Abstract
Understanding the makeup of mental disorders has great value for both research and practice in psychopathology. The richer and more detailed our compositional explanations of mental disorder—that is, comprehensive accounts of client signs and symptoms—the more information we have to inform etiological explanations, classification schemes, clinical assessment, and treatment. However, at present, no explicit compositional explanations of psychopathology have been developed and the existing descriptive accounts that could conceivably fill this role—DSM/ICD syndromes, transdiagnostic and dimensional approaches, symptom network models, historical accounts, case narratives, and the Research Domain Criteria (RDoC)—fall short in critical ways. In this article, we discuss what compositional explanations are, their role in scientific inquiry, and their importance for psychopathology research and practice. We then explain why current descriptive accounts of mental disorder fall short of providing such an explanation and demonstrate how effective compositional explanations could be constructed.
Mental disorder represents a serious and expanding global health problem; demonstrating high prevalence internationally (Steel et al., 2014) and accounting for a considerable proportion of global disease burden (Whiteford et al., 2015). Being able to assess and treat these problems effectively is therefore critically important. Researchers and clinicians also have an ethical obligation to seek accurate understandings of mental disorders, so as not to provide mistaken accounts of people’s genuine concerns. Both goals rely heavily on the explanations we have for mental disorders: our theoretical accounts of how they originated and why they persist.
Explanations are invaluable: they tell us what to look for in assessment, what to target in treatment, and how to do so successfully. At present, however, psychopathological explanations are often unsound: constructed in idiosyncratic ways and containing numerous conceptual flaws (see Hawkins-Elder & Ward, 2020a, 2020b). The comprehensive description of explanatory targets—known as compositional explanation—is significantly underemphasized, and often neglected, in psychopathology (Gillett, in press). Describing the nature of psychopathological symptoms and signs holds significant value for explanation, classification, research, and practice (Wilshire et al., in press). Despite this, no explicit compositional explanations of psychopathological phenomena have been created and existing descriptive accounts present relatively poor alternatives; lacking sufficient depth, theoretical structure, or evidence base.
Current diagnostic syndromes provide only a short list of vaguely defined “core” features (e.g., DSM-5, American Psychiatric Association, 2013; ICD-11, World Health Organization, 2019), thereby limiting the richness of information theorists possess about the disorder. Alternative classificatory perspectives, such as transdiagnostic, dimensional, and symptom network models, although in some cases providing greater taxonomic validity, still fail to richly describe psychopathological problems at all the relevant levels. In contrast, more “clinical” descriptions, such as historical accounts (i.e., the first identification or early scientific conceptualizations of a disorder) and clinical case narratives (e.g., case studies) tend to describe disorders in more depth, but lack detail and theoretical organization and are often empirically outdated. The Research Domain Criteria (RDoC; Cuthbert & Insel, 2013) is perhaps the current option that most closely approximates a compositional explanation. However, critically, this framework is not geared towards the theoretical conceptualization of psychopathological phenomena and therefore, although excellent at guiding empirical investigation into psychopathology, is unable to provide coherent compositional explanations—at least, not on its own.
Our intentions with this article are therefore threefold. First, to highlight the importance of compositional explanations of psychopathology by outlining their role in scientific inquiry and clinical practice. Second, to demonstrate that none of our current descriptive accounts (i.e., DSM/ICD syndromes, transdiagnostic and dimensional classifications, symptom network models, historical accounts, case narratives, and the RDoC) are capable of successfully acting as compositional explanations. Finally, based on this discussion, to suggest how we could construct effective compositional explanations of psychopathology.
Compositional explanations: What they are and what they do
In this section, we discuss the nature of compositional explanations and their role in scientific inquiry and clinical practice. However, we first must clarify what we mean by “explanation.”
What is an “explanation”?
There is some disagreement in science over what exactly constitutes an explanation, as well as how exactly explanations relate to models and theories. Trout (2016) provides a simple and helpful definition of explanation: “an explanation is the description of underlying causal factors that bring about an effect” (p. 18). More specifically, an explanation is an account that provides an understanding of a phenomenon’s causes, composition, context, or consequences (Faye, 2014; Ruphy, 2016). Models and theories, in contrast, are conceptual representations of phenomena in the world: tools that can be put to the task of explanation (Mantzavinos, 2016; Savulescu et al., 2020).
For our purposes, a theory is an integrated system of concepts and ideas that can be used to explain why some phenomena occur and persist. Theories are more general than models, usually detailing more abstract phenomena (e.g., “human behaviour” in general vs. specific types or instances) and able to explain different subsets of phenomena (Bailer-Jones, 2003). Models, in contrast, represent more concrete empirical phenomena (i.e., identified factors, systems, or processes such as “clinical depression” or “binge eating”); typically, in idealized and simplified ways (Bailer-Jones, 2003; Haig, 2014). Theories may inform the development of models, and models may represent localized applications of theories (Bailer-Jones, 2003). For example, the coercion cycle model represents the localized application of operant conditioning theory to the phenomenon of child conduct problems (Dishion & Patterson, 2015).
We define an explanation herein as the entire bank of explanatory knowledge about a phenomenon—that is, the most complete explanatory account (Craver & Kaplan, 2020). A model or theory may serve as an explanation if it represents the entire bank of knowledge about a phenomenon: when it adequately represents all aspects of the phenomenon without being overly complicated or sacrificing critical detail. However, this is rarely the case. More often, models and theories represent partial explanations (Bailer-Jones, 2003). When the phenomenon of interest is more complex, multiple theories and models will typically be needed to fully explain it (Kendler et al., 2020).
What are compositional explanations?
Philosophers of science make a distinction between causal and compositional explanations (e.g., Craver, 2007; Kaiser & Krickel, 2016). A causal explanation depicts the factors that result in a subsequent effect—for example, heating water (a cause) until it boils (an effect). A psychopathological example is the proposition that traumatic experience can cause individuals to experience intrusive memories, elevated arousal, and avoidance behaviour (i.e., a “posttraumatic stress” syndrome). Causal explanations in psychopathology may also include accounts of relations that maintain the disorder, such as mutual reinforcement between symptoms (e.g., insomnia and low mood; Konjarski et al., 2018) or behavioural reinforcement “cycles” (e.g., the coercion cycle; Dishion & Patterson, 2015).
In contrast, a compositional explanation describes underlying structures and interactions that make up a phenomenon; viewed as part of it rather than “causing” it (Craver, 2007; Gillett, in press; Kendler et al., 2020). For example, the symptom “low mood” is likely to be composed of processes at the phenomenological, subpersonal, neurobiological, and physiological levels (Ward & Clack, 2019). It is important to note that compositional explanations may, in some cases, contain causal relations, depending on the phenomenon being explained. For example, a compositional explanation of a syndrome, like clinical depression, would necessarily include description of causal relations between symptoms (e.g., low mood and insomnia), as these relationships are part of the constitution of that syndrome (although arguably do not cause it). However, these same relations could form part of an etiological explanation depending on the focus of inquiry. For instance, if, instead of describing the composition of depression, we were trying to explain the development and persistence of low mood in depressed individuals, we might ascribe causal or etiological significance to insomnia somewhere within that explanation, but we would not say that insomnia in any way constituted low mood. Hence, the role of factors and processes within an explanation varies depending on the question being asked.
What is the role of compositional explanation in theory, research, and practice?
Compositional explanations play a critical role in all aspects of clinical inquiry. First, they hold significant value for etiological explanation. The more detailed our compositional understanding of a phenomenon, the more information we have to provide clues about its etiology (Hawkins-Elder & Ward, 2020b). For example, if we were trying to explain the existence of a cake—knowing only that it was a cake—we could reason that its etiology probably involved components and processes common to most cakes (e.g., flour, sugar, being baked). However, with further detail about how it is composed—for example, chocolate sponge, cream-filled—we have additional clues to help refine our etiological reasoning; strengthening ideas about the involvement of some factors and processes (e.g., flour, being baked) and suggesting new ones (e.g., whipping cream, cocoa). When our compositional understandings are “thin” (less detailed) it can promote errors in causal reasoning: relevant causal factors and processes may be neglected or deliberately omitted, and flawed or irrelevant ones may be included. For example, we could develop multiple etiological theories about our cake, hypothesizing various baking processes, when all the while we were dealing with an ice cream cake—a type of cake indeed, but one involving none of our postulated causal processes.
Compositional explanations also hold value for research and practice. Their primary value for research is via classification. Although compositional explanations are attempts to describe disorder phenomena, they do not claim to know the best method of classifying the psychopathological phenomena with which they are concerned. However, because of their informational value, compositional explanations are highly useful for those aiming to develop taxonomies of mental disorder (Wilshire et al., in press): their rich descriptions may help to signal connections between syndromes or symptoms, and thereby suggest novel and improved ways of organizing them. Compositional explanations lay out the psychopathological landscape in comprehensive detail, allowing it to be thoroughly surveyed by those who wish to classify it.
In clinical practice, compositional explanations hold value for both assessment and treatment. For one, their value for etiological explanations and classification systems has flow-on effects for assessment and treatment. Improvement of classification systems will likely be beneficial for diagnosis and the prescription of appropriate clinical interventions. Likewise, better etiological explanations will likely improve clinical formulations and intervention strategies based on them. Compositional explanations also hold independent value for clinical practice. Possessing more information about a psychopathological problem provides us with more features to look for in assessment. It may also help identify potential therapeutic issues. For example, knowing that a particular disorder often involves cognitive inflexibility or poor attentional control might contraindicate interventions requiring high levels of cognitive effort. Likewise, knowing that a particular symptom involves difficulty sensing physical sensations may influence how an intervention is administered (e.g., devoting extra time during a mindfulness intervention to helping the client identify physical sensations).
Current “approximate” compositional explanations
There are several types of account within the psychopathology space that, due to their descriptive nature, could potentially serve as compositional explanations: namely, (a) DSM-5/ICD-11, (b) transdiagnostic and dimensional approaches, (c) symptom network models, (d) historical accounts, (e) clinical case narratives, and (f) the Research Domain Criteria (RDoC) framework. We will now address each in turn and explain why, on our view, none effectively serve as compositional explanations.
Diagnostic syndromes: DSM-5 and ICD-11
The DSM-5 (American Psychiatric Association, 2013) and ICD-11 (World Health Organization, 2019) are perhaps the most prominent attempts to conceptualize and describe mental disorders. Both group disorders into discrete syndromes (collections of symptoms and signs) comprising a set of diagnostic criteria (e.g., borderline personality disorder, anorexia nervosa). They are frequently used as compositional explanations in research and theory: empirical inquiry is often oriented around DSM-5 categories and etiological models commonly use them as the foundation for explanation. However, these syndromes fall short of providing a compositional explanation in two important ways.
First, DSM/ICD syndromes lack explanatory scope. Each is characterized by a relatively small number of descriptively “thin” criteria spanning but a few levels of analysis (e.g., behavioural, cognitive, emotional). For example, the criteria for anorexia nervosa (listed in Table 1) describe only a few features, despite research identifying many others common to these individuals, such as alexithymia (Nowakowski et al., 2013; Westwood et al., 2017), interoceptive deficits (Stinson, 2019), cognitive deficits (Hedges et al., 2019), and autistic traits (Westwood et al., 2016). Furthermore, both the DSM and ICD outline only those features that are most clinically salient—that is, those most readily observable in practice or perceived as “most central” to the disorder’s pathology. Although appropriate and often useful in practice, this omits other relevant features that may be harder to identify (e.g., emotional comprehension, executive functioning, interoceptive ability) or have less apparent relevance (e.g., attentional bias, central coherence), but are nonetheless characteristic of the disorder.
DSM-5 diagnostic criteria for anorexia nervosa and bulimia nervosa.
Second, DSM/ICD syndromes lack explanatory depth, as the features/symptoms listed by them are typically thinly described. For instance, “disturbance in the way in which one’s body weight or shape is experienced” (American Psychiatric Association, 2013, p. 339) is a necessary criterion for anorexia nervosa (see Table 1), however there is no detail about the exact nature of this “disturbance.” For example, is it a distortion in sensory perception or cognitive evaluation (Mölbert et al., 2017)? Does it encompass the body in general or does it tend to be focused on specific areas (Cash & Deagle, 1997)? Body image is recognized to be a “multi-faceted construct consisting of a variety of measured dimensions” (Thompson, 2004, p. 8), including perceptual, conceptual, and emotional (Stinson, 2019). Therefore, the precise nature of any proposed “disturbance” would need to be more specifically detailed.
Transdiagnostic approaches and dimensional approaches
Transdiagnostic approaches advocate dispensing with existing diagnostic syndromes in favour of broader classifications based on shared characteristics. In some cases, this involves collapsing said syndromes into a broader disorder category (e.g., anxiety disorders), in others basing classification on some common factor (e.g., the internalizing/externalizing model; Krueger & Eaton, 2015). Dimensional approaches are based around spectra or “scales” rather than discrete categories, such as the Five Factor Model of personality disorders (Widiger & Costa, 2013). Approaches may be both dimensional and transdiagnostic, such as the Hierarchical Taxonomy of Psychopathology (HiTOP) model: a hierarchical organization of mental disorder, consisting of transdiagnostic spectra at the top (e.g., general psychopathology, internalizing/externalizing) and syndromal subfactors (e.g., eating problems), shared symptoms/signs, and traits at progressively lower levels (see Kotov et al., 2017).
Although transdiagnostic and dimensional approaches may provide useful alternative means for classifying psychological problems, they do not necessarily describe psychopathological phenomena any more fully than diagnostic syndromes. In some cases, they even provide weaker descriptions. For example, the internalizing/externalizing model, although highlighting links between diagnostic categories and thus traversing arbitrary diagnostic boundaries, provides even less information about mental disorders. Describing a problem as an “internalizing disorder,” although useful for some purposes, gives very little information about its precise nature (e.g., whether it involves anxiety, mood, eating, etc.), or the minutiae of its presentation (i.e., the factors and mechanisms that comprise the problem). The HiTOP model provides somewhat more information than diagnostic syndromes thanks to its hierarchical structure, which conceptualizes psychopathological problems at both more general levels (e.g., spectra levels) and more specific levels (e.g., symptoms, signs, and traits). However, this model still lacks the richness of information necessary for a compositional explanation: symptoms, signs, and traits are not broken down into lower level factors or mechanisms, nor are any relevant relationships between them modelled. Furthermore, although disorders are conceptualized at broader, transdiagnostic scales, they are not described contextually at higher levels of analysis (e.g., sociocultural, interpersonal, political); layers of meaning necessary to fully comprehend any psychopathological problem.
Symptom network models
The network theory of mental disorder proposes that psychopathological symptoms should be conceptualized as causing each other (e.g., persecutory delusions resulting in paranoia, subsequently leading to social withdrawal) rather than caused by an underlying “disease” process (e.g., delusions, paranoia, and social withdrawal as arising from a common cause, such as a neurobiological dysfunction or genetic mutation; Borsboom, 2017). Symptom network models (SNMs) apply this theory to specific syndrome clusters—often, but not always, DSM/ICD syndromes. A network structure is generated by depicting the causal links between symptoms of that condition, including their strength and direction (Borsboom, 2017). SNMs can also model the relationships between symptoms across disorders (e.g., eating disorders and depression/anxiety; Smith et al., 2018), which makes them particularly useful in accounting for comorbidity (Fried et al., 2017). However, although SNMs provide a useful and interesting description of psychopathological symptom relationships, they still fail to provide effective compositional explanations.
SNMs, like DSM/ICD syndromes, lack explanatory scope and depth. Although the relationships between symptoms are elaborated within these models, the nature of the symptoms themselves is not fully explained: each is represented largely at the phenomenological level, rather than at each level of analysis (e.g., molecular, neural, physiological, cognitive/psychological, interpersonal, sociocultural). For example, anhedonia, a key symptom of depression, can be represented at the phenomenological level as involving both decreased “liking” and decreased “wanting,” at the cognitive level as a reduced hedonic capacity, reduced reward motivation, and errors in reward learning, at the neural level as dysfunction in the “hedonic network” and mesolimbic pathways, and at the molecular level as reductions in opioid and dopaminergic activity (see Clack & Ward, 2020). Compared to a full analysis such as this, the descriptions of symptoms given in SNMs are significantly underpowered. They may act as partial compositional explanations, certainly—as models depicting the relationships between psychopathological symptoms—but lack the depth of detail required to fully explain the constitution of the disorders with which they are concerned.
Historical accounts
We refer here to descriptions of disorder states that accompanied the first identification of a psychiatric syndrome or were developed around the time of the DSM-III (published in 1980), which represented a paradigm shift towards our current conceptualization of mental disorders (Mayes & Horwitz, 2005). Examples include Russell’s (1979) initial characterization of bulimia nervosa and Bruch’s (1973, 1978/1982) early descriptive accounts of anorexia nervosa, 1 considered the first “modern description” of the disorder (Marks, 2019). These sorts of accounts typically consist of a set of clinical case studies from which the author draws broader conclusions. For example, Russell’s (1979) initial characterization of bulimia nervosa involved 30 patients, three of whom were presented as illustrative case studies, from which he drew conclusions about the disorder’s typical features, such as demographics, symptomology, medical complications, and psychopathological correlates. These accounts are often more descriptively comprehensive than the classificatory approaches above. However, they are nevertheless unsuitable to serve as compositional explanations.
Most problematic is that their explanatory scope extends beyond composition. Although they do describe the presentation of a disorder—as a compositional explanation should—they often branch into hypothesizing its etiology as well. For example, as well as describing the disorder’s presentation, Bruch’s (1978/1982) account of anorexia nervosa makes numerous etiological claims—for example, “the child’s inability for constructive self-assertion and the associated deficits in personality development are the outcome of interactional patterns that began early in life” (p. 37)—including several chapters highlighting precipitating factors and speculating on the causal role of family dynamics (e.g., chapters “The Perfect Childhood” and “How It Starts”). Although this information may hold relevance in a clinical context, theoretically it conflates the theoretical tasks of compositional and etiological explanation. Although these tasks are related—each informing the other—they are conceptually distinct, requiring different modes of theoretical reasoning: causal versus compositional (see above). Attempting to achieve both within a single account is therefore likely to create convolution and promote logical errors, thereby impairing the integrity of each task.
Historical accounts also tend to lack empirical foundation: being either the first or one of the earliest descriptions of a disorder, there was typically little empirical research to inform their construction. They are therefore most often based on a small number of case studies which, although potentially the best option available at the time, falls short of modern scientific standards. For example, there is typically extensive sampling bias: samples are generally comprised solely of the author’s existing patients, and therefore (due to reduced access to psychiatric treatment at the time; Mechanic, 2007) likely to be skewed towards those of higher socioeconomic status and European descent. Furthermore, cases are often aggregated in pseudoscientific or anecdotal ways to illustrate the author’s points, rather than analysed in a valid statistical manner.
There has been little structured effort to update or expand such accounts in line with contemporary research, despite many still being used to inform it. Although some aspects of historical accounts can now be empirically verified, there are still many claims that current research fails to support or actively refutes. For example, Bruch’s (1978/1982) account of anorexia nervosa describes the disorder as affecting “the daughters of well-to-do, educated, and successful families, not only in the United States but in many other affluent countries” (p. vii), implying that anorexia nervosa exclusively affects WEIRD populations (i.e., Western Educated Industrialized Rich Democratic; Henrich et al., 2010). However, research now indicates anorexia nervosa is not a culture-bound syndrome (Keel & Klump, 2003; Pike et al., 2014), and shows no reliable association with ethnicity or socioeconomic status (Hadassah Cheng et al., 2018; Schaumberg et al., 2017).
Finally, as with DSM/ICD syndromes, historical accounts tend to refer to only the most salient features of a disorder (i.e., the phenomenological level) and therefore neglect those that are more deep-seated and less easily observed (e.g., neural network dysfunction, alterations in hormone or neurotransmitter systems), though nonetheless relevant.
Case narratives
Case narratives are often provided to students and clinicians in textbooks or treatment manuals to demonstrate how a disorder typically presents. Their descriptions usually include information about the characteristic symptoms and signs, demographics (e.g., age, gender), relevant history (e.g., familial, medical, psychiatric), and triggers for that disorder. For example: Anna, a 15-year old girl of European descent, presented with extreme weight loss and low appetite. Her BMI had fallen from 19 (healthy for her age group) to 16 within the last six months, such that she was substantially underweight. A recent check-up revealed no underlying medical explanations for her weight-loss. Anna’s mother reported she had been refusing to join family meals, confining herself often to her bedroom, and eating a drastically reduced diet. She had also stopped spending as much time with friends and increased her exercise regime significantly—running for 1–2 hours every day, in addition to competitive swimming training. When her parents expressed their concerns to Anna, she tended to either burst into tears or shout at them. Anna was unconcerned about her weight-loss and denied that her eating or exercise behaviour was a problem. She expressed significant body dissatisfaction and drive for thinness, complaining she was “too fat” and wishing she were thinner. According to her mother, Anna had had many challenges throughout her development and had previously seen a child psychologist for anxiety.
As with both diagnostic syndromes and historical accounts, case narratives tend to refer only to features of the disorder that are most clinically salient. For example, the exemplar above refers largely to the phenomenological experience of the client, Anna, and neglects to include information about any physiological, neural, or molecular processes. This makes sense given that such accounts are intended as prototypical examples of clinical presentation, and in practice one would not routinely engage in the methods of investigation required to identify more deep-seated structural phenomena (e.g., fMRI, CSF sampling). However, as previously discussed, it significantly limits their explanatory value.
Furthermore, although case narratives refer to significantly more features of the disorder than most classificatory approaches, they still lack depth in their descriptions. They fail to go into any features in detail—relying on brief, superficial descriptions despite the fact that these constructs are often multidimensional (e.g., body image; see above)—and continue to refer to the thinly defined constructs entrenched within psychopathological research (e.g., “drive for thinness,” “body image dissatisfaction”). Case examples are also just that: examples. Each presents a specific instance of a disorder. Hence, although many features may be represented, it is unlikely that all features relevant to the disorder will be included, as real-life cases seldom (if ever) present with every feature associated with the condition.
The Research Domain Criteria (RDoC)
The RDoC is a clinically independent research framework intended to guide empirical investigation into psychological mechanisms (Cuthbert & Insel, 2013). It was developed as a reaction to the publication of the DSM-5, which many perceived as being a conservative development on the previous edition (DSM-IV) that retained many of the problems originally identified (e.g., reification; 2 Whooley, 2014). The RDoC presents an “alternative nosological framework” (Whooley, 2014, p. 100) that seeks to advance psychopathology research—specifically, neurobiological investigation (Cuthbert & Insel, 2013; Whooley, 2014). The RDoC assumes that mental disorders are “brain disorders” born out of dysfunctions in neural circuitry, and therefore aims to build a nosology of mental disorder from the “bottom-up” using current neuroscience research (Cuthbert & Insel, 2013; Whooley, 2014). The hope is that by doing so we will develop more valid diagnostic categories, anchored in neurobiology (Lilienfeld, 2014; Whooley, 2014).
The RDoC framework provides a two-dimensional “matrix” to guide psychopathology research, consisting of six psychological “domains” of investigation—(a) negative valance systems (e.g., threat, loss), (b) positive valence systems (e.g., approach motivation, reward learning), (c) cognitive systems (e.g., attention, working memory), (d) systems for social processes (e.g., attachment, social communication), (e) sensorimotor systems (e.g., action selection, initiation, execution, habit development), and (f) arousal/modulatory systems (e.g., sleep-wake, arousal)—as well as seven “units of analysis”—(a) genes, (b) molecules, (c) cells, (d) neural circuits, (e) physiology, (f) behaviour, and (g) self-report (Cuthbert & Insel, 2013; Lilienfeld, 2014). The RDoC assumes that mental disorders result from disruptions in the normal-range functioning of these processes and thus applies basic understandings of psychology and neuroscience to psychopathological problems (Lilienfeld, 2014).
It is important that the role of the RDoC within the scientific inquiry process be accurately understood. The RDoC is a research framework intended to scaffold investigation into psychological processes—both their function and dysfunction—in order to obtain insight into psychopathology and thereby “inform future classification schemes [emphasis added]” (Insel et al., 2010, p. 748). Hence, although able to generate substantial data about psychological and psychopathological processes, the RDoC does not work to conceptualize these theoretically—that is, to create a coherent compositional account that links these findings together in relevant and meaningful ways. It therefore does not directly produce compositional explanations.
Even if it did include such a synthesis, the RDoC matrix is not directly geared towards studying psychopathological processes. Although intended to provide insight into mental disorder, one of the core philosophies of the RDoC is that investigation should be directed towards broader psychological processes (e.g., positive valence systems, cognitive systems)—how they both function and malfunction, and thereby may contribute to the development and maintenance of mental disorder—rather than specific psychopathological problems (Cuthbert & Insel, 2013). Hence, the picture developed by the RDoC framework is more likely to be a comprehensive understanding of these systems—including their role in psychopathology—rather than synthesized descriptions of particular mental disorders. This is not to say the RDoC cannot contribute valuably to the development of compositional explanations—the wealth of compositional data the framework has the power to generate would have great value for their construction. However, it is critical to note that the RDoC also encompasses etiological investigation (e.g., genetic research), and does not clearly distinguish between these two processes. As previously discussed, this conflates two distinct theoretical tasks (i.e., causal vs. compositional explanation) and may lead to problems farther along in the explanatory process.
The RDoC is also significantly neurocentric; asserting that mental disorders be considered “brain disorders” born out of dysfunctions in neurocircuitry (Cuthbert & Insel, 2013; Lilienfeld, 2014; Whooley, 2014). Although this approach has some benefits—for example, highlighting the role of neurobiological processes in mental disorder (at times discounted or neglected) and providing a platform for investigating neurobiological aspects of psychopathology—it largely sidelines other levels of explanation (e.g., phenomenological, sociocultural) despite their equal relevance to mental disorder. Of the seven “units of analysis” prescribed by the RDoC, five are biologically based—genes, molecules, cells, neurocircuitry, physiology—implying that neurobiological factors hold far greater explanatory weight (Lilienfeld, 2014).
Furthermore, even though one of the investigative domains concerns social/interpersonal phenomena (i.e., systems for social processes), broader sociocultural structures and influences are not addressed within the framework. Cultural factors are well-evidenced as playing a significant role in multiple aspects of psychopathology (e.g., etiology, maintenance, symptom expression) and we would argue that mental disorders cannot be understood independently of their social and cultural context. For example, some symptoms associated with posttraumatic stress disorder can be viewed as adaptive when viewed within specific contexts, including the precipitating trauma event (e.g., hypervigilance, physical hyperarousal, and emotional detachment may be useful in combat situations). Hence, even though the RDoC might lead to richer neurobiological understandings of psychopathology, it also risks decontextualizing mental distress such that lower levels of analysis (e.g., cellular, neural) are not considered within the broader context of the problem and higher levels (e.g., psychological, phenomenological, sociocultural, etc.) end up significantly underspecified despite their explanatory relevance (Whooley, 2014).
Building better descriptions
As we have seen, current options for compositional explanation are insufficient; failing to demonstrate the necessary empirical adequacy, depth, and scope. In this section therefore, we demonstrate how we believe effective compositional explanations could be constructed. To begin, we first outline the theoretical framework used to guide these ideas: the Phenomena Detection Method (PDM; Ward & Clack, 2019).
Guiding framework: The phenomena detection method
The PDM (Ward & Clack, 2019) is a metatheoretical framework for the detection and modelling of “clinical phenomena” (e.g., symptoms) which is not dependent on existing classification systems such as the DSM-5 or ICD (discussed above). Critically, it emphasizes the importance of developing compositional explanations of symptoms, making it highly relevant to the current problem. It has four phases: (1) formulating client complaints and/or accompanying signs, (2) discerning and analysing patterns in data related to these symptoms (i.e., detecting clinical phenomena), (3) constructing multiple models of the phenomenon using different levels or units of analysis, and (4) linking in etiological factors to develop causal explanations. Phases 1–3 are relevant to the construction of compositional explanations and thus inform our reasoning in this section.
Another important aspect of the PDM is its promotion of, and adherence to, a pluralistic account of scientific explanation, which states that scientific explanations should involve a collection of theoretical models that represent the constitution or causes of a symptom at and across different spatial and temporal scales, instead of trying to represent everything using a single model. This approach is known as model pluralism and has been widely recognized as a promising way forward in both the biological and social sciences, in which the phenomena of interest are of a high level of complexity (Hochstein, 2016; Mitchell & Dietrich, 2006; Potochnik, 2010; Ruphy, 2016). The PDM endorses this by prescribing the construction of multiple compositional or causal models of the explanatory target at a range of spatial scales and levels of abstraction.
From syndromes to symptoms
A key impediment to current descriptive accounts acting as compositional explanations is the fact that most are built around DSM/ICD syndromes, which are widely acknowledged to possess numerous conceptual flaws, such as symptomatic heterogeneity and rampant comorbidity (e.g., Nielsen & Ward, 2020; Whooley, 2014). Building compositional explanations using these constructs is therefore problematic: any resultant description, no matter how rich or detailed, will lack a certain amount of validity as the very construct being described is conceptually flawed. Continuing to use these constructs theoretically also further entrenches them in research and practice, thereby impeding the development of better classificatory approaches. Hence, on our view, the first step towards building better compositional explanations is to transition away from these syndromes as the foci of explanation.
Several theorists argue that, to move forward, psychological explanation should, at least for now, focus on symptoms rather than syndromes (Berrios, 2013; Borsboom, 2017; Ward & Clack, 2019; Wilshire et al., in press). For instance, instead of trying to describe and explain the syndrome bulimia nervosa, which comprises a cluster of diverse symptoms, one would instead focus on a single symptom of that pathology, such as binge eating. This approach makes sense conceptually, as symptoms and signs have greater validity than DSM/ICD syndromes; arguably representing genuine phenomena as opposed to artificial categories. Compared to these syndromes, symptoms have more defined boundaries, less heterogeneity, and greater stability. For example, the symptom binge eating is more obviously distinct from other symptoms (e.g., self-starvation, purging) than the syndrome bulimia nervosa is from other eating disorder diagnoses (e.g., anorexia nervosa). A client shifting from this symptom presentation to an alternative presentation (e.g., self-starvation) or to a state of recovery is also likely to be much more psychologically meaningful than a transition from one eating disorder diagnosis to another, which can currently be accomplished by changes in arbitrary factors like BMI. Practically, it is also useful to reduce the scope of our explanatory focus: it is much easier to detail the composition of a single symptom than a large and diverse collection of them (i.e., a syndrome).
Symptoms and signs also make for more appropriate foci at an ethical level, as they represent the actual concerns of clients: each is a valid and important aspect of the client’s difficulties that we should aim to understand. At the coarser grain size of syndromes, although we are getting a concise and practical account, we may neglect the description and explanation of some symptoms in favour of providing a brief and uncomplicated overall account. At finer grain sizes, such as the neurobiological (e.g., the RDoC), although generating useful and detailed information that can be used to inform theoretical conceptualization of clinical phenomena, we are no longer centring our accounts on client problems—which arguably should be our paramount concern as clinicians—and risk decontextualizing their distress (Whooley, 2014). Although to fully describe symptoms we no doubt need to investigate and describe phenomena at smaller scales (e.g., neurobiological, molecular) and consider their relationship to other symptoms (e.g., syndromes, symptom networks), we argue that the appropriate starting point for compositional explanations should, at least for now, be psychopathological symptoms.
Starting with data
Compositional explanations should be constructed using empirical evidence: reasoning abductively from data to identify constructs and processes relevant to the phenomenon in question (Haig, 2014). This begins with an unbiased gathering of relevant data—for example, cross-sectional research from a variety of disciplines involving those presenting with that symptom/sign—that is of high methodological quality—for example, RCTs, meta-analyses, systematic reviews, methodologically rigorous single studies—which is then mined for patterns that might represent compositional constructs or processes (Hawkins-Elder & Ward, 2020b).
For instance, within the symptom binge eating, we may theorize the existence of the phenomenon impaired inhibitory control based on meta-analyses and systematic reviews showing that individuals who exhibit binge eating demonstrate poorer performance on planning (e.g., Farstad et al., 2016), decision-making (e.g., Guillaume et al., 2015; Wu et al., 2016), and set-shifting tasks (e.g., Wu et al., 2016), higher self-reported impulsivity (e.g., Farstad et al., 2016; Steward et al., 2017), and frequent engagement in other impulsive or reckless behaviours (e.g., self-harm, substance abuse; Peebles et al., 2011). Having a range of different, and methodologically robust, data all pointing to the existence of impaired inhibitory control means we can be more confident that this phenomenon is genuinely present, rather than the false product of biased reasoning or methodological error. A full model constructed in this manner will therefore be a more accurate representation of the phenomenon of interest.
Multilevel explanation
Compositional explanations should describe their phenomenon at all relevant levels of analysis—for example, molecular, neurological, cognitive, phenomenological, interpersonal, contextual/sociocultural, and so forth. Indeed, Zachar (2008) describes psychopathological phenomena as structures with “many overlapping levels” and argues that “having alternative models better reflects the domain of psychiatric disorders” (pp. 339–340). Compositional explanations should therefore be similarly multilevel in order to adequately reflect this.
To accomplish this, we recommend building models in a “stacked” format, beginning with the phenomenological level (at which the symptom/sign is reported/observed) and moving outwards, considering each level of analysis in turn, to identify factors and processes that might be constitutionally relevant. This is useful because factors and processes at one level may partially constitute or be otherwise related to those at another and building outwards in this manner may help the researcher to make these connections.
Consider the symptom binge eating: at the phenomenological level, we can recount how this symptom is experienced by the client based on self-report data from empirical research: privacy is important (secretive eating; Lydecker & Grilo, 2019), emotion is often involved (emotional eating; e.g., Leehr et al., 2015; Ricca et al., 2012), individuals typically perceive a lack of control over their eating (e.g., Colles et al., 2008), and may experience strong physical hunger and hedonic craving for food (e.g., Ng & Davis, 2013; Witt & Lowe, 2014; see Figure 1). From this level, we can then consider factors and processes suggested by empirical research at lower levels that may comprise this symptom. For example, at a cognitive level, the reported hunger and craving could be represented as a heightening of appetite 3 (see Figure 1). This may be partially constituted at the physiological level by an impaired appetite feedback system (e.g., imbalances in hunger and satiety hormones, Culbert et al., 2016; altered vagal nerve transmission, Peschel et al., 2016) and at the neurological level by interoceptive network deficits (e.g., insular dysfunction, Gasquoine, 2014; Klabunde et al., 2017) and alterations in reward pathways (e.g., Avena & Bocarsly, 2012; Frank, 2013; Wierenga et al., 2014). Further down at the molecular level, serotonin dysregulation (e.g., Compan et al., 2012) may be partially responsible for the experience of hunger surrounding a binge (due to its role in appetite regulation; e.g., Lam et al., 2010), and the experience of intense craving may be influenced by dysregulation in opioid and dopaminergic systems implicated in reward and addiction (e.g., Berridge, 2009; Majuri et al., 2017). At a higher level, it is also worth considering how sociocultural factors may influence how the symptom is experienced (e.g., enabling or inhibiting certain behaviours, altering symptom content). For instance, overeating, for metabolic or hedonic reasons, is somewhat dependent on socioeconomic food security (e.g., Anderson-Fye, 2018), as food must be available in reasonable abundance for it to be overconsumed. Similarly, the content of cravings is likely to be influenced by the individual’s cultural environment (e.g., Osman & Sobal, 2006).
Illustration of a multilevel approach to compositional explanation, using the symptom binge eating.
We can conduct the same process for each aspect of a symptom to build a multilevel explanation of its constitution (see Figure 1). Such an explanation provides a rich description of the symptom, as it details the relevant factors and processes at all levels and considers how these may comprise or influence each other.
Detailing domains
An adequate compositional explanation should describe all aspects of the phenomenon in a high level of detail. One way of doing this is to build smaller compositional models of constructs “nested” within the larger account. This helps to avoid overcomplicating the broader model with specifics, allowing it to present a streamlined overview, but ensures that all constructs are sufficiently outlined and the overall explanation is descriptively rich.
Consider the multilevel model of binge eating previously sketched out: although this model provides a good overview of constitutional factors and processes at each level, the constructs referred to within each domain are still in need of further definition. By constructing nested “submodels” of these phenomena, we can more clearly define the factors and processes invoked and thereby enrich the overall account. For example, consider heightened appetite (identified at the cognitive level). Based on the literature, we can construct a more detailed compositional submodel of this construct (see Figure 2).
Example of a compositional submodel within the symptom binge eating, detailing the nested cognitive phenomenon heightened appetite.
Individuals who binge eat are more sensitive to the effects of reward (e.g., Harrison et al., 2010; Wierenga et al., 2014), including reward from food (e.g., Schag et al., 2013). They are therefore likely to have greater hedonic hunger (e.g., Witt & Lowe, 2014) than nonbinge-eating individuals. Binge eaters also demonstrate impaired satiety (e.g., Sysko et al., 2007). Their metabolic hunger is therefore also likely greater, as they have less indication when they are full. Finally, binge eaters reliably demonstrate poorer interoception—ability to sense and interpret internal sensations—than nonbinge eaters (e.g., Jenkinson et al., 2018; Klabunde et al., 2017). They may therefore struggle to detect physical cues, including appetitive signals, and accurately interpret them, sometimes misattributing physical sensations as hunger or satiety. This may make it harder for them to appropriately modulate eating behaviour according to their body’s metabolic needs (e.g., Herbert & Pollatos, 2018).
Submodels at one level can also be linked to submodels at other levels to further enrich the explanation. For example, we might construct a physiological submodel of the impaired appetite feedback system (see Figure 3), involving alterations in the baseline levels and responses of appetitive hormones (e.g., Culbert et al., 2016; Prince et al., 2009), increased gastric capacity and delayed gastric emptying (e.g., Klein & Walsh, 2004), and decreased ascending vagal nerve transmission (e.g., Simmons & DeVille, 2017). This can then be linked to the cognitive model by relating the lower level physiological processes to those at the higher level. For example, the impaired satiety described at the cognitive level is likely partially constituted by these hormonal and gastric differences (see Figure 3; e.g., Berthoud, 2008; Zanchi et al., 2017). Aberrant hormonal functioning could also partially comprise the increased sensitivity to food-related reward due to the influence of some hormones on dopaminergic networks (e.g., leptin; Cassioli et al., 2020). Vagal nerve dysregulation may likewise play a part in impaired interoception, as it plays a key role in transmitting sensory information from the body to the brain (Craig, 2002). This kind of intermodel linking further enriches the overall compositional picture, contributing to a more in-depth account of the symptom’s composition.
Example model linking cognitive (heightened appetite) and physiological (impaired appetite feedback system) submodels within the symptom binge eating.
Conclusion
In this article, we suggested how better compositional explanations could be constructed by focusing on symptoms rather than syndromes, using empirical research, and creating detail-rich models spanning all levels of analysis. At present, we lack theoretically oriented and descriptively rich accounts of how psychopathological problems are constituted. The absence of these compositional explanations is of significant concern, as they hold genuine value for both research and practice. The kind of nested modelling outlined above is a good example of how originally conceptually thin phenomena can be elaborated into rich, multilayered compositional accounts. Developing a network of compositional models at different levels of analysis may yield insight into the structures and processes constituting disorders that could, ultimately, result in stronger etiological explanations, more accurate taxonomies, and more precisely targeted treatment. In our view, greater attention to the compositional explanation of psychopathological symptoms is a crucial step towards ameliorating the social costs and personal suffering of mental illness.
Footnotes
Acknowledgements
The authors would like to thank the EPC Lab at Victoria, as well as Alexander Moses for his assistance in designing the figures for the paper.
Declaration of Conflicting Interests
The authors declare that there is no conflict of interest.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
