From the Five-Factor Model to AMPD to HiTOP: How Personality Disorder Diagnosis Went Dimensional

Personality disorder diagnosis spent most of the twentieth century behaving like a small specialty inside descriptive psychiatry: a list of categorical syndromes, each defined by a polythetic checklist, each in principle distinct, each in practice deeply unhappy with the others. By 2013, the DSM-5 task force had gotten close to retiring that whole edifice and replacing it with a dimensional, trait-based scheme drawn — in large part — from the Five-Factor Model (FFM). They lost the political fight, the dimensional model was shunted to Section III of the manual as the Alternative Model for Personality Disorders (AMPD), and the categorical system limped on. A decade later, the HiTOP consortium has been building, in parallel, an empirical hierarchy of psychopathology whose trait level recovers something very close to the FFM. The debate over what DSM-6 should do has become, more or less openly, a debate about how far to trust this structural picture.

This piece is an attempt to lay out that debate honestly. The dimensional case is genuinely strong, and the categorical system it would replace has problems that are not going away. But the trait architecture that is being offered as the replacement was not built for clinical purposes, and the question of how much weight a bottom-up factor-analytic model can carry inside a diagnostic system is more open than the most enthusiastic advocates sometimes allow.

What was actually wrong with the categorical system

The story that DSM-IV personality disorders were a mess is not a caricature. Three findings, repeated across enough independent samples to count as established, drove the dimensional turn.

First, comorbidity was endemic. A patient who met criteria for one personality disorder typically met criteria for two or three others; the average was somewhere around 1.4 additional diagnoses per index PD. If the categories were carving nature at the joints, this much overlap would be a scandal. The more parsimonious reading is that the categories were carving the same underlying continua at slightly different angles.

Second, the residual category swallowed the cases. Personality Disorder Not Otherwise Specified — PD-NOS — was, in many clinical settings, the most frequently assigned PD diagnosis of all. A diagnostic system in which the modal patient does not fit any of the named categories is operating well outside its design specifications.

Third, interrater reliability was unimpressive for several of the DSM-IV PDs once you moved outside structured-interview research conditions. The polythetic format — pick five of nine — produced category memberships that depended heavily on which subset of features the clinician chose to weight. Two clinicians could agree about everything they had observed and still disagree about whether the threshold had been crossed.

These findings did not rule out the existence of meaningful personality pathology. They did suggest that the categorical structure imposed on top of it was at best a coarse approximation and at worst an obstacle to seeing what was actually there.

Costa and Widiger and the FFM-PD translation project

The Five-Factor Model entered this story from the side. It was not built to describe personality pathology. It was built, by Costa and McCrae and a substantial supporting literature, as an empirical structure recovered from factor analyses of self- and peer-rated personality descriptors — initially from the lexical hypothesis tradition (every important individual difference will, eventually, be encoded in everyday language) and later from broader item pools. Five higher-order factors emerged consistently enough across samples, languages, and instruments that the structure became the de facto common metric of trait psychology.

Beginning in the early 1990s, Thomas Widiger and collaborators argued that this same structure could be repurposed to describe personality pathology. The argument had two parts. The empirical part was that DSM-IV personality disorder symptom counts correlated, in predictable patterns, with FFM domain scores: borderline features tracked Neuroticism and low Agreeableness, antisocial features tracked low Agreeableness and low Conscientiousness, schizoid features tracked low Extraversion and (variably) low Openness, and so on. The conceptual part was that this was not a coincidence — that personality disorders just are extreme, maladaptive expressions of the same trait structure that describes the rest of personality.

Widiger and Costa packaged this into an explicit translation: the FFM-PD model, complete with proposals for re-describing each DSM-IV personality disorder as an extreme profile on the five domains and their facets. The proposal was that clinicians could replace the categorical diagnosis with a profile, and that the profile would do everything the diagnosis did and more — capture the specific pattern of impairment, locate the patient in the same descriptive space as everyone else, and dispense with the comorbidity and NOS problems by treating the underlying continua as continuous in the first place.

It is worth being clear about what is and is not impressive in this proposal. The empirical mapping is real: extreme FFM profiles do recover a substantial portion of what DSM-IV PD criteria were trying to capture. The conceptual claim — that personality disorders are nothing more than extreme trait configurations — is a much stronger inference than the data support, and the field has been arguing about it ever since.

DSM-5 Section III: the AMPD as a hybrid

What ended up in Section III of DSM-5, the Alternative Model for Personality Disorders, was a compromise. The AMPD has two criteria.

Criterion A is level of personality functioning: a clinician-rated dimensional assessment of how well the patient does on identity, self-direction, empathy, and intimacy. This is the part of the model that is not a pure trait restatement. It comes out of the psychodynamic and self-development tradition — Kernberg's levels of personality organization, Bender's work on self and interpersonal functioning — and it is the AMPD's attempt to keep the pathology in personality pathology, rather than reducing it to extreme trait scores.

Criterion B is the trait part: 25 maladaptive trait facets organized into 5 higher-order domains — Negative Affectivity, Detachment, Antagonism, Disinhibition, and Psychoticism — measured with the Personality Inventory for DSM-5 (PID-5). These five domains are, with one important caveat, the FFM with the polarity flipped to the maladaptive end:

FFM to AMPD Criterion B crosswalk: Neuroticism maps to Negative Affectivity, low Extraversion to Detachment, low Agreeableness to Antagonism, low Conscientiousness to Disinhibition, and Openness loosely (and contestedly) to Psychoticism.

The crosswalk is approximate. Negative Affectivity is high Neuroticism. Detachment is, broadly, low Extraversion plus elements of low positive affect. Antagonism is low Agreeableness. Disinhibition is low Conscientiousness. The ragged edge is Psychoticism, which has been described as "Openness gone wrong," but the empirical loadings of PID-5 Psychoticism on FFM Openness are weak and inconsistent, and several authors place Psychoticism on a separate higher-order factor entirely. This is not a fatal flaw in the model — it is, more accurately, an honest signal that the FFM's fifth factor was not built to describe disordered cognition and is being asked to do work it was not designed for.

The AMPD then defines six specific personality disorder types (antisocial, avoidant, borderline, narcissistic, obsessive-compulsive, and schizotypal) as configurations on Criterion A and Criterion B, with the rest of the diagnostic territory covered by a residual trait-specified category. The structure is dimensional, but the surface still looks categorical to the clinician who wants a name to put in the chart.

HiTOP: a parallel bottom-up hierarchy

While the AMPD was being argued out inside the DSM process, a separate consortium — led by Roman Kotov, David Watson, and a large multinational group — was building the Hierarchical Taxonomy of Psychopathology (HiTOP) from the bottom up. HiTOP is not a personality disorder model. It is an empirical hierarchy of psychopathology in general, derived from large-scale factor analyses of symptom and trait data across the internalizing, externalizing, thought-disorder, and somatoform domains.

What is interesting for the present argument is that the HiTOP trait level recovers, with minor relabeling, the same five domains as Criterion B of the AMPD. HiTOP's spectra at the trait level are usually labeled Internalizing, Detachment, Antagonism, Disinhibition, and Thought Disorder, mapped to facet-level traits that look very much like the PID-5 facets and, behind them, very much like FFM facets. Two largely independent research programs — one driven from the top down by the DSM revision process, one driven from the bottom up by joint factor analysis of psychopathology data — converged on a trait architecture that is recognizably the same.

Convergence of this kind is the strongest empirical argument in favor of the dimensional turn. It is hard to attribute the structure to either tradition's idiosyncrasies if both keep finding it.

What is actually being signaled for DSM-6

The DSM-6 process is ongoing, and any specific claim about what will be in the manual should be hedged hard. What can be said is what the relevant work groups and commentators have publicly signaled.

The strongest signal is AMPD promotion. Several members of the original AMPD work group — Krueger, Skodol, Hopwood, and others — have argued, in published commentary, that the empirical case for the dimensional model has only strengthened in the decade since DSM-5, and that DSM-6 should move the AMPD out of Section III and into the main text. ICD-11 has already done something similar: its personality disorder chapter, which came into force in 2022, replaces categorical PDs with a single severity dimension plus five trait qualifiers (Negative Affectivity, Detachment, Dissociality, Disinhibition, and Anankastia) that overlap heavily with the AMPD domains. The international convergence is real and is being cited in DSM-6 deliberations.

The second signal is categorical retirement, or demotion to specifiers. If the AMPD is promoted, the categorical PDs would either be retired or kept as descriptive specifiers attached to the dimensional profile. This is the most contested piece of the package, because the categorical labels have substantial clinical, legal, and forensic infrastructure attached to them. Even advocates of full dimensionalization tend to acknowledge that "borderline personality disorder" is unlikely to vanish from clinical vocabulary regardless of what the manual does.

The third signal is harmonization with HiTOP. Some commentators have argued that DSM-6 should not just adopt the AMPD trait domains but also acknowledge the broader HiTOP-style hierarchical structure, which would situate personality pathology inside the same trait space as internalizing and externalizing psychopathology more broadly. This is the most ambitious option and the least likely to be adopted in full.

None of this should be read as established. The DSM revision process is governed by a working-group consensus that has historically defaulted toward continuity, and the categorical system has constituencies — particularly in forensic and insurance settings — that have nothing to do with the empirical merits of the dimensional alternative. A reasonable prior is that DSM-6 will move some distance toward the AMPD without going as far as its strongest advocates would like.

Where the FFM is genuinely doing useful work

Stepping back from the politics, what is the case for using the FFM (or its AMPD/HiTOP cousins) inside a diagnostic system?

The structural case is the strongest. A trait architecture that has been recovered from factor analyses of self-report, peer-report, lexical, behavioral, and clinical data across dozens of countries and languages is doing something real about the structure of individual differences. The convergence between the top-down AMPD and the bottom-up HiTOP is not the kind of thing one gets by accident. If the goal is to describe where a patient sits in the population on a small number of broadly orthogonal continua, the FFM-derived domains are, demonstrably, an efficient and replicable way to do it.

The clinical case rests on three more specific arguments. First, profile-based formulation: a five-domain profile carries more individuating information than a categorical label, and that profile is comparable across patients and across time in a way categorical labels are not. Second, trait-targeted treatment selection: there is preliminary evidence that some trait domains track differential response to specific interventions — for example, high Negative Affectivity profiles respond to interventions that target affect regulation, regardless of the categorical label. Third, fit with longitudinal data: trait-level structure is more stable across the lifespan than categorical PD diagnosis is, and stability is a feature one wants in a construct that is supposed to describe enduring patterns.

These are real points. None of them, taken alone, settles the question.

Where the FFM does not fit cleanly

The honest case against unrestricted clinical use of the FFM as the basis of a personality disorder system has at least three threads, and they are not the same complaint.

The first is the absence of theoretical grounding. The FFM is a structure recovered from factor analysis. It was not derived from a theory of personality, of psychopathology, or of human functioning. It does not specify what is supposed to vary, why those particular five dimensions and not others should be expected, what mechanism produces individual differences along them, or what their developmental origins are. Costa and McCrae's later five-factor theory is a post hoc attempt to provide that grounding, but the structure existed before the theory and would survive the theory being wrong. This is not, in itself, a fatal objection — the periodic table also predated its theoretical explanation — but it is a different kind of warrant than a model derived from a clinical theory carries. The FFM tells you reliably that people differ along these continua. It does not tell you why a particular person's position on those continua looks the way it does, what generated it, or what would change it.

A related point is worth being explicit about, because it tends to get glossed over in the more enthusiastic presentations of the model. The FFM is sometimes described as if its grounding in everyday language and in factor-analytic methods made it scientifically pure in a way that more theory-laden models are not — as if the data simply gave us the five factors and the researchers merely reported what they found. That framing is overstated. Factor analysis returns components, not names. The decision that a particular component is Conscientiousness rather than Constraint or Will or Industriousness, that another is Openness to Experience rather than Intellect or Culture or Imagination, and that a fifth labeled Psychoticism in the AMPD reasonably describes the same thing as Openness in the FFM, is in every case a human-judgment act of construal performed by researchers with theoretical commitments. The labels carry connotations, the connotations shape downstream interpretation, and the long-running disagreement between the Openness and Intellect camps about the fifth factor is the clearest demonstration that even at the highest level of the structure the meaning is not read off the data — it is assigned to the data, by people, working inside a tradition. The factor structure is genuinely robust. The semantic interpretation of that structure is a much more conventional act of scientific theorizing than the model's empirical reputation suggests.

The second is the non-clinical and largely lexical origin of the data. The factor structure was built primarily from self- and peer-rated personality descriptors collected from non-clinical samples. The lexical hypothesis is a serious idea, and the fact that natural-language adjectives encode something like the FFM is genuinely surprising and informative — but it is not the same thing as a model built to describe pathology. Symptoms and traits are not the same kind of entity. When you push the FFM into the pathological extremes, what you get is, at best, a description of the surface features that pathological people share with non-pathological people who happen to be unusually high or low on the same continua. Whether that description is sufficient to capture what is clinically distinctive about pathological functioning is exactly what is in dispute. The AMPD's Criterion A is, in part, an attempt to plug this hole — it imports a non-trait construct (level of personality organization) precisely because the trait domains were not doing the necessary clinical work on their own.

The third is the asymmetry between structural and clinical validity. The FFM has very strong structural validity: the five-factor solution replicates and the loadings are stable. Its clinical validity — its ability to do the work that diagnosis is supposed to do, in the consulting room and in treatment planning — is a much smaller and more contested literature. There is real evidence that FFM and AMPD profiles correlate with clinically meaningful outcomes; there is much less evidence that they outperform a thoughtful clinical formulation built around categorical or psychodynamic constructs in the cases where it actually matters. The structural case has been allowed, in some of the more enthusiastic writing, to do the clinical case's work for it. That is the point at which a Meehl-style caution becomes warranted: a measurement model that is well-validated as a measurement model is not, by that fact alone, well-validated as a clinical instrument.

None of this argues against dimensionalization in principle. It argues against pretending that the particular dimensional model on offer is a finished product, or that adopting it solves the deeper problem the categorical system was failing at — which is the problem of saying what, in any particular patient, is actually wrong.

What the dimensional turn would and would not solve

The clearest gains would be on the problems that drove the dimensional case in the first place. Comorbidity would dissolve into co-elevation on shared continua, which is what it should have been called all along. PD-NOS would disappear, because every patient would receive a profile rather than a category match. Reliability would improve at the trait level, because dimensional measurement is less sensitive to threshold artifacts than polythetic categorical measurement.

The harder questions would not go away. The question of what counts as a personality disorder would shift from "which categorical box does this patient fit" to "where on each domain, and at what level of impairment, does this patient sit" — and the second formulation is not obviously easier to answer in practice. The question of cultural validity — whether the trait structure, derived predominantly from Western and largely English-speaking samples, generalizes adequately to other cultural and linguistic contexts — has been argued for both sides and is not settled. The question of borderline personality organization, in the Kernberg sense, does not map cleanly onto any subset of the AMPD trait domains, which is one reason Criterion A exists; whether a hybrid model with a separate functioning criterion is a stable solution or a transitional compromise is an open question. And the question of psychoticism's status — whether it is a degenerate corner of Openness, a separate apex factor, or a categorical departure from continuous variation — is not going to be answered by adopting the AMPD; it is just going to be inherited.

Where this leaves the clinician and the curious reader

The most honest summary is something like the following. The categorical PD system has serious empirical problems. The dimensional alternative has a strong structural case and a weaker but real clinical case. The trait architecture that the alternative rests on — the FFM, the AMPD trait domains, the HiTOP trait level — is one of the most thoroughly replicated structural findings in personality psychology, and it is also a model that was not built to do clinical work and is being asked to do it. Both of those things are true, and the right response is neither to embrace the dimensional turn uncritically nor to resist it in defense of a categorical system whose problems are not going to be solved by waiting.

For research, the case for using the FFM-derived structure is overwhelming. It is the common metric the field actually has, it travels well across samples and studies, and it makes findings comparable in a way categorical PD research never quite achieved. For clinical use, the case is real but more modest: the dimensional profile is genuinely informative, it carries useful prognostic and treatment-relevant information, and it should probably be part of how serious personality assessment is done — alongside, not in place of, the clinical formulation that asks what is actually going on inside a particular life. That is the position the empirical record currently supports. Anything stronger is enthusiasm, and enthusiasm is not what the personality disorder literature has historically suffered from a shortage of.

If you want to see where you sit on the trait architecture this article has been describing, the Big Five (IPIP-50) test on this site uses Goldberg's public-domain item pool — the same family of items that underwrites most of the FFM research literature — to produce a five-domain profile in about ten minutes. For the schizoid/avoidant question that sits at the difficult interface between the dimensional and categorical traditions, the IPIP schizoid scale and the SADT walks through what trait-level measurement does and does not capture. For the broader interpersonal-style framing that AMPD Criterion A draws on, see the history of the interpersonal circumplex. And for the case that no single model — categorical, dimensional, or psychodynamic — should be doing this work alone, see the multimethod argument.

Selected references

• Bender, D. S., Morey, L. C., & Skodol, A. E. (2011). Toward a model for assessing level of personality functioning in DSM-5, Part I: A review of theory and methods. Journal of Personality Assessment, 93(4), 332–346.
• Costa, P. T., & McCrae, R. R. (1992). Revised NEO Personality Inventory (NEO-PI-R) and NEO Five-Factor Inventory (NEO-FFI) professional manual. Psychological Assessment Resources.
• Hopwood, C. J., Kotov, R., Krueger, R. F., Watson, D., Widiger, T. A., et al. (2018). The time has come for dimensional personality disorder diagnosis. Personality and Mental Health, 12(1), 82–86.
• Kotov, R., Krueger, R. F., Watson, D., et al. (2017). The Hierarchical Taxonomy of Psychopathology (HiTOP): A dimensional alternative to traditional nosologies. Journal of Abnormal Psychology, 126(4), 454–477.
• Krueger, R. F., Derringer, J., Markon, K. E., Watson, D., & Skodol, A. E. (2012). Initial construction of a maladaptive personality trait model and inventory for DSM-5. Psychological Medicine, 42(9), 1879–1890.
• Skodol, A. E., Bender, D. S., Morey, L. C., et al. (2011). Personality disorder types proposed for DSM-5. Journal of Personality Disorders, 25(2), 136–169.
• Watson, D., Stasik, S. M., Ro, E., & Clark, L. A. (2013). Integrating normal and pathological personality: Relating the DSM-5 trait-dimensional model to general traits of personality. Assessment, 20(3), 312–326.
• Widiger, T. A., & Costa, P. T. (Eds.). (2013). Personality disorders and the five-factor model of personality (3rd ed.). American Psychological Association.
• World Health Organization. (2019/2022). International Classification of Diseases, 11th Revision (ICD-11): Personality disorders and related traits.
• Wright, A. G. C., Thomas, K. M., Hopwood, C. J., Markon, K. E., Pincus, A. L., & Krueger, R. F. (2012). The hierarchical structure of DSM-5 pathological personality traits. Journal of Abnormal Psychology, 121(4), 951–957.