The DCIDE framework: systematic investigation of evolutionary hypotheses, exemplified with autism

ABSTRACT

Evolutionary explanations of mental disorders are a longstanding aim of evolutionary psychiatry, but have suffered from complexities including within‐disorder heterogeneity and environmental effects of contemporary societies obscuring possible ancestral functions. Studying the relevant processes of human evolution directly is not possible, so hypotheses have remained speculative, exaggerating “just‐so storytelling” critiques. This is despite significant evidence existing in genetics, neuroscience and epidemiology, all of which bears some inferential relevance to evolutionary hypotheses, but which is often not marshalled in a systematic way. To utilise this evidence best to investigate evolutionary explanations of psychiatric (or other) traits we present a novel framework of evidence synthesis and analysis and exemplify it by systematically reviewing evidence related to autism. In the five stages of this “DCIDE framework” analysis, Description identifies a trait to explain and Categorisation initially excludes verifiably non‐adaptive cases by utilising evidence from genetics, neuroscience, and environmental factors. Integration then hones a target for adaptive explanation by considering evidence of age of onset, environmental effects, duration, prevalence and sex differences, incorporating relevant correlated traits visible to selection. Evolutionary hypotheses are then Depicted and Evaluated for their ability to explain all the evidence at hand, using standardised areas of evidence and theoretically motivated principles (e.g. traits arising at birth and lasting for life have different plausible explanations than traits arising in adolescence and receding in adulthood). Competing evolutionary hypotheses can thus be systematically compared for their sufficiency in explaining a wide range of available evidence. In the DCIDE review of autism, when Described with current diagnostic criteria, up to 20% of cases Categorise as non‐adaptive, primarily caused by de novo mutations and environmental trauma. The remaining cases are eligible for adaptive explanation. For Integrating genetically correlated phenotypes, evidence of high prevalence of subclinical familial traits and camouflaged female cases is necessary. Competing Depictions contrast a high intelligence by‐product hypothesis with social niche specialisation for high “systemising” cognition. In Evaluation, broad evidence supports the social niche hypothesis while the intelligence by‐product hypothesis fails to predict various lines of evidence. This provides not only the most robust synthesis of autism research relevant to evolutionary explanation to date, but is a first example of how the structure of the DCIDE framework can allow improved systematic evolutionary analysis across psychiatric conditions, and may also be adopted to strengthen evolutionary psychology more generally, countering just‐so storytelling and cherry‐picking critiques.

I. INTRODUCTION

Inspired to explain biological form and variation, and in particular the sophisticated fit of organism's traits to their survival in their local environment, Darwin (1869) proposed his theory of evolution by natural selection. The 20th century then brought huge advances in mathematical models (Fisher, 1930), theoretical understanding (Maynard Smith, 1993) and empirical evidence for evolution by natural selection.

Psychologists have extended this explanatory theory to the human mind, in “evolutionary psychology” (Barkow, Cosmides & Tooby, 1995; Pinker, 1997; Buss, 2015). Despite capturing academic and public attention, evolutionary psychology has been subject to both scientific and socio‐political criticisms. Scientifically, there are questions over validity, rigour, and falsifiability; socio‐politically, there are concerns over interpretation in ways that are sexist, racist or eugenicist. These critiques are not unique to evolutionary psychology, but have perhaps uniquely hampered the acceptance of this particular subfield of evolutionary inquiry (Gould & Lewontin, 1979; Segerstrale, 2000).

More recently, the relevance of evolutionary theory for understanding health and disease has been gathering traction in the fields of evolutionary medicine and evolutionary psychiatry. However, these fields are liable to some of the same criticisms, and more. This article briefly reviews some of these before offering a novel framework of systematic review designed to address the major scientific problems affecting evolutionary psychiatry by enhancing evolutionary inference.

II. THE HISTORICAL CHALLENGE OF TESTING ADAPTIVE HYPOTHESES

Immediately after Darwin (1869) proposed natural selection as the force shaping apparent design of biological forms, he was subjected to various criticisms. Philosopher of science Karl Popper spent decades criticising evolutionary theory as being fundamentally beyond the reach of science by being confined to the past: “The evolution of life on Earth, or of human society, is a unique historical process… Its description, however, is not a law, but only a singular historical statement.” (Popper, 1957, p. 108). Popper eventually changed his mind, accepting evolution by natural selection as a falsifiable and valid scientific hypothesis (Sonleitner, 1986) – but the same vein of criticism is still common. Problematically, the theory is essentially historical. Evolutionary scientists have tried to address this in various ways.

George Williams' Adaptation and Natural Selection (Williams, 1966) is famous for outlining the expected characteristics of adaptations (as well as pointing out the improbability of group selection). To Williams, adaptation and adaptive design explain complex biological traits, and the necessity of believing adaptive design caused a trait is evident in their efficiency at performing some function or goal. Yet he recognises limitations in defining and defending the presence or absence of adaptation. “The formulation of practical definitions and sets of objective criteria will not be easy, but it is a problem of great importance and will have to be faced.” (Williams, 1966, p. 9).

Williams himself relies on informal arguments, especially analogy between organic systems and human technologies, which provide “indirect evidence of complexity and constancy” (Williams, 1966, p. 10). Mainstream evolutionary biology and psychology has carried this approach of “reverse engineering” forward (Dawkins, 1986; Pinker, 1997; Symons, 1990). Eyes are complex organs with an array of systems and components perfectly suited for vision (Goldsmith, 1990), so vision must be the function of eyes. Whilst this suffices to justify uncontroversial cases such as eyes, as Williams (1966) notes, it struggles where disputes arise, for example in reverse engineering the function of low mood – it is very likely some function exists, but this is much harder to discern and it is difficult to make an uncontroversial argument.

Although uncontroversially evolved psychological mechanisms with clear functions surely exist – for example, sexual desire and hunger – evolutionary psychology has primarily aimed to explore the mechanisms behind less‐intuitive examples, such as nuances in sex‐specific jealousy (Buss et al., 1992; Buunk et al., 1996), incest avoidance (Lieberman, Tooby & Cosmides, 2007), and shame (Sznycer et al., 2016). As such, it has been more open to the same century‐old criticism of forwarding speculative untestable hypotheses, often labelled as “just‐so stories” (Smith, 2019; Dupré, 2003; Gould & Lewontin, 1979). More specifically, there are continued claims that evolutionary hypotheses are unfalsifiable, lack sufficient testable predictions (McCain & Weslake, 2013), and are excessively reliant on modern observations of functioning (Henrich, Heine & Norenzayan, 2010).

Difficulty evidencing hypotheses in evolutionary psychology is a clear consequence of the study system. While evolutionary biologists have developed a strong toolkit for testing evolutionary hypotheses that avoids such critiques, including experimental evolution or measuring selection in natural populations, these methods are often not applicable to humans for ethical or practical reasons (although note that the field of human behavioural ecology often does measure selection; see Scelza, Koster & Shenk, 2024). Evolutionary psychology generally begins with theory‐derived hypotheses of mechanisms likely to have evolved in the mind and predicts specific experimental or observational outcomes if such mechanisms exist, or derives adaptive explanations for already well‐known mechanisms, which may lead to further, more nuanced experimental testing.

One of the problems here is that “storytelling” can occur in suggesting the adaptive problem and the functional design, with an experimental result proposed which is not exclusive to the explanation. Evolutionary psychologists recognise the need to design experiments with predicted outcomes that are as precise as possible to the hypothesis of the evolved psychological mechanism (Lewis et al., 2017), but this meets scientific limitations, because many experimental results are compatible with multiple hypotheses as to the underlying psychological architecture (e.g. if males and females experience jealousy in different situations, is this due to a general “mate guarding” mechanism or to several more specific mechanisms, for example for “noticing competitors” “identifying disengagement” “awareness of partner absence”, etc.) This has been called the “grain problem” (Atkinson & Wheeler, 2004). Even the strongest critics recognise that evolutionary processes to some extent shaped the mind (Smith, 2019); the contention is aimed at the scientific limitations for understanding how.

Furthermore, the pursuit of evolutionary psychology has been somewhat tarnished, both scientifically and socially, by concentration on specific hypotheses that have been both incorrect and damaging. A prominent example here is the “dual mating strategy” hypothesis of female mate choice (Gangestad, Thornhill & Garver‐Apgar, 2005). This argued that human females seek copulation with males of the highest genetic quality during ovulation to get pregnant, but whilst non‐ovulating will maintain relationships with other males to receive resources and support, including in childcare. This hypothesis was prominently referenced in books and evolutionary psychology papers, pointing to a handful of positive experimental results (e.g. Pillsworth & Haselton, 2006). Eventually it was co‐opted to denigrate women as innately deceptive and lament a fictional version of human hierarchies in the “manosphere” and “incel” communities. However, none of the core motivating evidence on female preferences changing during ovulation has replicated (Gangestad & Dinh, 2022; Alexander, 2022).

III. EVOLUTIONARY PSYCHIATRY

Psychiatric conditions such as autism and schizophrenia are generally of high cost, unknown aetiology and incomprehensible pathophysiology (Alawieh et al., 2012). Unsuccessful efforts towards explanations or treatments have been plentiful, with advances in neuroscience and genetics generally failing to identify pathology and blurring boundaries between disease states and normal variation (Geschwind & Flint, 2015). In tandem with burgeoning evidence in genetics, epidemiology and neuroscience in mainstream psychiatric research, the field of evolutionary psychiatry has been developing theoretical frameworks attempting to make sense of disorder at the evolutionary level (Abed & St John‐Smith, 2022a; Hunt, St‐John Smith & Abed, 2023; Nesse & Williams, 1996; Nesse, 2019), borrowing key principles from the better established field of evolutionary medicine (Gluckman et al., 2016). Psychiatry and clinical psychology have concentrated almost exclusively on the individual, their social surroundings, psychological processes, and biological make‐up, but very rarely considered the “ultimate” level of explanation of evolutionary history and adaptive function. Psychiatry has lauded the biopsychosocial model (Hunt, Abed & St John‐Smith, 2022), but it should expand to be “evobiopsychosocial” (Hunt et al., 2023).

As Jerome Wakefield's (Wakefield, 1992, 1997, 2005) “harmful dysfunction” definition of disorder highlights, evolutionary perspectives on whether a trait is an adaptation or not are critical to its status as a true “disorder”. Disorders are dysfunctional in the evolutionary sense and harmful in an evaluative sense, where an individual or society believes the dysfunctional trait is harmful. The evolutionary sciences are the only paradigm to lend a truly objective standard for health and disease – that defines what biology “should” be doing, as has been recognised by many philosophers of biology and medicine (Neander, 1991; Millikan, 1989; Wright, 1976).

Where evolutionary psychology concentrates on describing functional design, evolutionary psychiatry and medicine aim to describe why natural selection has left us vulnerable to disorders, providing both general explanations and hypotheses about specific conditions. Nesse (2005) synthesised such explanations into six types: mismatch with modern environments, host–pathogen co‐evolution, harmful trade‐offs of adaptations, constraints on what natural selection can achieve, defences that carry aversive symptoms and traits that increase fitness at the expense of health (see also Abed & St John‐Smith, 2022b; Del Giudice, 2018a). These general explanations are not mutually exclusive: for example, a specific disorder could be a mismatched defence [e.g. social anxiety (defence) overactivated by social media (mismatch)].

In comparison to evolutionary psychology, evolutionary psychiatry is arguably much more scientifically complicated to test (Zachar & Kendler, 2017). It carries the same difficulty of hypothesising about a historical process, with additional complexities. Adaptations are inherited fitness‐enhancing traits, and their functions are the processes that improve their fitness (Hunt, 2023a; Sober, 2000). The presence of adaptation is often obvious (zebras' stripes are surely adaptations) and the function of adaptation is the interesting and more complex question [are zebras' stripes for camouflage, predator avoidance, heat management, social interaction or protection against biting insects; apparently, the latter (Caro et al., 2014)]. However, in evolutionary psychiatry the definition of a trait as an adaptation or not is itself a controversial question.

The likelihood that disorders are not functional adaptations (although some may be) undermines the reverse‐engineering strategy of hypothesising of complex design features to be ascertained experimentally. Mismatch with modern environments might mean that adaptive traits are now functioning quite differently – even dysfunctionally – despite historically being adaptive. On top of this, there are complexities regarding the correct objects of explanation. Disorder categories are highly heterogenous and may contain both functional and dysfunctional subtypes – some cases of “social anxiety” may be adaptive, some maladaptive (Nesse, 2022). Relevant subclinical manifestations of a particular disorder may be disregarded from analysis due to clinical focus on disease categories rather than quantitative traits, despite genetic relationships and shared causality between subclinical and clinical forms (Smoller et al., 2018). If so, evolutionary explanations aimed at the disorder itself will be missing the key phenotype with adaptive benefits – akin to trying to explain sickle cell disease without considering the heterozygous carriers benefitting from malaria resistance. The over‐a‐century long development of the evolutionary sciences has concentrated almost exclusively on adaptation and function; evolutionary medicine and psychiatry require a different methodological approach.

Motivation for developing the fields is strong, particularly in psychiatry, where previous paradigms have been disappointing. If there is a single critical justification for serious consideration of evolutionary psychiatry, it is seen in a paradox recognised for many decades as the “schizophrenia paradox” (Huxley et al., 1964), now framed more generally as the paradox of common, harmful, heritable mental disorders (Keller & Miller, 2006). Given that all mental disorders have a genetic component – recognised for over a century by observing familial inheritance, now certified in genome‐wide association studies (GWAS) – why has natural selection not removed the contributing alleles from the population? Keller & Miller (2006) suggested that mutation–selection balance would be sufficient, but this has not been supported by recent genetic data (Keller, 2018). This remains a paradox that the correct explanation must answer – whether neurobiological, or psychological, or genetic – the persistence of these conditions in the human species is an evolutionary mystery. This is particularly paradoxical because psychiatric conditions, unlike most disease, usually arise early in life, before the age of reproduction, so are entirely visible to forces of natural selection.

IV. MEDICINE, SCIENCE, AND THE NATURALISTIC FALLACY

Beyond scientific problems, a contention with evolutionary approaches to psychiatry and medicine is that evolutionary classifications revolve around facts of evolutionary history, which is medically irrelevant: medicine seeks to heal suffering. Much suffering comes from pain, but pain is evolutionarily functional, and often deemed worth treating medically, as researchers in evolutionary medicine and psychiatry note (Nesse & Schulkin, 2019). Even if biological function is defined most reliably through the evolutionary account, medical decisions do not, and should not take it as directive; we should not restrict treatment to conditions classified as evolutionarily dysfunctional. Leading philosophers of medicine defending the value of evolutionary approaches, including Wakefield, recognise this (Wakefield, 2021; Boorse, 2016).

There is an important ethical point here regarding the “naturalistic fallacy”: believing that what is natural is good or right. This is both immoral and inappropriate. In the “natural” hunter–gatherer state without vaccinations, antibiotics or sanitation, 50% of humans die before they are 25–30 (Gurven & Kaplan, 2007). To some extent, the entire point of medicine is defying what is “natural” – its purpose is to intervene and alter natural outcomes (excepting instances of treating “unnatural” – or artificial – harms). Utilising evolutionary analysis to understand the cause of disorder bears no necessary connotation to its treatment. This is why some philosophers of medicine support value‐based rather than science‐based approaches to “disorder” attribution (Glackin, 2019).

If evolutionary explanations are disconnected with treatment decisions, why investigate them? The answer is the same that justifies the relevance of any science to medicine, namely that scientific explanation can inform the values upon which we justify treatment decisions.

Medicine is fundamentally disconnected from science. Indeed, medicine is many millennia older than science, and practised by non‐scientific communities and individuals – clearly medical decision‐making is not inherently scientific (Murphy & Woolfolk, 2000). However, scientific information can be useful in directing both the development of treatments and treatment decisions. The dynamic here is not one of equally necessary components, as Wakefield suggests defines true “disorder”. Instead, medicine acts solely upon values of alleviating suffering and harm. However, in forming those values, scientific understanding (causal explanations, medical research, patient experience, and so on) is considered in integration with all other values (religious, traditional, political, humanist; social, familial, individual, and so on) (Hunt, 2023b).

Revealing an evolutionary dysfunction or function may plausibly alter our values regarding a trait [e.g. evidence that attention deficit hyperactivity disorder (ADHD) had an evolutionary function (Williams & Taylor, 2006; Shelley‐Tremblay & Rosén, 1996) may make us think differently about stimulant medication – perhaps seeing it as a useful modern solution to a problem of evolutionary mismatch, perhaps as overmedicalisation]. Different formulations of scientific explanation may also have interesting downstream effects via stigmatisation and self‐stigmatisation. Neuroscientific and genetic (“biogenetic”) versus environmental explanations (e.g. stress) have different effects on stigma, where stress explanations reduce hopelessness and biogenetic explanations increase endorsement of medical intervention (Loughman & Haslam, 2018). Functional explanations of depression also seem to reduce self‐stigmatisation (Schroder, Devendorf & Zikmund‐Fisher, 2023). Further indications that evolutionary explanations may be quite meaningful across a range of psychiatric conditions are seen in the many overlaps between evolutionary psychiatry and the “neurodiversity” social movement (Hunt & Procyshyn, 2024).

In medical research, recognition of evolution has less diffuse consequences, particularly in directing towards useful animal models (Hunt et al., 2023). Animal models are only useful insofar as the system is conserved between species – for many psychiatric traits, which appear to be human specific and affect higher cognitive and social function, this implies that rodent models are useful for testing toxicology but not effects on symptomology. This differs between disorders – anxiety has clear correlates in other species, but schizophrenia does not (Geyer et al., 2012). Furthermore, an evolutionary approach has the potential to provide an improved system for classifying disorders, in line with the different forms and causes of dysfunction. Evolutionary theory provides a causal theory which could help subtype broad diagnoses (e.g. looking for 12 subtypes of “depression”; Rantala et al., 2018), allowing theory‐informed “precision medicine”. Grounding research into psychopathology in the theory which explains all biological variation may be the solution to the continuous failures in psychiatric research so far. Thomas Insel, who led the National Institute of Mental Health between 2002 and 2015 and oversaw around $20 billion of research spending (Rogers, 2017) lamented, “Whatever we've been doing for five decades, it ain't working … Maybe we just need to rethink this whole approach … With no validated biomarkers and too little in the way of novel medical treatments since 1980 … it is time to rethink mental disorders” (Greenberg, 2013). However, to seriously rethink mental disorders through an evolutionary lens, improved methods of analysis are necessary.

V. THE DCIDE FRAMEWORK

Although the historical process of evolution is often unobservable directly, a range of evidence can be marshalled in a process of “inference to the best explanation” (Douven & Wenmackers, 2017). As Lewis et al. (2017, p. 370) note in their exposition of methods in evolutionary psychology, “Ultimately, the merit of a hypothesis (…) should be determined based on the cumulative body of evidence.”

The pursuit of evolutionary psychiatry facing the complexities of evolutionary psychology and more, improved methods for developing, assessing and comparing hypotheses are sorely desirable. Here it is particularly critical to move beyond intuitions of functional design and formulate practical definitions and sets of objective criteria, as Williams (1966, p. 9) called for.

Here we provide a novel framework for rigorous and systematic formulation and evaluation of such evolutionary hypotheses. While this article is written specifically with human psychiatric conditions in mind, many of its principles could also be adapted for studying other species for which direct tests of evolutionary hypotheses are not feasible, but a wealth of more circumstantial evidence is available.

The “DCIDE” framework is a novel approach to testing evolutionary explanations and distinguishing the causes of disorder by evolutionary principles (Fig. 1), making use of available empirical evidence in a systematic way (Fig. 2). Briefly, a trait is Described; subtypes within it are Categorised as requiring adaptive explanation or not; for those requiring adaptive explanation, relevant genetically related traits are Integrated to capture the broader phenotype visible to selection; various hypotheses explaining the traits' reproductive success are Depicted; and those hypotheses are Evaluated in a standardised manner. Its framework allows for a single condition's heterogeneity to be parsed out (e.g. different cases of autism could be primarily caused by vulnerability, mutation, by‐product or adaptation). This is essentially a philosophically rigorous formalization of the process which some diligent researchers in evolutionary psychology and psychiatry may utilise anyway (e.g. Del Giudice, 2018a), that makes key steps and standardisable interpretations of evidence explicit, mitigating concerns of cherry picking and just‐so storytelling.

Fig. 1.

Common kinds of explanations for medically relevant conditions. Bold text indicates distinct kinds of explanation. Bullet points list common examples from evolutionary medicine, which are not mutually exclusive with each other. For a closer breakdown of mutually exclusive evolutionary explanations, see Hunt (2023a). Note that mutation–selection balance is easily identifiable in cases implying historical negative reproductive success, but non‐adaptive genes can also be mutations awaiting negative selection that are reproductively successful due to other circumstances.

Fig. 2.

The DCIDE framework. Green rectangles are steps for honing target of explanation. Blue rectangles are steps attempting adaptive explanation. Yellow text indicates possible outcomes of analysis. Ovals indicate sufficient explanation.

A major initial hurdle in hypothesising in evolutionary psychiatry is the heterogeneity within single mental disorder labels. When causing mental disorders, mutation, vulnerability and non‐adaptive genes are likely fitness reducing, so can be defined as disruptions leading to dysfunctional processes (Hunt, 2023a): fairly plain cases of disorder, and investigable with existing biomedical tools. However, biomedicine has not been able to explain most of the variance behind diagnosable mental disorders – plausibly because they are related to adaptations and by‐products of adaptations. We call these “adaptive explanations” as they postulate positive selection on the causal alleles, although note that many traits requiring adaptive explanation may be harmful by‐products, not adaptive in themselves (considered further in Section V.3). Non‐adaptive explanations referencing evolutionary processes such as genetic drift can also be assessed and compared with adaptive explanations.

The DCIDE framework is fundamentally a way of organising evidence to assist evolutionary explanations of complex biological phenomena by integrating multiple areas of incommensurable relevant evidence: helping the organisation and interpretation of evidential clues as to the historical evolutionary process. Its basic principles and format could adapt flexibly around new findings, analytical methods, or even entirely new areas of evidence (e.g. microbiome research). Its structure is suitable for reviews of literature of various lengths – from full systematic reviews to introductions in empirical papers. With minor adjustments it could help illuminate adaptive explanations of healthy psychological or physiological traits, or vulnerability to simple diseases. Its usefulness in enhancing evolutionary psychology will be briefly exemplified, although the overall method's primary scientific value is in assessing the phenomena of heterogenous aetiology that remain confusing to psychiatry.

This paper applies the DCIDE framework to assess potential evolutionary explanations of autism [“autism spectrum disorder” (ASD) in current diagnostic manuals; “autism” and other de‐pathologised terminology (Bury et al., 2023) used here where possible]. Autism is exemplified for several reasons. Firstly, it is highly heterogenous, in both phenotypic presentation and known aetiologies (Lord et al., 2020), which is highly problematic (Mottron, 2021), and the DCIDE framework is particularly suitable for untangling such heterogeneity. Second, autism is widely researched, providing a large pool of available evidence. Third, autism presents a profound evolutionary paradox (Keller & Miller, 2006), as one of the most highly heritable mental disorders (Baselmans et al., 2021), and is the subject of various evolutionary hypotheses (Del Giudice, 2018b), which can be assessed to exemplify the framework's potential in discerning between hypotheses. Fourth, autism is a central concern of the “neurodiversity” movement (Kapp et al., 2020) which seeks its reconceptualisation as difference rather than disease, often emphasising associated cognitive strengths. Critics of this movement argue that highly debilitating cases cannot be construed as mere “difference” and warrant separate categorisation (Singer, 2022), potentially by recognising heterogeneous causation in the autism spectrum (Baron‐Cohen, 2019). This indicates the importance of resolving scientific questions in an area of such critical social importance as psychiatry.

(1). DCIDE: describe

The first necessary step to all scientific studies is deciding on a target of investigation. This occurs in Description, which identifies a phenotype to analyse. The DCIDE framework is specifically formulated to evaluate adaptive explanations, so Described phenotypes should ideally be considered plausible candidates for such explanations – however, later stages of Categorisation (Section V.2), Integration (Section V.3) and Evaluation (Section V.5) provide specific strategies for honing coherent targets of explanation. If an inappropriate target of explanation is Described, Evaluation will encourage re‐Description (Fig. 2), and in an iterative process, coherent targets should be identified. Hunt (2023a) considers nuances of altering trait descriptions and downstream effects on evolutionary analysis in more detail.

Initial Description could identify a trait or cluster of traits forming a syndrome (or any biological or psychological characteristic). Modern mental disorder manuals provide descriptivist criteria based on clinical experience and expert consensus (Stein et al., 2010). Research on such conditions may use questionnaires, confirmed diagnosis, self‐reported diagnosis or even self‐diagnosis as inclusion criteria.

(a). Describe: autism

In the case of autism, a diagnosis based on the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM‐5; American Psychiatric Association, 2013) requires persistent deficits in social communication and social interaction across multiple contexts as well as restricted, repetitive patterns of behaviour, interests, or activities. These symptoms must be present early in development and cause clinically significant impairment in social, occupational, or other important areas of current functioning, and not be better explained by another intellectual disability or developmental delay disorder. These criteria can be taken to Describe autism, although research inclusion criteria differ from study to study. Ideally, studies from genetics, epidemiology, and psychometrics would share inclusion criteria, making their results more comparable.

Diagnostic norms shift due to a variety of cultural reasons, including political goals such as access to healthcare and accommodations (Fletcher & O'Brien, 2008). The DSM‐5 combined Asperger's syndrome, Autism and Post‐Developmental‐Disorder Not‐Otherwise‐Specified (PDD‐NOS) into a more general “Autism Spectrum Disorder”, undoubtedly combining multiple aetiologies under a single diagnosis (Lai et al., 2013); researchers now recognise they need to explain the “autisms” (Geschwind & Levitt, 2007; Happé, Ronald & Plomin, 2006). This change has accompanied an apparent rise in autism prevalence as the diagnosis widens to encompass behaviours closer to “normality” (Rødgaard et al., 2019; Idring et al., 2015), with loosening interpretations of core behavioural and social features (Zaroff & Uhm, 2012) and leading to much lower rates of accompanying intellectual disability (Lyall et al., 2017).

Although this has been reported by some as implying the “autism” diagnosis will eventually become meaningless (Knapton, 2019), there is in principle no loosening, tightening, subtyping or removing of subtypes that makes a Described trait beyond the scope of the DCIDE framework. However, loosening criteria can increase causal heterogeneity. Tackling this begins in Categorisation.

(2). DCIDE: categorise

It is widely acknowledged that single labels such as “autism” or “schizophrenia” contain individuals whose qualifying symptoms have distinct causes – for example, there are hundreds of distinct de novo mutations causing phenotypes that can be classified as “autism” (Yoon et al., 2021). Discovering verifiable dysfunction causing specific cases of autism or correlations with harmful personality traits has led to claims that all autism (Keller & Miller, 2006) and personality differences (Verweij et al., 2012; Zietsch, 2024) are simply maladaptive variation. The DCIDE framework avoids this deduction. That de novo mutations cause phenotypes diagnosable as autism does not mean every instance of autism is caused by de novo mutations, or even that every instance of autism is due to non‐adaptive processes. Down's syndrome is associated with high reported happiness (Skotko, Levine & Goldstein, 2011), traumatic brain injury with aggression (Rao et al., 2009) and Williams syndrome with gregariousness (Järvinen, Korenberg & Bellugi, 2013) – happiness, aggression and gregariousness cannot therefore be assumed as non‐adaptations, and nor are these verifiable dysfunctions relevant to evolutionary explanations of the wider functional traits (albeit potentially valuable for understanding proximate mechanisms – perhaps we can gain insight into gregariousness by studying Williams syndrome).

Such heterogeneity makes unitary explanations impossible for many Described disorder categories, but evolutionary principles provide the framing for a simple initial distinction: between the certainly non‐adaptive and the plausibly adaptive. The purpose of Categorisation is distinguishing this major class of heterogeneity in causation.

Reliance on intuition as to whether cases seem functional or not is insufficient, particularly due to the likelihood of cases of evolutionary mismatches and costly by‐products of adaptations being focussed on in Described criteria. The simple test for Categorisation is to ask whether the trait is inherited or not. Not all heritable traits will be adaptations, but all adaptations are heritable traits. By narrowing analytic focus upon inherited traits, we can distinguish the definitively non‐adaptive from the possibly adaptive. Later steps of the DCIDE framework involve asking more detailed questions about plausible evolutionary history: this is simply not necessary for uninherited traits.

Categorisation needs to occur on a case‐by‐case basis, asking whether the reason a case is meeting the Description criteria is inherited or uninherited factors. In Categorising psychiatric disorder, some cases (often the most debilitating) can be clearly identified as primarily explained by vulnerability, mutation or non‐adaptive genetic factors (Fig. 1). These can be considered sufficiently explained within a biomedical paradigm – they involve significant discernible pathology which certifiably has no evolutionary history, so do not need further DCIDE analysis. They can be categorised as non‐adaptive, considered explained, and excluded, so that the inherited traits which are eligible for adaptive explanation are focussed on.

A few complexities need noting: firstly, some inherited traits are very unlikely to have an adaptive history (e.g. rare mutations which have arisen very recently), and if they can be clearly identified, the variation they cause can also be Categorised as non‐adaptive, despite being inherited (presuming they convey neutral or negative effects on fitness). Second, inherited traits which require adaptive explanation may manifest differently between generations due to environmental influences (including the “environment” of the surrounding genome) and developmental stochasticity, so simply ascertaining that a Described trait is not itself visible in parents is insufficient – the question is whether the Described trait is caused by heritable factors. Third, Categorisation removes cases clearly without any need for evolutionary analysis, but identifying some small effect of non‐adaptive processes (e.g. toxins) does not mean excluding cases. Every organism is a cumulation of traits with adaptive histories of being shaped by natural selection, and also experiences infections, toxins, and de novo mutations which have small effects on those traits. All eyes have experienced some slight damage, but still need an adaptive explanation. A genetic background of inherited subclinical autistic traits might mean that a set of small‐effect de novo mutations or exposure to environmental toxins could cause diagnosable autism in one person, but would not cause autism in an individual with a different genetic background. In these cases, the inherited traits require an evolutionary analysis, but the specific manifestation does not. The degree to which the Described case is inherited influences the necessity of keeping it in for further analysis. The point of Categorisation is primarily to remove clearly non‐adaptive cases.

Categorisation utilises familiar biomedical variables from neuroscience, genetics and epidemiology (Table 1). In doing so, it utilises the unique benefit of an evolutionary framing in parsing such data, which allows clear conceptual differentiation between “difference” and “disease”. Following the key insight of Wakefield (1992), without this grounding in evolutionary history, any brain differences or genes can be deemed as “disease mechanisms” or “dysfunctional” because of their causation of a state pre‐determined by society to be a disease. Only the evolutionary framing can ground such designations objectively.

Table 1.

Biomedical variables allowing Categorisation as confirmable disruption and thus exclusion from further analysis.

Area of evidence	Indications of non‐adaptive cause
Proximate system (brain)	Differences not caused by inherited ontogenetic processes (e.g. physical damage, tumours, lesions).
Genes	Uninherited genetic variants (e.g. de novo mutations of SNPs, CNVs or chromosomal abnormalities); rare genetic variants with traceable history of persistent harmful effects. Other non‐adaptive alleles identified as successful due to population history, linkage disequilibrium or drift.
Environment	Damage with direct pathological effects (e.g. toxins, physical traumas, infections) not mediated by inherited genetic pathways.

CNV, copy number variation; SNP, single nucleotide polymorphism.

Although methods for distinguishing adaptive and non‐adaptive cause via biomedical variables could be developed much further, a basic outline for Categorising each area of evidence now follows, exemplified with autism.

(a). Brain categorisation

Neuroscientific studies of mental disorders can reveal clear dysfunctions, or differences that blend with normality. The assumption that brain differences associated with diagnosed disorder prove the presence of a brain disease is a serious (if tempting) error (Protopapas & Parrila, 2019). In Categorisation, the question asked of such findings is whether they result from a history of positive selection or vulnerabilities (Fig. 1). The key available heuristic is to ask whether broadly similar brain states or processes are visible in relatives and seem to result from normal heritable processes. We can clearly categorise severe disruptions of an individual's normal ontogenetic developmental process (e.g. toxins, physical traumas, tumours) in this way – we can confirm they are not inherited. Stochastic differences occurring in neurodevelopment are plausible future findings to allow Categorisation as disruption, but will require development of specific methods discerning normal ontogenetic variation occurring due to generally functional alleles from dysfunctional stochastic errors.

(i). Brain categorisation: autism

Common examples of non‐adaptive cases of autism with clear neuropathology are seen in various cases of so‐called “syndromic autism”. For example, Rett syndrome meets an autism diagnosis in between 40 and 97% of cases (Lyst & Bird, 2015). Caused by mutations in the MECP2 gene, it is associated with reduced brain volume, abnormally small and densely packed neurons with reduced dendritic complexity and spine density, and shorter pyramidal neurons in the motor and frontal cortices (Ip, Mellios & Sur, 2018). Categorising these cases is possible by observing their significant contrast with relatives. In cases when syndromes with gross morphological and functional brain differences are inherited, Categorising could require reference to genetics (see Section V.2.b). Evidence converges in such cases – Rett syndrome could be Categorised as a disruption on both neurological and genetic grounds.

Although autism of all forms is undoubtedly associated with brain differences, rarely does neuroscience discover autism caused by tumours or brain trauma (partly because criteria require symptoms from an early age). Certain syndromic cases show exceptional neuropathology but the majority of cases, particularly without intellectual disability, show diffuse and complex differences in brain growth and connectivity (Muhle et al., 2018), plausibly visible to selection via genetics and selectively neutral or positively selected (Fig. 1). Most cases of autism thus cannot be Categorised as explained by disruption at the neuroscientific level.

(b). Genetic categorisation

Although tracing direct causal pathways from genotypes to phenotypes is highly complex, finding genetic associations with traits is now possible, and we can use genetics to infer basic facts about a trait's evolutionary history (Schraiber & Akey, 2015). Most common mental disorders have a genetic component (Baselmans et al., 2021). Such apparently harmful genetic predispositions are explainable as positively selected (e.g. caused by antagonistic pleiotropy, mismatch, life‐history trade‐offs or balancing selection), neutral, or awaiting negative selection (e.g. under mutation–selection balance or genetic drift) (Durisko et al., 2016).

Genetic evidence is critical in Categorisation. Alleles are usually classified as common, rare or de novo (although researchers differ on the prevalence cut‐off between common and rare (Raychaudhuri, 2011; Glinton & Elsea, 2019)). Large‐effect de novo variants, including single‐nucleotide polymorphisms (SNPs) or copy number variants (CNVs) are the simplest cases explainable as non‐adaptive mutations (Fig. 1), because de novo variants have no history of reproductive success by definition.

Rare and especially common variants are more complicated, as they have been inherited, and are thus reproductively successful – they have increased in frequency since arising de novo, rather than being selected out – perhaps because they are adaptive. Their individual effects on fitness may be neutral and drifting, positive and selected, or weakly negative but persisting due to inefficient selection (Fig. 1). Approximately 130,000 variants (∼4% of human genomic variation) are shared with 0.5–5% of the population (Bourgeron, 2015). It is possible these result from positive or balancing selection (see Section V.3.e ), so ideally they require individual analysis. Specific techniques from population genetics (Adrion et al., 2020) may potentially be used to analyse whether such alleles have increased in frequency due to hitchhiking, population growth or bottlenecks, or simple drift and inefficient negative selection, and Described traits explained entirely by these processes can be Categorised as non‐adaptive. We suggest a placeholder heuristic that rare alleles (Hill et al., 2018) present in less than 0.5% of the population are likely non‐adaptive.

For inherited traits caused by common alleles, adaptive explanations become plausible. Ubiquitous or common genetic variants are evidence of significant historical reproductive success (inferred from their significant increase in frequency since arising de novo) and an apparent lack of negative selection (Uher, 2009). Presuming broad‐sense heritability is correctly and not over‐estimated, this also applies in cases of “missing heritability” (Manolio et al., 2009). Despite their potential to discover additive genetic variation, GWAS lack the power to detect specific associated alleles if there are complex gene‐by‐gene or gene‐by‐environment interactions (Parker & Palmer, 2011), expected of many adaptations, especially of highly plastic (e.g. psychological) phenotypes. Trait heritability observed in population and quantitative genetics need not be fully explained by molecular genetics to be meaningful and imply adaptive explanation is required. Although common alleles are not sufficient to prove adaptation (for example, alleles may have drifted to high frequencies or fixation, or be selfishly replicating transposons) they are at least necessary for it – we cannot Categorise traits caused by common alleles as non‐adaptive without further analysis.

(i). Genetic categorisation: autism

Autism has high heritability – approximately 80–90% (Baselmans et al., 2021). Subgrouping the autism spectrum by genetic evidence is crucial (Lai et al., 2013; Geschwind & State, 2015). In a large Swedish study, Gaugler et al. (2014) estimated approximately 6% of cases of autism are caused by de novo or rare genetic variants (excluding confirmable syndromes such as Williams or Downs). Glinton & Elsea (2019) estimate over 400 forms of “monogenic” autism exist, caused by de novo CNVs and de novo single nucleotide variants (SNVs), and estimate that these forms account for 10–20% of all autism cases. These cases are often highly penetrant and syndromic, with various other cognitive and physical effects, often associated with intellectual disability (Ronemus et al., 2014; Iossifov et al., 2015; Robinson et al., 2014). These cases can be Categorised as disruptions and considered explained. Authors such as Mitchell (2015) develop Keller & Miller's (2006) mutation–selection balance account to propose all unexplained cases of autism may similarly be caused by disruptive rare mutations, potentially of digenic or oligogenic causality, avoiding detection. If so, Categorisation of autism will continuously reduce the variance eligible for adaptive explanation. However, further developments in molecular genetics have failed to find sufficient causal variants (Keller, 2018), and indeed find distinct genetic underpinnings for cases of autism with distinct phenotypic signatures: common SNPs being associated with higher intelligence, less developmental delays and more core autism features whilst de novo mutations associate with lower intelligence and more developmental delay (Warrier et al., 2022). This supports the long‐standing recognition that familial cases associate with normal or higher intelligence (DeLong & Dwyer, 1988), and implies that explaining autism‐causing alleles requires reference to both non‐adaptive and adaptive explanations.

(c). Environmental categorisation

Cases of environmental causes that Categorise as disruption include physical trauma, infections and toxins, and various widely recognised extrinsic causes of disease. Categorising a Described trait as a disruption by environmental evidence requires certifying that an environmental factor is causal without acting via an adaptive mechanism. Reactions to environments can often be fitness‐enhancing examples of adaptive plasticity. Evidently fitness‐reducing environmental factors (e.g. bullet wounds) can be causal of adaptations (e.g. blood clotting and healing), so identifying harmful environmental causes is not sufficient to prove any downstream effects are fitness‐reducing disruptions. Assuming that environmental influences such as stress and trauma, which cause harmful phenotypes, are self‐evidently diseases is tempting, but that need not be the case (Belsky, 2012). Even in cases harming health and fitness in a particular individual [possible through chronic exposure, e.g. to stress (Beauchaine et al., 2011) or unusual underexposure (Seery, Holman & Silver, 2010)], such maladaptive manifestations can be by‐products of adaptation and require adaptive explanation.

The critical question is whether the precise aspect of the phenotype being Described (for further discussion see Hunt, 2023a) and resulting from the environmental factor is ontogenetically mediated and the mechanism could plausibly result from a history of positive reproductive success (bleeding from a bullet wound could not; blood clots could). Categorising by environment therefore ideally requires identifying environmental factors (likely correlating with exposure prior to Described trait appearance) and assessing the pathway of causation.

(i). Environmental categorisation: autism

Modabbernia, Velthorst & Reichenberg (2017) reviewed meta‐analyses and systematic reviews of environmental risk factors for autism. Perinatal and neonatal risk factors (Gardener, Spiegelman & Buka, 2011) include birth injury or trauma [relative risk (RR): 4.90], maternal haemorrhage (RR: 2.39) and umbilical cord complications (RR: 1.5). There is some evidence that heavy metals, most importantly inorganic mercury and lead, play a role in some cases. Valproate use during pregnancy causes foetal valproate syndrome, which regularly meets autism diagnostic criteria (RR: 2.9–5.2; Christensen et al., 2013). These types of cause can be comfortably Categorised as disruptions; they involve identifiable pathological processes with large individual effects and no potential functional history or direct inheritance. There are, to our knowledge, no good summary estimates of the proportion of autistic cases which would be Categorised via these mechanisms, but they are small in comparison to genetic factors.

Another key environmental factor associated with autism is increasing parental age: every 10‐year increase in age increases ASD risk by about 18% for mothers and 21% for fathers (Wu et al., 2017). Fathers over 50 have more than twice the chance of having an autistic child compared with fathers under 30 (Hultman et al., 2011). This affects autism without intellectual disability as well (Tsuchiya et al., 2008; Croen et al., 2007). There have been attempts to find disease processes behind this effect. Older parents give birth to children with more mutations, but de novo mutations cannot account for most of the increased risk (Gratten et al., 2016; Taylor et al., 2019). Alongside other subtler risk factors, such as maternal infection during pregnancy and maternal obesity, the obscurity of these pathways and the possibility they result via positively selected heritable mechanisms means these additional cases cannot be Categorised as non‐adaptive without further research.

(d). Concluding categorisation

Categorisation resolves a major fundamental form of causal heterogeneity, distinguishing Described trait variance explainable by dysfunction and non‐adaptive processes from potentially adaptive variance requiring further analysis.

In the case of autism, broad diagnostic criteria capture disparate phenotypes and aetiologies. Social and behavioural capacities are clearly highly sensitive to interruption via various pathways. Genetic syndromes linked to specific CNVs or rare or de novo mutations, and cases caused by early life traumas or prenatal toxin exposure are obvious cases of autism which Categorise as disruptions. Less‐common individual cases of unusual brain development or other discoverable pathologies may exist and could Categorise as disruptions. Confirmable disruptive cases are more often associated with intellectual disability and co‐occurring health conditions such as epilepsy (Lord et al., 2021). They likely account for between 5 and 20% of currently diagnosable autistic cases, although this is highly dependent on inclusion criteria, which has recently been expanding to include less intellectually disabled cases (Lyall et al., 2017). The majority of cases of autism, however, Categorise as potentially requiring adaptive explanation. Investigating this is the task of the remainder of the DCIDE framework.

(3). DCIDE: integrate

Integration aims to hone Description beyond disorder categories to broader phenotypes visible to selection. Reproductively successful heritable traits are not necessarily adaptations themselves. Evolutionary medicine and psychiatry recognise various by‐products of adaptations causing harm and diagnosed as disorder. Adaptive alleles can cause various harmful by‐products via: costly trade‐offs; mismatch to novel environments; maladaptive extremes of adaptive spectrums; maladaptive outcomes of developmentally plastic adaptive systems; pleiotropic effects including sexual antagonism; heterozygote advantage; and more (Abed & St John‐Smith, 2022a).

By‐products are reproductively successful via related adaptations without lending positive fitness themselves. They can be common, heritable, and in themselves costly, but persist and be positively selected nonetheless, especially if the benefits of a related adaptation historically outweighed their harms and selection could not (or randomly did not) mitigate them. They could be caused by common alleles, or be universal amongst a species. Their costs may incline us towards diagnosing them as simple disorder, whilst showing no traditionally recognised signs of pathology – they can be direct products of healthy function – confusing normal biomedical investigation.

Adaptationist accounts are necessary for explaining by‐products. If anxiety disorders are by‐products of adaptive anxiety, explaining anxiety disorders requires referencing adaptive anxiety (Nesse, 2023). Explaining obesity requires understanding the adaptation of fat storage (Wells, 2006). Explaining sickle cell disease requires understanding malaria resistance. Where disorder is by‐product, the related adaptation needs identification for explanation. Problematically, disorder‐focused Description may include by‐products but exclude relevant adaptations (e.g. Describing sickle cell disease ignores heterozygous individuals), making ultimate explanation impossible. If this has been a common problem, and many mental disorders are by‐products of adaptations, it would explain why previous psychiatric paradigms could never sufficiently explain disorders – they were ignoring the relevant related adaptations, and the biological mechanisms were not simply pathological. This is also a major problem with unsophisticated approaches to evolutionary psychiatry: the temptation is to explain “schizophrenia”, “anorexia” or any of the other diagnostic labels. Nesse (2023) calls this the mistake of “viewing disorders as adaptations”. If these are by‐products, the adaptation which conveys the vulnerability to the disorder may be entirely missed from the analysis.

To mitigate the possibility of Description criteria containing by‐products and excluding relevant adaptations, Integration assesses a trait's visibility to selection (the extent to which it can affect reproductive success) via various areas of evidence (Table 2). These “visibility variables” allow inference of selective pressures applied to a given phenotype. Very low visibility in any area implies adaptations could be missed, and so an Integrated trait including genetic and phenotypic correlations with greater visibility to selection should be analysed along with the originally Described trait (Fig. 3). For example, because schizophrenia is so rare (Hunt & Jaeggi, 2022), and anorexia seems related to specific novel environmental contexts (Abed et al., 2012), they are not plausible targets for adaptive explanation in themselves. The phenotype predominantly visible to selection needs consideration for effective evolutionary explanations: the more common correlates of schizophrenia and cross‐culturally invariant traits behind anorexia need to be analysed. Distinguishing the adaptation from the by‐product occurs in following stages: Integration is purely about ascertaining that the correct target of explanation is in view.

Table 2.

Visibility variables with their implications of high or very low visibility to selection.

	Area of evidence	Indications of high visibility to selection	Indications of very low visibility to selection
Phenotype features	Age of effect	Active before and prime years of fertility and child‐raising	Effects in old age, especially after grandparental support declines
	Duration	Long duration	Duration too insignificant to affect reproductive success
	Sex differences	Present in both sexes	Low visibility in one sex; none in the other
Causal factors	Environmental plasticity	Appears and reacts under evolutionarily conserved environmental features	Contingent on evolutionarily novel environmental features
	Prevalence of genetic propensity	High prevalence	Too rare to exist in most hunter–gatherer agglomerated‐band social groups

Note that environmental visibility variables differ substantially from environmental biomedical variables in their lack of reference to biological mechanisms. Phenotype features are combinatorial (e.g. longer duration means more visibility to selection, but not if only in old age).

Fig. 3.

Genes have multiple effects across the visibility variables, and Integrating phenotypes most visible to selection provides the appropriate target of adaptive explanation. Dark coloured ellipses indicate targets of high visibility – see Table 2 for details.

(a). Integrating by age of effect

Natural selection acts on reproductive success, so the ages leading up to and during reproductive potential are the most important, that is youth and early‐to‐mid adulthood (Williams, 1957), and visibility to selection is highest around that age (Hamilton, 1966). The relative importance of early life is so high that deterioration in aging could be a by‐product of genes increasing early life fitness, that is ‘antagonistic pleiotropy' (Austad & Hoffman, 2018). The longer life goes on, the higher the likelihood of prior reproduction and cumulative extrinsic mortality risk, reducing that age's visibility to selection.

Generally, seeking earlier correlates of late‐onset traits optimises analysis for visibility to selection. The likelihood of a Described trait being an adaptation decreases the further it appears after reproductive age. Grandparents can increase their inclusive fitness, so traits appearing in post‐reproductive age may be adaptations in exceptional cases (Hill & Hurtado, 2009), but Integration of effects during youth and prime reproductive age is usually necessary (Brown & Aktipis, 2015).

(i). Integrating by age of effect: autism

Autism is an early‐onset developmental condition, present essentially from birth, with lifelong duration (Lord et al., 2021). Relevant to later Evaluation of function and by‐product effects (see Section V.5), symptoms often change over the lifespan (Waizbard‐Bartov & Miller, 2023), but as age of effect makes it almost completely visible to natural selection, it does not require Integration of earlier correlates.

(b). Integrating by duration

Integrating correlates by duration encourages the inclusion of longer‐lasting phenotypes. Described traits could be invisible to selection if they are sufficiently short‐lasting (depending on the phenotype this may be minutes or months). Stochastic temporary bodily or psychological phenomena may occasionally arise as by‐products. For example, adaptive capacities for fear can result in startles when random harmless sights or sounds are briefly misinterpreted as representing dangers; a specific instance can be useless or harmful, but over a whole life course the system is adaptive. Integrating correlated traits that last long enough to have significant effects on reproductive success increases the likelihood of including relevant adaptations.

(i). Integrating by duration: autism

Autism is associated with cognitive and behavioural effects of various durations (e.g. bursts of behaviour in meltdowns, phases of obsession, lifelong difficulties in certain specific social domains (Lord et al., 2021); also see Section V.5.c.i). Overall, this makes it highly visible to selection by duration criteria and does not need Integration of longer‐lasting correlates.

(c). Integrating by sex differences

Selection upon (non‐Y chromosome) genes occurs in both sexes. Identical DNA can lead to similar or different phenotypes between sexes, and constraints can result in manifestation of sexually selected traits in both sexes despite positive selection in only one sex (Wittman et al., 2021). Sexual selection is unlikely to optimise phenotypes to the extent that the sexually selected trait only has effects in one sex (Lande, 1980; Williams & Carroll, 2009), so by‐product effects are expected (when the manifestation in the other sex is costly, we label it “sexually antagonistic”).

Sex difference Integration forces consideration of whether sex‐biased traits are the result of genes with different effects in the opposite sex, and attempts to include the manifestation between sexes. If single‐sex traits with low visibility (e.g. appearing late in life in females) are Described, possible higher visibility in the other sex should be carefully assessed (e.g. it is possible late‐life female manifestation is a by‐product of early‐life male adaptations).

(i). Integrating by sex differences: autism

Autism shows a pronounced sex ratio difference, with estimates of a 3:1 male to female ratio (Solmi et al., 2022). However, females may be more likely to “camouflage” and show different symptom profiles which are then underdiagnosed (Hull, Petrides & Mandy 2020). Indeed, females require a higher polygenic score to meet autism diagnostic criteria (Warrier et al., 2022). Selection on autism acts on the sum of the male and female manifestation, even if male presentation meets diagnostic criteria more regularly, so Integrating camouflaged female cases would better reflect total phenotypic impact of autism genes for informing adaptive explanations.

(d). Integrating by environmental plasticity

Environmental factors leading to trait manifestation imply trait visibility to selection if they are common and ancestrally relevant (e.g. stress, diet, infection). Traits appearing specifically in evolutionarily novel environments (e.g. high sugar diets, written language) cannot be an adaptation for that specific environmental factor. Disorders appearing in these specific circumstances are products of mismatch (Chaudhary & Salali, 2022; Lea et al., 2023). When affected by novel environments, Integrating the relevant adaptation requires identifying the system's manifestation as visible to ancestral selection: its ancestral reaction norm. Obesity only exists in high numbers in evolutionarily novel contexts, so fat tissue growth in response to excess calories is the relevant plausible adaptation.

(i). Integrating by environmental plasticity: autism

The question of whether autism exists or presents differently outside of novel environments is essentially unresolved because of lack of research in small‐scale societies (Gurven & Lieberman, 2020). Among industrialised societies, autism prevalence differs quite significantly (Solmi et al., 2022), but this is likely due mainly to diagnostic uncertainty and cultural differences in condoned behaviour (Zaroff & Uhm, 2012). Further cross‐cultural research would help in confirming whether and how autistic traits show visibility across cultures (and, presumably, evolutionary history). Otherwise, advancing parental age is a known prominent relevant factor (see Section V.2.c.i ), but as hunter–gatherer adults do have children into late life and autistic children are born to young couples, this does not imply autism is a product of mismatch or force Integration.

(e). Integrating by prevalence of genetic propensity

Adaptive genotypes often reach fixation and become species‐wide, and thus are clearly visible to selection. However, important counter‐examples exist of adaptive strategies via individual differences, especially in cases of social selection, social niche specialisation and negative‐frequency dependency (Hunt & Jaeggi, 2022; Martin, Jaeggi & Koski, 2023), which can be maintained at the genetic level by balancing selection (Wolf et al., 2007). Such strategies cannot be arbitrarily rare: an approximate “minimum adaptive prevalence” is of one individual per social group (Hunt & Jaeggi, 2022).

Integration already seeks relevant within‐individual effects of genotypes by considering duration and age of effect, but in cases of heritable individual differences, between‐individual effects are also important. Selection acts on the total effects of genes across a population, so their visibility to selection depends on the sum of their common and rare manifestation. Higher prevalence indicates higher visibility to selection, so Integrating more common correlates of rare traits is a useful strategy for enabling adaptive explanations – Description criteria may be excluding adaptations. This may be particularly relevant for explaining rare psychopathological traits with more common subclinical correlates. Simplistically, an allele with certain effects in 5% of individuals and different (perhaps more extreme or harmful) effects in 1% of individuals is five times more visible to selection in the 5% (although specific phenotypic effects may mean stronger selection in the rarer cases).

(i). Integrating by prevalence of genetic propensity: autism

Recent epidemiological studies of autism find a prevalence of around 1% (Fombonne, Macfarlane & Salem, 2021) although rates differ widely by study, ranging from 0.01 to 4.36% (Zeidan et al., 2022). In the UK, rates increased 787% between 1998 and 2018 (Russell et al., 2021), primarily because inclusion criteria loosened. Cases included now are disproportionately without intellectual disability (Lyall et al., 2017), and more likely to Categorise as requiring adaptive explanation.

At the 1% rate, autism meets minimum visibility criteria – given typical hunter–gatherer group sizes of approximately 165 (Hamilton et al., 2007), about one or two of these autistic individuals are on average expected in every extended social group (considered further in Section V.5.f). However, although adaptations can plausibly persist in around 1% of the population, they are more likely to do so at higher prevalence, implying Integration of more common genetically correlated traits is necessary. This could include cases incorporated in more permissive inclusion criteria and subclinical traits distributed throughout the population, especially in family members, which has led to the identification of a Broad Autism Phenotype (BAP) in 5–9% of the population (Morrison et al., 2018; Sasson et al., 2013). The genetic basis for autism and the BAP is shared, so capturing the relevant selected phenotype requires Integrating this wider autism spectrum – it is possible that adaptation visible in the BAP explains autism as a by‐product.

(f). Concluding integration

Integration expands Described traits to incorporate relevant correlated phenotypes using visibility variables. After Integration the paradox of common, harmful, heritable mental disorders is refined into the paradox of mental disorders with early onset, long‐lasting, ancestrally activated, common, heritable correlates, with no discoverable pathology.

In autism's case, its age of effect and duration imply high visibility to selection. Environmental evidence does not suggest complete novelty. However, its prevalence and sex differences imply current criteria may be missing relevant related phenotypes, so the Integrated trait should include subclinical cases – both camouflaging cases in females and broader autism phenotype traits in family members. Their shared genetics require shared evolutionary explanation.

Unfortunately, prior research into autism largely has not Integrated autism's subclinical or female manifestations, which are under‐researched (Hull et al., 2020). Nonetheless, to exemplify the DCIDE framework, Depiction and Evaluation will proceed, recognising the limitations of available evidence – with a strong recommendation that future empirical investigation into autism's origins consider effects of the genetic predisposition across the population.

(4). DCIDE: depict

Once an appropriate target of explanation is identified, the task which evolutionists have long been interested in, of hypothesising about the functional design of adaptations, is possible. In Depiction, explanatory hypotheses are presented to explain why the Described, Categorised, and Integrated phenotype persists in the population. There is a rich history and varied methods across evolutionary biology, psychology and anthropology in developing functional hypotheses for observed phenomena. As a minimum criterion for a Depiction, processes of gene, individual or group dynamics explaining the reproductive success of the Described and Integrated traits are sought. These will often reference function, but could also reference non‐adaptive processes such as genetic drift. When medical or psychiatric conditions are under consideration, functional effects should plausibly justify any by‐product dysfunctional effects or associated vulnerabilities.

As noted in Section II, hitherto evolutionary hypothesising, on psychiatric traits and elsewhere, has often referenced varied evidence in an unprincipled way, potentially cherry‐picking and becoming open to “just‐so storytelling” critiques. The DCIDE framework tackles this by systematically referencing “direct evidence” in Depiction and “circumstantial evidence” in Evaluation. Direct evidence is supposed to be of the function in action; the eye seeing, the wing flying, the jealousy causing mate guarding. This is the evidence often sought by evolutionary psychology (e.g. in experiments designed to reveal the mechanics of cognitive faculties such as anger; Sell et al., 2017), or evolutionary biology (e.g. in observing that peppered moth Biston betularia wing colour affects survival; Majerus, 2008). The key aim of Depictions is to reference direct evidence of the functioning adaptation (or, if proposing a non‐adaptive hypothesis, direct evidence of the trait's neutral or negative effects on fitness), proposing how it attained positive fitness and thus shaped the observed traits via natural selection. Varying relevance of potential forms of direct evidence and the circumstances of evidence collection are noted in Fig. 4. In Evaluation, Depicted hypotheses will be systematically compared and assessed using circumstantial evidence, potentially including biomedical and visibility variables, which have specific inferential connotations. Hypotheses are eventually judged on their sufficiency to explain this totality of evidence.

Fig. 4.

The varying relevance of direct evidence of functioning traits, also considering the circumstances within which that evidence is gathered.

This basic structure of analysis could be adapted to support or question existing hypotheses in evolutionary psychology or non‐evolutionary hypotheses of mental disorders. The multitude of proximate biological or psychological accounts [e.g. referencing lower oxytocin levels (John & Jaeggi, 2021) or weak central coherence in autism (Frith & Happé, 1994)] would, however, need expansion to explain heritable persistence in the species, which is often overlooked. Pre‐existing adaptive accounts are disadvantaged if they rely purely on diagnostic criteria, lacking Categorisation of subtypes or Integration of correlated phenotypes. In autism's case, various accounts have recognised dysfunctional subtypes and the heritable spectrum (informally applying Categorisation and Integration). However, the wider literature of evidence they draw from relies on varying inclusion criteria, limiting interpretability of the analysis. Future research should correct this, but the basic structure of the DCIDE framework will remain a useful tool to incorporate various areas of evidence in a principled way.

(a). Depict: autism

Hypotheses presented to explain autism are numerous. Here we Depict three prominent accounts (Baron‐Cohen, 2008, 2020; Crespi, 2016; Del Giudice, 2018b) referencing evolutionary function in their explanation of autism. They agree on symptoms of social disability as costly by‐products, but specifics regarding functional processes and evolutionary forces differ (Table 3). Classic hypotheses (e.g. of refrigerator mothers; Kanner, 1968) or recent neurobiological hypotheses (e.g. Markram & Markram, 2010) could equally be Depicted and Evaluated if expanded to address the evolutionary paradox of autism's persistence.

Table 3.

A summary of competing Depictions of autism direct evidence and adaptive explanations.

Author	Direct evidence	Adaptive explanation
Baron‐Cohen (2008, 2020)	Lack social abilities but excel in systemising abilities; family members often successful in systemising careers	Systemising ability filled an inventing/expert social niche; systemising overexpressed without balancing empathising abilities in individuals with autism
Del Giudice (2018b)	Systemising abilities; rational thinking; tendency towards committed, long‐lasting relationships	Slow life‐history skilled/provisioning strategy, with systemising strengths and altered relationship patterns as valuable for reproductive success and inclusive fitness
Crespi (2016)	Enhanced abilities in specific areas of intelligence; particular benefits of correlated traits across the population	Strong recent positive selection on human intelligence, cases of autism as a by‐product via overexpression and vulnerability

Baron‐Cohen (2008, 2020) frames autistic traits as an exaggeration of “if–then” systematic thinking, which exists on an empathising–systemising dimension. Autistics are hypoempathisers, lacking various empathic abilities including complex emotion recognition, recognition of faux pas and spontaneous ascription of internal states, but hypersystemisers, with associated advantages in folk physics and attention to detail, sometimes allowing them to achieve high levels of domain expertise in fields like mathematics, physics or computer science. Obsessions cluster in systemising domains. Regarding the broader autism phenotype he points to evidence of relatives of autistics being engineers, mathematicians and scientists, strong systemisers, and that scientists and mathematicians score higher on the Autism Spectrum Quotient. Baron‐Cohen's (2008, 2020) general hypothesis is that high systemisers filled an ancestral social niche as tool‐makers, inventors and experts in areas of their obsession. He accounts for disabling cases meeting autism diagnostic criteria (without Categorisable pathology) as resulting from systemising overexpression and a lack of empathising, perhaps due to assortative mating.

Del Giudice (2018b) explicitly addresses autistic heterogeneity and the necessity of differentiating cases caused by mutations and associated with intellectual disability from cases caused by common alleles, also citing evidence that those common alleles are associated with higher IQ and educational attainment in the general population. He references Baron‐Cohen's (2008) systemising account of autistic strengths and talents, adding that autistic symptoms relate to a logical, deliberative style of reasoning, and embellishes on the account by arguing for autism as a slow life‐history skilled/provisioning reproductive strategy, particularly for males. Elsewhere he suggests systemising abilities would be sexually selected as attractive traits in males (Del Giudice et al., 2010), and that autistic‐like traits delay reproduction and increase potential investment in family, noting that autistic‐like traits correlate with restricted sociosexuality, preserved or heightened interests in romantic relationships, and increased investment in long‐term partners. Maladaptive cases again arise from overexpression, exacerbated by assortative mating, mutations and environmental insults. Baron‐Cohen and Del Giudice therefore agree on autistic traits as cognitive specialisations (Hunt & Jaeggi, 2022) but differ on specifics of function.

Crespi (2016) provides a “high intelligence imbalance” hypothesis, where strong recent selection for intelligence in humans leads to occasional dysregulation and autism as a by‐product. Specifically, he relates autism to exaggerated “perceptual” intelligence. He cites evidence of autistic reductions in verbal skills but increase in focus of attention, enhanced perceptual and spatial abilities and superior ability in non‐rotational aspects of the mental rotation task. He further notes that autism genetically correlates with IQ, childhood IQ, college attendance and years of education, cognitive function in childhood and educational attainment, and verbal–numerical reasoning and educational level reached. Sensory discrimination and acuity are enhanced. He notes that family members of autistics share cognitive strength profiles. These strengths are paradoxical in the face of apparently lower autistic IQ, on average, on most standardised tests. Referencing various neurological similarities in brain size and growth, connectivity and neuronal function between high intelligence and autism, he hypothesises that instances of autism are costly by‐products of general selection for intelligence in humans. Thus, despite noting the same cognitive strengths as relevant to adaptive processes causing autism, the hypothesis does not claim that autistic or BAP individuals were filling a particular social niche.

(5). DCIDE: evaluate

The goal of Evaluation is deciding the optimum coherent explanation given the available evidence. A systematic standardised method is critical here, to avoid just‐so storytelling or cherry‐picking, and allow fair comparison of Depicted hypotheses. The stages of Description, Categorisation and Integration are particularly critical for evolutionary explanations of disorders, which are likely to include specific non‐adaptive cases and exclude relevant adaptations from initial criteria. Systematic Depiction and Evaluation is useful across all traits, disorders or obvious adaptations. This is where George Williams' call for rigorous methodology that goes beyond intuition or analogies to engineering can be answered, and hypotheses can be tested with standardisable interpretations of objective evidence.

Evaluation proceeds in three steps (Fig. 5). The first two are standard existing scientific practice not requiring extended exposition. Firstly, asking about reliability. Do Depictions reference reliable direct evidence (e.g. reproducible, replicable, methodologically sufficient, evolutionarily relevant) and reliable explanatory theory (e.g. does not rely on naïve group selection)? Second, asking whether competing Depictions' direct evidence is compatible. Does any referenced reliable evidence falsify a competing hypothesis (e.g. one hypothesis claims jealousy enhances affectionate attachment but another cites evidence that jealousy causes fear and reduces affection)? Once viable Depictions remain, the third, more novel stage of Evaluation progresses. Do the Depicted hypotheses sufficiently explain the circumstantial evidence? Existing evolutionary hypotheses have often cited areas of circumstantial evidence [e.g. male preponderance and low prevalence to justify sociopathy as a cheating strategy (Mealey, 1995) or cross‐culturally shared sex differences to justify jealousy as a mate guarding and retention strategy (Buunk et al., 1996)] but the DCIDE framework's Evaluation assesses such evidence systematically, utilising standardised implications from each area of evidence via functional or by‐product explanations (Table 4).

Fig. 5.

Three steps for Evaluation. Possible outcomes of each step, represented by arrows, can occur simultaneously (e.g. a Depiction can mostly explain the circumstantial evidence, and to some extent be explained, but also show mixed results requiring updating of Description).

Table 4.

Examples of circumstantial evidence and the general Evaluative implications for each area of evidence to be explained as either functional or by‐product.

Area of evidence	Functional explanation	By‐product explanation	Example considerations
Age of effect	Adaptation for that stage of life history	Appears at ages when non‐functional	Pre‐reproductive age; adolescence and early reproduction; prime reproductive age; grandparental investment
Duration	Adaptation functions for this duration	Has non‐functional effects for this duration	Lifelong traits; occurrence over long (e.g. life stages), medium (e.g. seasons) or short term (e.g. immediate events)
Sex differences	Sex‐specific function	Selected for in other sex; consequences of other sex effects	Arising in sexual‐selection specific circumstances (e.g. sexual competition); infidelity; sexual jealousy; sexual attraction; childcare
Environmental plasticity	Functional reaction; adaptive developmental plasticity; degree of plastic strategy	Non‐adaptive consequences of plasticity; mismatch	Presentation cross culturally. Effects of ancestrally relevant factors (e.g. diet; illness; stress; social status; relationship status; interpersonal conflict; entrapped; danger; opportunity)
Prevalence of genetic propensity	Functional strategy optimised for a particular prevalence within a group	Appears in cases and groups where non‐functional	0.5–1% = one individual per extended social group; 3–5% = one individual per band; 10–20% = one individual per family/friend group, 50% = one in a pair; 100% = everyone. (Intermediate percentages unclear)

Circumstantial evidence can subsume all non‐direct evidence associated with Described and Integrated traits, but ideally they should have standardisable interpretations (e.g. Tables 1, 2 and 4). The critical question is whether Depictions sufficiently explain that circumstantial evidence. Visibility variables alone are considered as relevant circumstantial evidence herein, but biomedical variables could also be utilised (e.g. genetic selection histories implying environments of adaptiveness; observed pleiotropic effects via mechanisms such as hormones implying trade‐offs).

Visibility variables are circumstantial facts applying to every phenotypic characteristic. Function depends upon visibility: a function can only occur to the extent that it is visible (e.g. at a particular age, for a particular duration). By‐products are visible, but that visibility needs to be excused as non‐functional. Advantageously, visibility variables have fairly standard implications (e.g. traits appearing for 10 min versus several years, or in early versus late reproductive age, are eligible for specific functional or by‐product explanations; Table 4). This requires framing in terms relevant to human evolution rather than modern descriptive norms: for example, age of effect should be considered in terms of different human life stages instead of decade (Haig, 2010); prevalence in terms of absolute numbers per human ancestral group size instead of percentages (Hunt & Jaeggi, 2022). Depictions can be compared in their sufficiency of explaining the specific visibility of Described and Integrated traits; the correct hypothesis should explain them coherently. Insufficiency here should appear either as lacking explanation or just‐so storytelling, excessively wrangling the explanation to fit the evidence without prior theoretical grounds. The correct hypothesis should find each circumstantial fact justifying rather than challenging.

Three main possibilities arise during Evaluation (Fig. 2). Described and Integrated traits may be considered sufficiently explained; further evidence or development of hypotheses may be required; or the results may be mixed (likely if attempting to explain multiple adaptations simultaneously) and a different Description required. These are not mutually exclusive. Part of the explanation may be satisfactory, other questions left unanswered, and some subset of cases not fitting that explanation, requiring re‐Description.

A notable point of philosophy of science is that each stage of Evaluation, like all science, eventually relies on the subjective judgement of the scientist/s as to whether the evidence or theory is adequately supportive. We should be able to agree on objective observations, but not necessarily its interpretation as evidence for or against a particular hypothesis: different parties can maintain different opinions or standards. In Evaluation, questions of whether direct evidence is amply replicated, or methodology is appropriate, or theory is sound, are debatable in themselves. Judgements of compatibility, whether Depictions cite coherent or contradictory direct evidence, could vary. Although circumstantial evidence carries standardised theory‐derived implications, in considering whether each area of evidence is sufficiently explained by a particular Depiction, again there can be disagreement. It is exactly because we rely ultimately on subjective evaluation of hypotheses' worth that a structure within which to engage in analysis and disagreement is valuable. Evaluation makes explicit the distinct points of theory and evidence that can be questioned in assessing the validity of a Depicted hypothesis – this is useful to encourage productive debate, and prevent the problems of cherry‐picking and unsophisticated hypothesising, but is no guarantee of unequivocal agreement.

In this review, Evaluation has important limitations. Neither empirical research nor theoretical models have been formulated specifically for a DCIDE analysis (e.g. autism research has largely concentrated on pre‐reproductive age; Integrated traits are rarely considered). Depiction's accounts of evidence are necessarily assumed if unmentioned by the authors. Despite these limitations, exemplifying Evaluation is essential, and assessing Depictions of autism still leads to useful conclusions.

(a). Evaluate depiction reliability and compatibility: autism

Initially, Depiction reliability and compatibility must be assessed. It should be noted that most cited studies come from before the replication crisis (Shrout & Rodgers, 2018) and improvement in practices (e.g. open data and pre‐registration), so Evaluation may change substantially if evidence is revised.

In terms of reliability, direct evidence of autistic strengths is plagued by complexity and heterogeneity between individuals, but holds up to meta‐analysis (Muth, Hönekopp & Falter, 2014). Reliance upon novel experimental tests implies questionable evolutionary relevance (Fig. 4), but relationships between successful technical careers and systemising (Baron‐Cohen & Hammer, 1997; Baron‐Cohen et al., 2001) imply general aptitudes in valued areas with possible ancestral correlates. Expanding research to non‐industrialised societies would dramatically improve confidence here. The reliability of evidence for autistic mating and parenting strategies seems questionable, with Del Giudice et al.'s (2010) study, using a relationship questionnaire in 199 students, correlating autistic‐like traits with a preference for committed long‐term relationships, while the broad autism phenotype has also been associated with avoidant and less‐secure romantic attachment (Lamport & Turner, 2014). Due to this mixed evidence, the mating/parenting aspect of Del Giudice's (2018b) Depiction is too weakly supported to warrant full separate Evaluation currently. With this aspect called into question, his account more directly aligns with Baron‐Cohen's (2008, 2020) (Table 3), so for simplicity they can be considered equivalent. Evaluation can proceed comparing the systemising social niche specialisation with the high intelligence by‐product hypotheses.

In terms of relying on plausible evolutionary theory, Del Giudice (2018b) and Crespi (2016) specify evolutionary dynamics via verbal models, while Baron‐Cohen (2008, 2020) is more implicit, not mentioning social niche specialisation or social selection explicitly, but providing the same basic argument (that high systemisers were valuable to the social group for their specific abilities). One possible criticism of the validity of the theory behind the social niche specialisation hypotheses could be raised against the viability of balancing selection at the genetic level maintaining adaptive individual differences in complex polygenic traits (Zietsch, 2024). This takes the strong position that adaptive personality differences cannot be polygenic, but there is evidence that great tit (Parus major) personality is under precise balancing selection (Mouchet et al., 2021) and has a polygenic basis (Santure et al., 2015). Furthermore, evidence for frequency‐dependent selection across species is ubiquitous (Gómez‐Llano et al., 2024). These are the sorts of specific debates that can develop and affect Evaluation, and which different parties may disagree on. For the sake of exemplifying the DCIDE framework here, we shall assume both the by‐product and social niche specialisation hypothesis rely upon viable theory.

In terms of immediately accounting for one another's direct evidence, the overlap between hypotheses relieves potential conflict between them – their primary discrepancy is in the evolutionary model. These competing Depictions can be Evaluated for their sufficiency explaining the circumstantial evidence.

(b). Evaluate: age of effect

The relevance of age of effect requires referencing average challenges facing different ages of humans over evolutionary history. Functions of adaptations should appear at appropriate times. For example, traits appearing early in reproductive age could function in courtship and initial acquisition of mates; later ages carry trade‐offs between supporting existing offspring and investing in new offspring. If there is selection for some function at a particular age in life (e.g. high sociosexuality in early adulthood) by‐products might appear throughout the life course, despite not being fitness enhancing. By‐product accounts could refer to costs of plasticity (e.g. if it is more costly than beneficial to be able facultatively to turn off high sociosexuality when fitness‐enhancing benefits disappear), physical constraints (e.g. if high sociosexuality could not evolve restricted to early reproductive age) or trade‐offs optimised for reproductive success early in life (e.g. if high sociosexuality causes health problems in old age). Depictions referencing such by‐products would benefit if research reveals constrained physiological mechanisms.

(i). Evaluating age of effect: autism

Autism is often considered to have lifelong effects, but its specific manifestation often varies substantially by age. In infancy and childhood, social and behavioural difficulties may be quite severe, but during adolescence and adulthood many autistics, especially those of higher cognitive ability, reduce in symptomology (Waizbard‐Bartov & Miller, 2023). Individuals scoring high in autistic traits often show reduced social difficulties over time (Riglin et al., 2021), and differences in oxytocin levels are only found in children, not in adults (John & Jaeggi, 2021). This trajectory may relate to apparent advantages in autistic and high‐systemising individuals of accurately predicting social psychological phenomena (Gollwitzer et al., 2019), implying they recruit deliberation and reasoning abilities in social situations instead of automatically empathising. This change in characteristics over time is important for adaptive accounts, because behaviours occurring during the period of pre‐reproductive age only need to prepare an individual for reproduction, whilst behaviours at reproductive age more directly affect fitness.

Depictions by Del Giudice (2018b) and Baron‐Cohen (2008, 2020) relate to systemising and specialist skills serving useful social functions – the benefits primarily arising in adulthood, when their abilities are appreciated, leading to reproductive success. Baron‐Cohen (2020, Ch. 2) proposes that hypersystemising in childhood makes exceptional ability more likely. This is coherent with slow development of systemising abilities and costs that alleviate with age. Crespi (2016) might account for this developmental trajectory as a by‐product of intelligence imbalance slowly compensating over time. Still, harmful by‐products are more likely to appear late in life, where they are less visible to selection, so some constraint explanation of why costly early‐life autistic traits persist is required.

(c). Evaluate: duration

Duration is most functionally informative if aligned with immediate environmental stimuli or life circumstances. Durations of adaptations should relate to their adaptive function: adrenaline's effects are short‐lasting, love is longer‐lasting, capacity for language longer, because of their respective functions. By‐products could theoretically occur for any duration, but the specific duration should still be justified.

Described and Integrated characteristics often have different durations – for example, schizophrenia may consist of delusional thoughts and hallucinations for long periods, with brief episodes of intense psychosis requiring hospitalisation or increased medication. Simplistically, visibility to selection implies longer effects are more likely to be adaptations, but is contingent on phenotype specifics and strength of selection. Ten minutes of severe anxiety may have a stronger effect on fitness than ten years of moderate anxiety if the ten minutes involve life‐saving behaviour.

(i). Evaluating duration: autism

Autism is characterised by general differences in cognition which seem long‐lasting, especially in restrictive and repetitive behavioural domains, with more regular improvement in social functioning over time (Waizbard‐Bartov & Miller, 2023). Characteristic behaviours also involve bursts of extreme behaviour (e.g. meltdowns, sensory overload) and periods of obsession with particular systems, subjects or activities which may last from days to years.

Systemising associates with differences in occupations, interests and hobbies across the lifespan (Svedholm‐Häkkinen & Lindeman, 2016) which fits Baron‐Cohen (2008, 2020) and Del Giudice's (2018b) Depictions of high‐systemising phenotypes filling a particular social niche. If periods of obsession lead to long‐lasting expertise, this would also support the hypothesis. Various shorter‐lasting behaviours such as repetitive play, craving movement (e.g. spinning, jumping) and stereotypies (e.g. hand flapping) reduce with age (Mayes & Calhoun, 2011), and along with meltdowns and sensory overloads could be acceptable by‐products. To Crespi (2016), the duration of autism and Integrated traits is due to the developmental imbalance of intelligence which lasts for life. Bursts or phases of behaviour are by‐products of this imbalance. The long duration and thus high visibility are surprising for costly by‐products, which should be strongly selected against – again, more clarity as to the mechanisms that have constrained selection from maintaining high intelligence without regular cases of autism is needed.

(d). Evaluate: sex differences

Sex differences may relate to sex‐specific functions of the characteristics in question or come downstream of other sex differences, environmental or biological. Functional accounts should explain differential presentation (either in prevalence or manifestation) as solving an adaptive challenge experienced uniquely or disproportionately by one sex. Differential challenges between sexes are a common topic of evolutionary psychology, involving intersexual or intrasexual competition, infidelity, courtship, mate guarding, attractiveness and more (Trivers, 1972; Stewart‐Williams & Thomas, 2013).

By‐product approaches to explaining sex differences might reference adaptive sex differences with costs (potentially in either sex) or differential vulnerability due to other sex effects. As noted in Integration, biological constraints mean by‐products in the non‐ or less‐benefitting sex are expected. Biomedical variables may allow validation of such hypotheses (e.g. finding testosterone is critical to trait manifestation implies positive selection in males with by‐product effects in females).

(i). Evaluating sex differences: autism

As noted in Integration, autism sex ratios are male‐biased at approximately 3:1, with female autistics more successfully camouflaging and requiring a higher polygenic score to meet diagnostic criteria (Warrier et al., 2022). 44% of general population males also show either a systemising or high systemising bias in the UK Brain Type Study (Greenberg et al., 2018) in comparison to 27% of females. Baron‐Cohen's (2008) account particularly concentrates on this difference, postulating that higher empathising in females relates to evolutionary pressures related to childcare, whilst systemising may help in male‐biased activities of hunting, tracking, toolmaking and tool use; while this may sound like stereotyped gender roles, and deviations are certainly possible (or even desirable), the ethnographic record generally shows that among hunter–gatherers women are primarily responsible for childcare (and typically nurse infants for several years), and that this constrains (although not entirely precludes) women's ability to engage in high‐risk activities like hunting or warfare (Kelly, 2013; Bird, 1999). High female empathising may also mean that being shifted towards a high systemising phenotype makes diagnosable autism more likely in males than females, because female empathising mitigates social disability (Baron‐Cohen, 2020). Camouflaged female cases are thus explained as a combined reduced benefit to females of systemising and increased benefit of empathising. The social niche specialisation hypothesis explains the general spectrum of sex differences as due to systemising traits and niches as particularly fitness‐enhancing for males. This explanation of the Integrated spectrum is more complete than Crespi's (2016), which briefly mentions sex differences in autism rates as possibly related to reduced verbal and higher rotational intelligence in males. To make a stronger case, the by‐product account would have to spend more time explaining why the costly by‐products of high intelligence are much more likely to cause diagnosable autism in males – and would have to provide a more convincing explanation of this fact than the social niche specialisation hypothesis, which considers this at length.

(e). Evaluate: environmental plasticity

Environmental effects can precipitate, prevent or alter presentation of traits – they affect phenotype manifestation within a reaction norm. Functional explanations of environmental effects should explain why this reactivity was itself fitness enhancing. This can be obvious in short‐lasting reactive states such as emotions. Suites of feelings, physiological arousal and behaviour arise and dissipate in particular circumstances of adaptive challenge – for example, anger arising to rectify social undervaluation (Sell et al., 2017) or fear for protection (Stankowich & Blumstein, 2005). For long‐lasting traits, predictive adaptive responses (Gluckman, Hanson & Spencer, 2005) utilise environmental cues to provide information about future environments, allowing functional developmental trajectories via adaptive developmental plasticity (West‐Eberhard, 2003). This may affect whether a trait appears at all (e.g. early life experiences causing phobias) or presentation of a trait (e.g. language capacity developing different languages and accents). Assessing the range of presentation of the trait in question allows observation of its fixed and flexible aspects, which inform functional hypotheses regarding the fixed and flexible aspects of the adaptive challenge. Cross‐cultural research can be particularly informative here.

By‐product explanations need to account for specific effects as non‐functional. These are expected – external influences have stochastic effects, and functional plastic responses need only be beneficial on average. Arousal of short‐lived reactions can be inappropriate, and in longer‐lasting plastic responses, developmental trajectories may end up non‐functional or harmful in the future environment [so called “developmental mismatch” (Gluckman, Hanson & Low, 2019)].

(i). Evaluating environmental plasticity: autism

Environmental factors precipitating cases of autism eligible for adaptive explanations are limited – advancing parental age is the clearest (Wu et al., 2017). Whether advancing parental age also increases systemising and BAP traits is unclear. If so, investigating functional explanations for why systemising traits are more beneficial for children born to older parents is prudent for the social niche Depiction (e.g. perhaps older siblings support less‐social younger siblings, mitigating harms). If not, undiscovered dysfunction is more likely, and both Depictions would need to explain this effect as a vulnerability trade‐off. This awaits further research for clarity.

Autism, BAP and systemising presentation do depend on environment, however. Although autistic and systemising cognitive differences are fairly stable over the life course, the behavioural manifestation of systemising is environmentally dependent, concentrating on local lawful, repeatable, predictable systems (Baron‐Cohen, 2008). Systemisers are “pattern seekers”, but precise systemising interests may range between mechanical systems, weather, music, animals, and more. This fits social niche specialisation hypotheses, which necessitate flexibly adapting to local ecologies. Although research on autistic presentation in evolutionarily relevant societies is lacking (Spikins, Wright & Hodgson, 2016), there is an ethnographic example of an unusually antisocial reindeer herder, highly valued for memorising parentage, medical history and character of 2600 individual reindeer. Similar modern success stories are often in technical careers. Where such tendencies lead to socially valued expertise, this could be functional. If the BAP and systemising were costly by‐products away from an optimum cognitive type, reliable inclinations to attach to potentially valuable local systems and gain above‐average ability are somewhat surprising. Still, by‐product explanations could plausibly explain this as a mix of extreme cognitive strengths associated with the alleles and universal tendencies for individuals to gravitate towards their areas of ability, rather than actually fitness enhancing by being valued above neurotypical cognition. This area of evidence less clearly favours either hypothesis, but neither are seriously challenged.

(f). Evaluate: prevalence of genetic propensity

Genotypes interact with environments in developing phenotypes, and the range of resulting phenotype expressions – the reaction norm – is the genetic effect that selection acts upon (Martin et al., 2023) and which must relate to functional explanations. Depictions should explain why heritable reaction norms do not exist at lower or higher frequencies.

Species‐wide traits (e.g. eyes) or states (e.g. hunger) are eligible for functional explanations relating their species‐wide positive effect to fitness. Species‐wide by‐products may be trade‐offs of such adaptations, unavoidable biological constraints, or potentially non‐adaptive yet drifted to fixation. Heritable traits below species‐wide prevalence are more complicated. They may have fixated within sub‐groups (e.g. of geographic location or sex), thus requiring population‐specific functional explanations. They may result from “reactive heritability” (Tooby & Cosmides, 1990), species‐wide strategies arising due to other heritable phenotypes (e.g. extraversion arising in attractive people; Lukaszewski & Roney, 2011). Alternatively, they may represent adaptive individual differences in functional strategy or by‐products of such strategies (Fig. 1). Functional genetic individual differences are mostly theoretically predicted and empirically observed in social animals with predictable social environments, where costs or constraints prevent plasticity, and social niches exist for alternative strategies (Hunt & Jaeggi, 2022). These likely require referencing negative frequency‐dependent selection, social niche specialisation and social selection, with alleles under balancing selection.

When Evaluating hypotheses' sufficiency at explaining specific prevalence of genetic propensity, ancestral social group size is the relevant framing. Hunter–gatherer social structures are composed primarily of residential units (“bands”) of individuals (mean size of 28) interacting frequently with other bands in metagroups (“tribes”) with a geometric mean of 165 individuals (95% confidence limit: 152–181; Hamilton et al., 2007). Gender ratios are usually equal, although can fluctuate stochastically (Kramer, Schacht & Bell, 2017), with population composition about half adults and half children/adolescents (Kelly, 2013). Assuming these broadly represent the average group sizes of relevance to human evolution, we can infer prevalence per social group throughout human evolutionary history, with standardised functional implications – traits appearing, on average, in five individuals in every band are eligible for different explanations than traits appearing once per band. In general, higher prevalence indicates higher visibility to selection and thus higher likelihood to be an adaptation, but rare strategies may be adaptive at a minimum level of once per aggregated bands (approximately 1%) or once per band (approximately 4%) where specific social niches exist (Hunt & Jaeggi, 2022; Martin et al., 2023). Mealey (1995) argued, for instance, that sociopathy is an adaptive cheating strategy, optimal in around 1% of the population to avoid cheater‐detection.

(i). Evaluating prevalence of genetic propensity: autism

Autistic traits are distributed continuously throughout the population. Assuming that high heritability implies similar cross‐cultural rates of relevant heritable cognitive differences (although precise manifestation may differ), trait distributions should be Evaluated referencing hunter–gatherer social group size. The most extreme 0.6% are ineligible for functional explanation (Hunt & Jaeggi, 2022), but if identifying, for example, 3.14% of individuals (4.64% of boys) as autistic (Li et al., 2022), approximately one adult in every two or three bands would meet diagnostic criteria. Stricter criteria including 1% of individuals implicate one individual per tribe. At 5–9% of the population (Morrison et al., 2018; Sasson et al., 2013), one BAP adult per band is expected.

Del Giudice (2018b) and Baron Cohen (2008, 2020) are not precise about the adaptive range of autistic traits, but such frequencies are coherently explained if one BAP adult finds a social niche within most hunter–gatherer bands, with the possibility that manifestations meeting autism diagnostic criteria find niches in the wider social group. It is also possible that band‐level niches are advantageous enough to justify costly extremes manifesting more rarely in the wider social group – or indeed, if assortative mating is important, perhaps such phenotypes were essentially non‐existent in the past, due to restricted mate choice (Kramer et al., 2017). Success at different levels of group size could vary between environments and generations – the social niche hypothesis merely implies balancing benefits and costs which prevent persistent increase or decrease in frequency, aligning with various versions of functional explanations of the observed prevalence. Crespi's (2016) model emphasises that autism‐causing alleles increase intelligence and have been positively selected across the general population, but autistic trait distribution represents prevalence of costly by‐products, persisting due to inefficient negative selection. This makes autism and BAP trait presence in every hunter–gatherer group surprising – the proposed dynamic is that the non‐autistic intelligence increase is significant enough that it justifies a constant costly by‐product – the persistent churn of selection for intelligence and against autistic individuals just happens to remain stable enough to maintain a rate of almost exactly one BAP adult per band.

(g). Concluding evaluation

Evaluating the high intelligence by‐product hypothesis against the social niche specialisation hypothesis with respect to the visibility variables, a general pattern emerges: for the by‐product hypothesis, the circumstantial evidence is theoretically surprising and needs to be excused as a failure of natural selection, but for the social niche hypothesis, the evidence is generally expected. Note that if any single area of evidence was substantially different, the social niche hypothesis could be excluded: if prevalence too low, or onset too late, or duration too short, or phenotype not appropriately environmentally reactive. Instead, the evidence fits the Depiction, so we Evaluate the systemising social niche specialisation hypothesis favourably.

Biomedical variables have been excluded from this Evaluation, which could unfairly disadvantage the by‐product model. In particular, the by‐product model predicts positive selection for autism alleles associated with intelligence, and neurological similarities between high intelligence and autism implying exaggerated features. Some research has found positive selection on autism risk variants associated with intelligence (Polimanti & Gelernter, 2017), but only 14 SNPs out of 446 showed such signals (Prakash & Banerjee, 2021). With regard to exaggerated features, the strongest findings relate to brain growth, but one large project found only 15% of autistic boys and 6% of girls show precocious brain growth (Amaral et al., 2017), and a very large (N = 7005) study found no significant relationship between autistic traits and subcortical region size (Sharp et al., 2023). Future findings might identify the exaggerating brain differences Crespi (2016) proposes. This would not necessarily exclude the social niche specialisation hypothesis (they could be the mechanism of adaptation) but would provide a mechanistic explanation of the high intelligence by‐product.

The major existing alternative to these two models is mutation–selection balance models. Although the DCIDE framework could identify harmful mutations during Categorisation, it has been proposed (Mitchell, 2015) that the genetic basis is too complex, and single gene effects too small for natural selection to remove the variants effectively. If such variants are suggested to be common and heritable, and could never be reliably Categorised as non‐adaptive, this hypothesis can still be Depicted and Evaluated. To do so briefly – presuming the theory is sound, the compatibility of the mutation–selection balance hypothesis with direct evidence of autistic cognitive strengths cited by the adaptive hypotheses is questionable. Explaining why harmful mutations should associate with traits such as higher educational attainment and improvements in various specific cognitive tasks is necessary, and existing accounts do not attempt to explain these findings. Presuming a passable response could be formulated, then accounting for the specific circumstantial evidence faces similar challenges as Crespi's (2016) by‐product model: the high visibility of autistic traits is surprising if they are simply costly and supposedly avoiding negative selection. It would be particularly surprising if the mutation–selection balance rate maintains a prevalence of autistic traits which is high enough for a BAP individual to be present in every band, and one or two autistic individuals in every wider social group. Why would the prevalence not be higher or lower? The social niche specialisation hypothesis is the only Depiction which predicts and explains this explicitly. Both the by‐product and mutation–selection balance hypotheses must excuse this fact (and all the circumstantial evidence) as essentially happenstance – harms happen, selection is inefficient – but the proposition that this randomly causes exactly the evidence we expect to see if the social niche specialisation hypothesis were true makes this unconvincing. Any other neutral or negative hypotheses are put in the same difficult position.

Given this DCIDE review, we suggest that the best existing evolutionary explanation for autistic traits, including diagnosable autism but excluding non‐adaptive cases, is that advantages in systemising and related cognitive enhancements led to individuals consistently filling valued roles throughout societies over human evolutionary history, with social disadvantages that came as trade‐offs. The primary reason the human population differs cognitively upon the autism spectrum is because the benefits of autistic thinking are balanced by costs: only so many spaces exist for these types of minds before strengths are outweighed by weaknesses, so the individual difference persists.

Given the sufficiency of this explanation in accounting for the available evidence, it would be surprising, but certainly not impossible, if a competing Depiction usurped it. Most plausible would be some alternative account of the precise role or niche that autistic or BAP individuals fill, perhaps concentrating on some non‐systemising trait such as attention to detail or sensory sensitivities. The circumstantial evidence (especially prevalence) would be hard to explain with completely different evolutionary models to social niche specialisation, but the precise phenotypic traits of relevance, and the dynamics which occur regarding the positive and negative selection effects may differ. For example, although existing accounts concentrate on male preponderance and male niches, it's possible that autistic cognition is more positively selected in females (e.g. for food processing or tool‐use skills) with males bearing costly by‐products due to background higher systemising cognition.

Despite the reasonable sufficiency of the systemising social niche specialisation hypothesis in explaining the direct and circumstantial evidence, some complexities should be noted. Firstly, not all ASD or BAP individuals show the systemising behavioural phenotype core to the Depiction. This should warrant Evaluation as “mixed results” – the original Description may be too broad, capturing multiple adaptations or by‐products of adaptations, and benefit from tighter Description. Altering Description specifiers to differentiate high and low systemisers may result in a different adaptive explanation for low‐systemising autistics. Evidence of distinct heritability between these groups would justify this separation. There is emerging evidence that genetics could be used to subtype cases of autism caused by common alleles, and that these cluster into groups which show different age of onsets and ability profiles (Zhang et al., 2024). If an efficient DCIDE analysis were to be conducted on these re‐Described groups, we should seek both circumstantial and direct evidence of the phenotypes associated with such clusters.

Furthermore, Evaluation does not narrow down the adaptive range of systemising–BAP–autism Integrated traits. The systemising social niche specialisation hypothesis fits the evidence, but exact functional explanations are not constrained to “the BAP is adaptive, autism is not” or vice versa. This is acceptable – indeed, precise function may vary between generations and environments – only average inclusive fitness matters. Relatedly, although the high intelligence by‐product explanation alone struggles to sufficiently explain the circumstantial evidence, positive selection on intelligence across the population may simultaneously be a factor increasing autism‐predisposing alleles, accompanying balancing selection for systemising social niche specialisation. Functional accounts are often not mutually exclusive. Non‐adaptive processes such as mutations and toxins also likely contribute to some cases, in a diathesis‐stress type way (Rende & Plomin, 1992). Just as Baron‐Cohen (2008, 2020) and Del Giudice (2018b) suggest assortative mating could explain certain dysfunctional outcomes, costs and vulnerabilities are inevitable to adaptations. In questioning the primary functional story explaining the evolutionary paradox of autism's persistence, however, the social niche specialisation hypothesis best fits the available evidence.

VI. FURTHER APPLICATIONS

The DCIDE framework is a formal framework for formulating and assessing adaptive hypotheses, particularly suitable for psychiatric traits, for which rich biomedical and epidemiological data often exist. Attempting to explain psychiatric conditions must start by recognising the stark heterogeneity between non‐adaptive and potentially adaptive cases within broadly defined disorder categories. Once non‐adaptive cases are considered explained, any account that does not recognise the relationship between diagnosable traits and their more evolutionarily visible correlates will fail to capture the critical phenomena. Then, given various hypotheses attempting to explain those phenomena, without using a systematic method of analysing the available evidence and comparing hypotheses with standardised inferences, progress towards the true explanation is difficult. Although autism serves as a useful exemplification, the DCIDE framework could be widely applied. Indeed, beyond its use in whittling down the many existing evolutionary hypotheses for psychopathological traits [see Del Giudice (2018a) for a comprehensive review], its role in strengthening existing work in evolutionary psychology may also be valuable.

Evolutionary psychology could benefit from assessing circumstantial evidence and visibility variables, expanding the scope of useful experiments and data collection beyond trying to precisely evoke psychological function, and encouraging collecting simpler behavioural or demographic data to support hypotheses. This is yet another reason to encourage research beyond traditional convenience samples of students (Henrich et al., 2010), especially cross‐culturally and across age groups. Although such evidence is sometimes collected and referenced, this has occurred without a framework of standardised analytical principles. To briefly DCIDE review a classic target of evolutionary psychology, romantic jealousy may be Described as thoughts or feelings of insecurity, fear, and concern over a relationship. Presumably few cases will Categorise as non‐adaptive. Jealousy is certainly visible to selection, so Integration of associated traits is not required to encompass a likely target phenotype of selection. A leading Depiction (Buss & Haselton, 2005) is that jealousy is activated by threats to a valuable relationship, functioning to protect it from partial or total loss, with many specific features of jealous emotions and behaviours noted as direct evidence of the functional trait. In Evaluation, this hypothesis amply explains the age of effect (from sexual maturity for romantic jealousy, earlier for friendship jealousy, and youngest for parental investment jealousy), duration (relating to perceived threat), sex differences (more concerned with sexual infidelity in men and emotional infidelity in women), environmental effects (relating to threat likelihood) and disposition prevalence [the capacity for jealousy seems essentially universal, showing fairly low (29%) heritability (Kupfer et al., 2022), which is plausibly explained as reactive heritability]. The sufficiency of the evolutionary account in explaining this circumstantial evidence makes it clear that this is not mere “just‐so” storytelling.

Competing hypotheses must explain the same evidence – the DCIDE framework provides a structure of systematic review to compare evolutionary hypotheses, applicable to both psychology and psychiatry. In psychiatry, when forced to try and explain not only the direct evidence of the trait's suggested function, but also the circumstantial evidence implying visibility to evolutionary selection, as well as the lack of biomedical evidence of Categorisable non‐adaptive causes, many existing approaches from non‐evolutionary fields will be shown for blindness to critical available evidence. Neuroscientific investigations of autism generally assume discovered circuits are dysfunctional and do not ask why the brain differences are so common, heritable and associated with certain cognitive strengths. GWAS studies identify alleles correlated with psychiatric traits but generally do not ask about why the alleles persist, what their effects are in the general population, and what it means that the resulting traits are early onset and highly prevalent. When the broad available evidence is laid out, simple pathological hypotheses denying any role for adaptive explanations of many common mental disorders may often begin to sound like the much more speculative option – perhaps such speculations should be labelled “just‐no” storytelling.

VII. CONCLUSIONS

• (1)Evolutionary explanations of psychiatric conditions require untangling heterogeneity and incorporating subclinical forms with shared genetic relationships.
• (2)Autism spectrum disorder includes 5–20% of cases of clear dysfunction without need for adaptive explanation – explaining the remaining cases needs to incorporate the broad autism phenotype and camouflaged cases in females.
• (3)Standardised analytical principles for assessing evolutionary explanations can be derived from recognising visibility to selection and its connotations.
• (4)Of existing adaptive explanations for autism, systemising social niche specialisation and high intelligence by‐product hypotheses are compared.
• (5)Systemising social niche specialisation better explains the available evidence around autism, while the by‐product hypothesis struggles to incorporate circumstantial evidence.
• (6)Adopting this method of systematic review could reduce problems of just‐so storytelling and cherry picking in evolutionary psychology and psychiatry.