earlier psychiatry had distinguished depression that arose from within, often called endogenous, from depression that was a reaction to loss, failure, or adversity. The distinction was contested, but it embodied the assumption that the meaning of a symptom depended on what it was a response to. DSM-III abandoned this assumption. With a single exception, its criteria for major depression asked only whether the features were present, and never why a person was sad, sleepless, or unable to concentrate.
The exception is revealing. DSM-III excluded uncomplicated bereavement from the diagnosis, so that a person who met the symptom criteria after the death of someone close was not, for a limited period, to be counted as depressed. No equivalent exclusion was made for the loss of a job, the end of a marriage, a diagnosis of serious illness, or any other major adversity. This asymmetry between the death of a loved one and every other loss has attracted much subsequent research and argument (Wakefield et al. 2007; Kendler et al. 2008). Horwitz and Wakefield argued that the decontextualized criteria confused intense normal sadness with disorder and produced a seeming epidemic of depression (Horwitz and Wakefield 2007). Spitzer, the manual’s principal architect, wrote the foreword to their book.
In the terms used here, the bereavement exclusion is the one place where the manual acknowledged that what a person’s distress is depends on what it is engaged with. Everywhere else, the criteria treated a symptom as a feature that is either present or absent independently of the life in which it occurs. That is the definition of a nonrecursive aspect: something that does not change in response to how it is engaged. By stripping context, the manual made the symptom behave as one. The surviving exception shows that its makers knew the assumption could not hold everywhere, and the fact that it survived alone shows how much had been given up. When the bereavement exclusion was itself removed in 2013, the last trace of the old assumption disappeared from the criteria for major depression.
D. From predicament to units
The fourth operation divided entangled predicaments into discrete units. DSM-III was built of categories, each with its own criteria, and it originally used hierarchical exclusion rules to manage their overlap: one diagnosis was not to be made if its features were “due to” another disorder. These rules soon proved difficult to sustain. Using data from a community survey of more than eleven thousand people, Jeffrey Boyd and colleagues found that the presence of any disorder increased the odds of having almost any other disorder, including disorders that the manual did not treat as related (Boyd et al. 1984). Many of the hierarchical exclusions were dropped in the revised edition of 1987, and the term comorbidity entered general psychiatric use to describe the resulting co-occurrences.
The sequence is instructive. People in distress rarely present with one problem. Low mood, worry, sleeplessness, pain, drinking, and the difficulties of work, money, and family come together and act on one another. A classification of discrete categories must either decide which of these is primary, which is what the hierarchical rules attempted, or record each as a separate disorder that happens to co-occur with the others. Either way, the predicament is divided into units that can be counted one by one. The units are nonrecursive in the relevant sense: a recorded diagnosis of generalized anxiety disorder does not change in response to a recorded diagnosis of major depression in the same person. The living predicament, in which the worry feeds the sleeplessness and the sleeplessness the despair, is precisely the kind of interaction between aspects that a set of separate units cannot contain. Comorbidity is the statistical name for the trace that this interaction leaves when it has been divided into units.
E. From encounter to procedure
The fifth operation is the one on which the others depend. It redesigned the diagnostic encounter so that its output would not depend on who conducted it. The measure of success was interrater reliability, expressed as the kappa coefficient, which corrects the proportion of agreement between two raters for the agreement expected by chance. In the first phase of the DSM-III field trials, the overall kappa for axis I diagnoses of 281 adult patients was 0.78 when two clinicians attended the same interview and 0.66 when they interviewed the patient separately, figures that were described as higher than those previously achieved (Spitzer, Forman, and Nee 1979). The trial’s organizers, following Fleiss, treated kappas above 0.7 as good agreement (Spitzer, Williams, and Endicott 2012). The pattern within these figures is itself telling. Agreement was highest when both clinicians heard the same answers in the same interview, and it fell when each met the patient separately, which is to say when the patient had the opportunity to respond differently to different people. For the personality disorders, the categories most explicitly concerned with how a person characteristically engages with others, agreement was lower still. The more a category depended on what happens between people, the harder it was to make its record independent of who was in the room.
Reliability in this sense is a property of the relation between records. It shows that two people who apply the same procedure to the same patient tend to write down the same category. It says nothing about the patient, and it is achieved by making the procedure as independent as possible of the interrecursive features of the encounter: the particular rapport between this clinician and this patient, the clinician’s sense of what lay behind an answer, the patient’s response to being asked. The more fully specified the procedure, the more the encounter resembles the evaluation of a function, and the less the outcome depends on who meets whom.
This is the precise sense in which DSM-III made the diagnostic record behave as if it were nonrecursive. The patient remained recursive, and so did the clinician. What was engineered was an encounter whose recursive features would leave as little trace as possible in its output. Reliability was then taken as evidence that the procedure was measuring something stable. But the stability had been built into the procedure, and the procedure had been built to produce it.
VI. From Manual to Survey: How the Clinician Was Removed
Had the operations of DSM-III remained confined to the clinic, they would have changed psychiatric practice without necessarily changing what could be counted in populations. The decisive step came when the criteria were turned into a questionnaire.
Even before DSM-III was published, Lee Robins and her colleagues at Washington University were developing the Diagnostic Interview Schedule, an instrument designed to allow lay interviewers, after brief training, to ask all the questions needed to evaluate most of the disorders in what was about to become the standard American nomenclature (Robins et al. 1981; Robins 1994). It was first used in the Epidemiologic Catchment Area program of the National Institute of Mental Health, which surveyed some twenty thousand people in five American cities and their surroundings in the early 1980s (Robins 1994). For the first time, the prevalence of specific mental disorders, defined by explicit criteria, could be estimated in a general population.
The interview completed what the manual had begun. In the clinic, DSM-III criteria were still applied by a clinician who saw the patient and could judge whether a reported symptom was clinically significant. In the survey, a trained interviewer read fixed questions in a fixed order, recorded the answers, and passed them to a scoring algorithm that determined whether the criteria had been met. The clinician’s judgement, the last recursive element in the diagnostic act, had been replaced by a procedure that could be run by anyone and scored by a computer. The diagnosis had become, in the most literal sense, a mathematical function applied to a set of recorded answers.
The structured interview also changed who could be counted. A clinical diagnosis required a person to arrive at a clinic, which is itself a decision shaped by symptoms, money, family, stigma, and access to care. A household survey required only that a person open the door and answer. For the first time, the people whose distress had never become a clinical event entered the record, and psychiatric epidemiology could report that most people who met criteria for a disorder had received no treatment for it. This finding, later called the treatment gap, became one of the most consequential numbers in mental health policy. It rested on the assumption that meeting the criteria of a lay interview and needing treatment were the same thing.
The question of whether the function captured what clinicians recognized as disorder was examined at once, and the results were sobering. When psychiatrists re-examined respondents in the Baltimore site of the survey, prevalence estimates based on the lay interview differed significantly from those based on psychiatric examination, and the people identified as cases by the two methods were often different people (Anthony et al. 1985). The investigators concluded that agreement was never high enough to encourage the confident replacement of a psychiatric diagnosis with one derived from the interview (Folstein et al. 1985). Other comparisons, in other settings, produced better and worse agreement.
These findings did not halt the method’s spread. In 1988 the Composite International Diagnostic Interview, written at the request of the World Health Organization and the United States Alcohol, Drug Abuse, and Mental Health Administration, combined the American instrument with questions drawn from European diagnostic traditions, fully structured to allow administration by lay interviewers and scoring of diagnoses by computer (Robins et al. 1988). Its descendants were used in the National Comorbidity Survey in the early 1990s and in the World Mental Health Surveys, which administered the instrument in dozens of countries (Kessler et al. 1994; Kessler and Üstün 2004). The estimates of prevalence that these surveys produced became the inputs from which the Global Burden of Disease calculated the burden of depressive and anxiety disorders, and the burden estimates became the founding evidence of the movement for global mental health (Murray and Lopez 1996).
The chain is therefore continuous. Petty excluded the mutable minds from political arithmetic in 1690. DSM-III, in 1980, gave them criteria that could be counted. The structured interview, in 1981, made the counting independent of clinicians. The international interview, in 1988, made it independent of place. The Global Burden of Disease, from 1996, converted the resulting prevalence into disability-adjusted life years and ranked depression among the leading causes of disability in the world. Each step depended on the one before it, and each extended the reach of a procedure designed, at the outset, to make the record of a recursive process behave as if it were nonrecursive.
The numbers that resulted are widely divergent. Surveys using the same nominal instrument have produced twelve-month prevalence estimates for major depression ranging, across countries, from fractions of a percent to around ten percent, and even the most carefully harmonized surveys show a severalfold range (Bromet et al. 2011). Some of that variation reflects real differences in living conditions. Much of it reflects differences in how distress is recognized, named, admitted to a stranger, and translated into answers to fixed questions in another language. Those are the recursive aspects of the process. Standardizing the interview left them in place, and they reappear in the estimates as variance that no one can apportion.
VII. What the Criteria Do Once Released
A classification of nonrecursive objects can be released into the world without changing the objects. A list of minerals does not alter the minerals. A classification of recursive beings cannot be released without becoming part of what it classifies. Ian Hacking described this as the looping effect of human kinds: people who are classified come to understand themselves through the classification, change their behaviour accordingly, and thereby change the kind, which must then be reclassified (Hacking 1995). Hacking’s account captures one important loop, the passage from a classification to self-understanding and back to the classification. The DSM-III criteria entered many more.
The criteria were designed as a clinician’s tool. Within two decades they had become a public vocabulary. Screening questionnaires were built directly on them. The Patient Health Questionnaire, whose nine depression items correspond to the nine criteria of the fourth edition of the manual, was developed by Kurt Kroenke, Spitzer, and Janet Williams for use in primary care and is now administered by general practitioners, employers, universities, and smartphone applications around the world (Kroenke, Spitzer, and Williams 2001). Campaigns against stigma taught people to recognize the criteria in themselves and others. Pharmaceutical advertising described them. Young people grew up with them as part of the ordinary language in which they articulated how they felt.
Each of these is an uptake of a nonrecursive deposit by recursive beings. The criteria do not change because they are read; the readers do. A person who has learned that poor sleep, low energy, and difficulty concentrating are signs of depression will attend to her sleep, energy, and concentration differently, articulate her felt misalignments in those terms, and present them to a doctor or a survey in that form. The articulation is real, and it may be helpful. But it is shaped by the classification it will then be measured against. The next survey, using questions derived from the same criteria, will record answers from people whose way of recognizing and naming their own states has been formed, in part, by the criteria and the earlier surveys.
This is one reason why the trend lines of depression are so difficult to interpret. Reported rates have risen in several of the countries that have invested most in recognizing and treating depression, which is the reverse of what one would expect if recognition and treatment were reducing a fixed quantity of disorder. Several explanations are possible, and they are not mutually exclusive: more distress, more recognition of the same distress, broader criteria, more willingness to report, more people who have learned to articulate their distress in the criteria’s terms. The data alone cannot decide between them, because the data are produced by the same encounter that the criteria have reshaped. An instrument designed to make the record nonrecursive cannot register the recursive effects of its own circulation. Those effects return as changes in the thing measured, and they are counted as changes in prevalence.
The effect is most visible among young people. A generation has grown up with the criteria as part of everyday language, learned in school assemblies, on social media, and in the screening questionnaires of university health services. The distress they report is real. But the forms in which it can be recognized, the words in which it can be said, and the thresholds at which it is taken to require help have all been shaped by a classification designed, in 1980, for research on hospital patients. When a survey of young people finds that a quarter or a third meet criteria for anxiety or depression, it measures distress and the circulation of the criteria at once, and it has no way of separating the two.
Nothing in this argument implies that the distress is unreal or that the people who report it are mistaken. The point concerns what the numbers can establish. A rising prevalence of kidney stones in a population can be read as a rising occurrence of kidney stones, because no one forms more stones by learning about them. A rising prevalence of depression, measured by criteria that have entered the language in which people recognize and articulate their states, cannot be read in the same way, because the measure is part of the process it measures.
VIII. Why It Went Unnoticed
A change of this magnitude might have been expected to provoke an explicit debate about what kind of thing a mental disorder is. There were vigorous debates about DSM-III, about its neglect of psychodynamic understanding, its treatment of particular categories, its political origins, and its relationship to the pharmaceutical industry. What did not take place was a debate about the ontological status of what the manual had made countable. Three features of the reform help explain why.
The first is the language of theoretical neutrality. The task force presented DSM-III as atheoretical with regard to aetiology, a descriptive nosology that took no position on the causes of the disorders it described (American Psychiatric Association 1980; Frances and Cooper 1981). Abstaining from claims about causes sounded like modesty. But a classification that defines a disorder by the presence of listed features, without regard to what they are a response to, has taken a position on what the disorder is: it is the features. Kenneth Kendler has argued that this position became entrenched over the following decades. Comparing the DSM criteria for major depression with descriptions in textbooks published between 1900 and 1960, he found that several features those textbooks described were absent from the criteria, and he warned that psychiatry had moved toward treating the criteria as constituting disorders rather than indexing them (Kendler 2016). The atheoretical stance concealed an ontological commitment by presenting it as the absence of one.
The second is the virtue of reliability. Reliability is a real good. Clinicians who cannot agree on what they are treating cannot compare their results, and patients who receive different diagnoses from different doctors are poorly served. Because reliability was so obviously desirable, the operations that achieved it appeared as improvements in method rather than as decisions about the domain. The question of what had been held fixed in order to produce agreement, and what had been lost by holding it fixed, was rarely asked. Kirk and Kutchins showed how far the reliability claims themselves rested on rhetoric (Kirk and Kutchins 1992). The deeper point is that even perfect reliability would show only that a procedure produces consistent records. It would show nothing about whether the thing recorded behaves consistently.
The third is institutional reinforcement. Once explicit criteria existed, a great many institutions found them useful. Insurers and government programmes could pay for the treatment of defined disorders. Pharmaceutical trials required entry criteria for their participants, and since the Kefauver-Harris amendments of 1962 American regulators had required evidence of effectiveness for a drug’s claimed indication, so that a defined disorder became the unit for which a drug could be tested, approved, and marketed. Research funders could support studies of homogeneous diagnostic groups. Epidemiologists could estimate prevalence. Each use confirmed the categories and created further reasons to keep them stable (Mayes and Horwitz 2005; Wilson 1993). The categories became infrastructure, and infrastructure rarely invites questions about the nature of what it carries. This is the pattern of mathematical selection: aspects of living processes that can be bounded and recorded accumulate evidence, institutions, and resources faster than aspects that cannot, until their prominence is taken for their importance (Ecks 2026a).
IX. What a Recursion-Literate Classification Would Require
None of this is an argument for abandoning psychiatric classification or for ceasing to count. Diagnoses allow people in distress to be referred, treated, paid for, and studied. Population surveys revealed that much suffering never reached any clinic, and the burden estimates that followed gave a neglected field an argument it had lacked. The question is whether classification and counting can be practised in a way that acknowledges what kind of process they are applied to. Four requirements follow from the argument of this article.
The first is to state the composition. For any category, it is possible to ask which of its aspects answer back to how they are engaged and which do not, and how much of what matters each carries. A disorder whose decisive aspects are nonrecursive, such as the cognitive decline that follows a particular neurodegenerative process, can be counted with considerable confidence. A disorder whose decisive aspects are recursive, such as most of what is called depression, cannot. A classification could record this difference rather than presenting all its categories in the same form, with the same appearance of solidity.
The second is to report reliability as a property of the procedure. Agreement between raters shows what a procedure produces; it does not show what the patient has. Field trials could report what had to be held fixed to achieve agreement, which aspects of the encounter were excluded, and which kinds of patient or setting produced disagreement. Disagreement could then be treated as evidence about the domain, as the reliability crisis of the 1970s could have been, rather than as error to be minimized.
The third is to distinguish counting a record from counting a condition. A survey that finds that a proportion of a population meets the criteria of a structured interview has established something about the answers given to that interview, in that place, at that time. Whether it has established the prevalence of a condition depends on how well the answers correspond to the living process, and that correspondence varies with language, setting, and the history of the categories in the population concerned. Reports of prevalence could say which of the two they have counted.
The fourth is to record trajectories as well as states. Recovery, relapse, adaptation, and the slow return to coordinated living are transitions between states, and they are where most of what matters in mental distress takes place. Breakdown leaves records; healing tends toward the silence of ordinary life and leaves none (Ecks 2026a). A classification built for episodes will always see more breakdown than recovery. Longitudinal and qualitative methods that follow people through time, and the clinical practice of holding diagnoses provisionally and revising them as the person changes, are better suited to the recursive aspects of distress than any count of episodes (Ecks 2026b).
Far from making psychiatric numbers less useful, these requirements would make it clearer what the numbers establish, and so make it harder to extend them beyond what they can bear.
X. Conclusion
What exactly happened in DSM-III? Without discovering that mental disorders were discrete, countable entities, and without making them so, the manual performed five operations on the record of mental distress. It turned diagnosis into a count of features reaching a threshold. It cut trajectories into episodes with edges. It removed the context that made distress intelligible, retaining a single exception for bereavement. It divided entangled predicaments into separate units. And it redesigned the encounter between clinician and patient so that its output would no longer depend on who conducted it. Each operation made some feature of the record behave as if it belonged to an object that does not answer back. Structured interviews then removed the clinician altogether, international instruments removed the dependence on place, and the Global Burden of Disease converted the resulting estimates into the most influential numbers in the history of mental health policy.
In doing so, the manual reversed the exclusion with which political arithmetic began. Petty had kept the mutable minds, opinions, appetites, and passions out of his calculations because they answered back. Three centuries later, psychiatry found a way to count them by building a record in which their answering back would leave as little trace as possible. The minds did not become less mutable. Their mutability was moved out of the record and into the spaces between records, where it continues to shape the trends that the records report.
The reversal went unnoticed because it arrived as a solution to a problem of reliability, under the banner of theoretical neutrality, and was quickly absorbed into the infrastructure of insurance, drug regulation, research, and epidemiology. The unreliability that prompted it was itself evidence of the recursive character of the domain, and that evidence was engineered away rather than examined. The consequence is a field whose central numbers carry more political force than any comparable figures in medicine while resting on the weakest fit between record and process. Recognizing what was done in 1980 asks for something more modest than undoing it: knowing, whenever a number about the mind is cited, which part of a recursive process that number has been built to hold still.
References
American Medico-Psychological Association and National Committee for Mental Hygiene. 1918. Statistical Manual for the Use of Institutions for the Insane. New York: National Committee for Mental Hygiene.
American Psychiatric Association. 1980. Diagnostic and Statistical Manual of Mental Disorders. 3rd ed. Washington, DC: American Psychiatric Association.
Anthony, James C., Marshal F. Folstein, Alan J. Romanoski, Michael R. Von Korff, Gerald R. Nestadt, Raman Chahal, Altaf Merchant, et al. 1985. “Comparison of the Lay Diagnostic Interview Schedule and a Standardized Psychiatric Diagnosis: Experience in Eastern Baltimore.” Archives of General Psychiatry 42 (7): 667–675.
Bayer, Ronald. 1981. Homosexuality and American Psychiatry: The Politics of Diagnosis. New York: Basic Books.
Boyd, Jeffrey H., Jack D. Burke, Ernest Gruenberg, Charles E. Holzer, Donald S. Rae, Linda K. George, Marvin Karno, Roger Stoltzman, Larry McEvoy, and Gerald Nestadt. 1984. “Exclusion Criteria of DSM-III: A Study of Co-occurrence of Hierarchy-Free Syndromes.” Archives of General Psychiatry 41 (10): 983–989.
Bromet, Evelyn, Laura Helena Andrade, Irving Hwang, Nancy A. Sampson, Jordi Alonso, Giovanni de Girolamo, Ron de Graaf, et al. 2011. “Cross-National Epidemiology of DSM-IV Major Depressive Episode.” BMC Medicine 9: 90.
Cooper, John E., Robert E. Kendell, Barry J. Gurland, Lawrence Sharpe, John R. M. Copeland, and Ronald Simon. 1972. Psychiatric Diagnosis in New York and London. Maudsley Monographs 20. London: Oxford University Press.
Decker, Hannah S. 2013. The Making of DSM-III: A Diagnostic Manual’s Conquest of American Psychiatry. New York: Oxford University Press.
Ecks, Stefan. 2026a. “The Mathematical Selection of Life: Macroeconomics and Public Health as Forms of Governing.” Living Value Theory, livingvaluetheory.org.
Ecks, Stefan. 2026b. “The Recursive and the Nonrecursive.” Living Value Theory, livingvaluetheory.org.
Feighner, John P., Eli Robins, Samuel B. Guze, Robert A. Woodruff, George Winokur, and Rodrigo Munoz. 1972. “Diagnostic Criteria for Use in Psychiatric Research.” Archives of General Psychiatry 26 (1): 57–63.
Folstein, Marshal F., Alan J. Romanoski, Raman Chahal, James C. Anthony, Michael R. Von Korff, Gerald Nestadt, Altaf Merchant, et al. 1985. “Brief Report on the Clinical Reappraisal of the Diagnostic Interview Schedule Carried Out at the Johns Hopkins Site of the Epidemiological Catchment Area Program of the NIMH.” Psychological Medicine 15 (4): 809–814.
Frances, Allen, and Arnold M. Cooper. 1981. “Descriptive and Dynamic Psychiatry: A Perspective on DSM-III.” American Journal of Psychiatry 138 (9): 1198–1202.
Hacking, Ian. 1995. “The Looping Effects of Human Kinds.” In Causal Cognition: A Multidisciplinary Debate, edited by Dan Sperber, David Premack, and Ann James Premack, 351–383. Oxford: Clarendon Press.
Horwitz, Allan V., and Jerome C. Wakefield. 2007. The Loss of Sadness: How Psychiatry Transformed Normal Sorrow into Depressive Disorder. New York: Oxford University Press.
Kendler, Kenneth S. 2016. “The Phenomenology of Major Depression and the Representativeness and Nature of DSM Criteria.” American Journal of Psychiatry 173 (8): 771–780.
Kendler, Kenneth S., and Charles O. Gardner. 1998. “Boundaries of Major Depression: An Evaluation of DSM-IV Criteria.” American Journal of Psychiatry 155 (2): 172–177.
Kendler, Kenneth S., John Myers, and Sidney Zisook. 2008. “Does Bereavement-Related Major Depression Differ from Major Depression Associated with Other Stressful Life Events?” American Journal of Psychiatry 165 (11): 1449–1455.
Kendler, Kenneth S., Rodrigo A. Muñoz, and George Murphy. 2010. “The Development of the Feighner Criteria: A Historical Perspective.” American Journal of Psychiatry 167 (2): 134–142.
Kessler, Ronald C., Katherine A. McGonagle, Shanyang Zhao, Christopher B. Nelson, Michael Hughes, Suzann Eshleman, Hans-Ulrich Wittchen, and Kenneth S. Kendler. 1994. “Lifetime and 12-Month Prevalence of DSM-III-R Psychiatric Disorders in the United States: Results from the National Comorbidity Survey.” Archives of General Psychiatry 51 (1): 8–19.
Kessler, Ronald C., and T. Bedirhan Üstün. 2004. “The World Mental Health (WMH) Survey Initiative Version of the World Health Organization (WHO) Composite International Diagnostic Interview (CIDI).” International Journal of Methods in Psychiatric Research 13 (2): 93–121.
Kirk, Stuart A., and Herb Kutchins. 1992. The Selling of DSM: The Rhetoric of Science in Psychiatry. New York: Aldine de Gruyter.
Kroenke, Kurt, Robert L. Spitzer, and Janet B. W. Williams. 2001. “The PHQ-9: Validity of a Brief Depression Severity Measure.” Journal of General Internal Medicine 16 (9): 606–613.
Mayes, Rick, and Allan V. Horwitz. 2005. “DSM-III and the Revolution in the Classification of Mental Illness.” Journal of the History of the Behavioral Sciences 41 (3): 249–267.
Murray, Christopher J. L., and Alan D. Lopez, eds. 1996. The Global Burden of Disease. Cambridge, MA: Harvard School of Public Health on behalf of the World Health Organization and the World Bank.
Petty, William. 1690. Political Arithmetick. London: Robert Clavel and Hen. Mortlock.
Robins, Lee N. 1994. “The ECA Study: What Next?” Epidemiologia e Psichiatria Sociale 3 (1): 14–21.
Robins, Lee N., John E. Helzer, Jack Croughan, and Kathryn S. Ratcliff. 1981. “National Institute of Mental Health Diagnostic Interview Schedule: Its History, Characteristics, and Validity.” Archives of General Psychiatry 38 (4): 381–389.
Robins, Lee N., John Wing, Hans-Ulrich Wittchen, John E. Helzer, Thomas F. Babor, Jay Burke, Anne Farmer, et al. 1988. “The Composite International Diagnostic Interview: An Epidemiologic Instrument Suitable for Use in Conjunction with Different Diagnostic Systems and in Different Cultures.” Archives of General Psychiatry 45 (12): 1069–1077.
Rosenhan, David L. 1973. “On Being Sane in Insane Places.” Science 179 (4070): 250–258.
Spitzer, Robert L., Jean Endicott, and Eli Robins. 1978. “Research Diagnostic Criteria: Rationale and Reliability.” Archives of General Psychiatry 35 (6): 773–782.
Spitzer, Robert L., and Joseph L. Fleiss. 1974. “A Re-analysis of the Reliability of Psychiatric Diagnosis.” British Journal of Psychiatry 125 (587): 341–347.
Spitzer, Robert L., Janet B. W. Forman, and John Nee. 1979. “DSM-III Field Trials: I. Initial Interrater Diagnostic Reliability.” American Journal of Psychiatry 136 (6): 815–817.
Spitzer, Robert L., Janet B. W. Williams, and Jean Endicott. 2012. “Standards for DSM-5 Reliability.” Letter. American Journal of Psychiatry 169 (5): 537.
Wakefield, Jerome C., Mark F. Schmitz, Michael B. First, and Allan V. Horwitz. 2007. “Extending the Bereavement Exclusion for Major Depression to Other Losses: Evidence from the National Comorbidity Survey.” Archives of General Psychiatry 64 (4): 433–440.
Wilson, Mitchell. 1993. “DSM-III and the Transformation of American Psychiatry: A History.” American Journal of Psychiatry 150 (3): 399–410.