Cramer et al.'s network perspective is a welcome innovation in a field whose problems and practices are equally stubborn. As a methodological tool, network analysis could enhance psychiatric diagnosis by clarifying symptom centrality and improve classification by mapping the “symptom space” in ways that may reduce comorbidity. The network perspective also provides an appealing rationale for paying attention to symptoms as phenomena that matter in themselves rather than merely being superficial and causally impotent markers of reified “disease entities” (Hyman Reference Hyman2010). This ontological upgrading of symptoms would be popular with behavioral clinicians (Persons Reference Persons1986), whose treatments directly target symptoms and who tend to find talk of “latent variables” a little beside the point. Indeed, I suspect that many clinicians would agree with the authors in comprehending mental disorders as constituted by their symptoms rather than revealed in them. Diagnostic labels are often treated as useful shorthands, rather than as natural kind terms.
Although focusing attention on symptoms and their interrelations has scientific and clinical advantages, it is questionable to go further and rule out in principle the existence of latent psychopathologies. Cramer et al.'s elimination of latent variables goes too far, and as a result the authors' account overreaches. There is no intrinsic incompatibility between network modelling and recognition of latent categories (Schweinberger & Snijders Reference Schweinberger and Snijders2003), nor should there be any incompatibility between believing that some symptom clusters reflect latent classes and that symptoms within those clusters may have causal interrelationships. Measurement models may demand local independence (i.e., no association among symptoms after the latent variable is statistically controlled) and equal centrality of all symptoms, but these are limitations in their formalism, not a reason for dismissing the existence of latent variables in the domain of psychopathology.
The authors present a starkly dichotomous view of disorders. On the one hand, there are medical disorders such as Down syndrome that have discrete boundaries, essence-like etiologies, defining features, and isolated symptom clusters. The authors recognize that such disorders are latent categories, and symptom networks will therefore represent them incompletely. On the other hand are psychiatric disorders, which have “necessarily fuzzy” boundaries, diffuse etiologies, and no defining features. Their symptom clusters form densely interconnected webs that can be separated into distinct disorders only by arbitrary division. This view can be challenged because the distinction between essentialized natural kinds and arbitrary symptom clusters is too polarized, and because psychiatric disorders are not uniformly of the latter type.
Cramer et al. present psychiatric conditions as all alike in lacking sharp, non-arbitrary category boundaries. No doubt this is true of many disorders, which simply represent quantitative extremes on continuous dimensions. However, as I have argued elsewhere (Haslam Reference Haslam2002), mental disorders are structurally diverse. Many are differentiated from normality by an imposed and convention-based decision rule. Others have boundaries that are intrinsic rather than conventional, but those boundaries are themselves indistinct or fuzzy, representing gradations of abnormality. Still others are latent classes with discontinuous boundaries. The parameters of the category (e.g., its prevalence) are empirical matters that do not simply reflect where a Diagnostic and Statistical Manual of Mental Disorders (DSM) committee chose to set a diagnostic threshold.
The evidence that some mental disorders are latent categories comes largely from research using taxometric methods, which Paul Meehl and colleagues (Meehl Reference Meehl1995) developed to distinguish between (latently) categorical and dimensional models of psychiatric phenomena. The preponderance of taxometric findings support dimensional models of mental disorders, consistent with Cramer et al.'s view, but several latent categories have been found (e.g., autism, schizotypal personality; for a review, see Haslam Reference Haslam2003). It is implausible that these latent categories represent natural kinds in the essentialist sense – they may arise from threshold effects, complex interactions among multiple causal factors, and so on, rather than from a single underlying causal factor, or “specific etiology” – but they have non-arbitrary category boundaries and some kind of underlying causal process or mechanism that makes them coherent. The philosophical concept of “homeostatic property clusters” (Kornblith Reference Kornblith1993), intended to strike a middle path between essentialism and conventionalism, is somewhat related. These clusters are “real divisions in the structure of the world” (Craver Reference Craver2009, p. 577) that cohere not because they share an essence but through the operation of a similarity-generating mechanism. Demonstrating that some psychiatric disorders are taxonic in Meehl's sense does not commit one to essentialism, but it does invalidate any claim that all disorders are arbitrary and all boundaries fuzzy.
Taxometric research has direct relevance to questions of comorbidity (Meehl Reference Meehl2001; Waldman & Lilienfeld Reference Waldman and Lilienfeld2001). Comorbidity is meaningful as a concept only if the supposedly comorbid conditions are latent categories. Two disorders cannot truly co-occur in a person unless both are discrete and separable. If disorders are not taxa, then comorbidity merely represents overlap of symptoms from different diagnostic lists. Cramer et al. hold to this view of comorbidity, according to which it is primarily a nuisance to be eliminated by nosological revision. By excluding on principle the existence of latent psychiatric categories, they fail to acknowledge that some comorbidity may be real and meaningful.
I share Cramer et al.'s belief that most current psychiatric diagnoses do not pick out natural categories and that essentialist and reifying accounts of them should be vigorously challenged. For this reason, and because network analysis is a promising tool for mapping symptom space and reining in the promiscuous comorbidity of current diagnostic practice, I applaud their work. However, I hold a more pluralistic view of psychiatric classification than they do, believing that some psychiatric conditions approximate real categories with non-arbitrary and non-fuzzy boundaries. For this reason, I question the ontological position that they adopt, according to which all psychiatric boundaries are fuzzy, all distinctions arbitrary, and all comorbidity spurious. It should be possible to reap the benefits of network analysis without committing to a position that the symptom level is the only one that is real, and that latent categories could not occur within psychiatry.
You have
Access
Cramer et al.'s network perspective is a welcome innovation in a field whose problems and practices are equally stubborn. As a methodological tool, network analysis could enhance psychiatric diagnosis by clarifying symptom centrality and improve classification by mapping the “symptom space” in ways that may reduce comorbidity. The network perspective also provides an appealing rationale for paying attention to symptoms as phenomena that matter in themselves rather than merely being superficial and causally impotent markers of reified “disease entities” (Hyman Reference Hyman2010). This ontological upgrading of symptoms would be popular with behavioral clinicians (Persons Reference Persons1986), whose treatments directly target symptoms and who tend to find talk of “latent variables” a little beside the point. Indeed, I suspect that many clinicians would agree with the authors in comprehending mental disorders as constituted by their symptoms rather than revealed in them. Diagnostic labels are often treated as useful shorthands, rather than as natural kind terms.
Although focusing attention on symptoms and their interrelations has scientific and clinical advantages, it is questionable to go further and rule out in principle the existence of latent psychopathologies. Cramer et al.'s elimination of latent variables goes too far, and as a result the authors' account overreaches. There is no intrinsic incompatibility between network modelling and recognition of latent categories (Schweinberger & Snijders Reference Schweinberger and Snijders2003), nor should there be any incompatibility between believing that some symptom clusters reflect latent classes and that symptoms within those clusters may have causal interrelationships. Measurement models may demand local independence (i.e., no association among symptoms after the latent variable is statistically controlled) and equal centrality of all symptoms, but these are limitations in their formalism, not a reason for dismissing the existence of latent variables in the domain of psychopathology.
The authors present a starkly dichotomous view of disorders. On the one hand, there are medical disorders such as Down syndrome that have discrete boundaries, essence-like etiologies, defining features, and isolated symptom clusters. The authors recognize that such disorders are latent categories, and symptom networks will therefore represent them incompletely. On the other hand are psychiatric disorders, which have “necessarily fuzzy” boundaries, diffuse etiologies, and no defining features. Their symptom clusters form densely interconnected webs that can be separated into distinct disorders only by arbitrary division. This view can be challenged because the distinction between essentialized natural kinds and arbitrary symptom clusters is too polarized, and because psychiatric disorders are not uniformly of the latter type.
Cramer et al. present psychiatric conditions as all alike in lacking sharp, non-arbitrary category boundaries. No doubt this is true of many disorders, which simply represent quantitative extremes on continuous dimensions. However, as I have argued elsewhere (Haslam Reference Haslam2002), mental disorders are structurally diverse. Many are differentiated from normality by an imposed and convention-based decision rule. Others have boundaries that are intrinsic rather than conventional, but those boundaries are themselves indistinct or fuzzy, representing gradations of abnormality. Still others are latent classes with discontinuous boundaries. The parameters of the category (e.g., its prevalence) are empirical matters that do not simply reflect where a Diagnostic and Statistical Manual of Mental Disorders (DSM) committee chose to set a diagnostic threshold.
The evidence that some mental disorders are latent categories comes largely from research using taxometric methods, which Paul Meehl and colleagues (Meehl Reference Meehl1995) developed to distinguish between (latently) categorical and dimensional models of psychiatric phenomena. The preponderance of taxometric findings support dimensional models of mental disorders, consistent with Cramer et al.'s view, but several latent categories have been found (e.g., autism, schizotypal personality; for a review, see Haslam Reference Haslam2003). It is implausible that these latent categories represent natural kinds in the essentialist sense – they may arise from threshold effects, complex interactions among multiple causal factors, and so on, rather than from a single underlying causal factor, or “specific etiology” – but they have non-arbitrary category boundaries and some kind of underlying causal process or mechanism that makes them coherent. The philosophical concept of “homeostatic property clusters” (Kornblith Reference Kornblith1993), intended to strike a middle path between essentialism and conventionalism, is somewhat related. These clusters are “real divisions in the structure of the world” (Craver Reference Craver2009, p. 577) that cohere not because they share an essence but through the operation of a similarity-generating mechanism. Demonstrating that some psychiatric disorders are taxonic in Meehl's sense does not commit one to essentialism, but it does invalidate any claim that all disorders are arbitrary and all boundaries fuzzy.
Taxometric research has direct relevance to questions of comorbidity (Meehl Reference Meehl2001; Waldman & Lilienfeld Reference Waldman and Lilienfeld2001). Comorbidity is meaningful as a concept only if the supposedly comorbid conditions are latent categories. Two disorders cannot truly co-occur in a person unless both are discrete and separable. If disorders are not taxa, then comorbidity merely represents overlap of symptoms from different diagnostic lists. Cramer et al. hold to this view of comorbidity, according to which it is primarily a nuisance to be eliminated by nosological revision. By excluding on principle the existence of latent psychiatric categories, they fail to acknowledge that some comorbidity may be real and meaningful.
I share Cramer et al.'s belief that most current psychiatric diagnoses do not pick out natural categories and that essentialist and reifying accounts of them should be vigorously challenged. For this reason, and because network analysis is a promising tool for mapping symptom space and reining in the promiscuous comorbidity of current diagnostic practice, I applaud their work. However, I hold a more pluralistic view of psychiatric classification than they do, believing that some psychiatric conditions approximate real categories with non-arbitrary and non-fuzzy boundaries. For this reason, I question the ontological position that they adopt, according to which all psychiatric boundaries are fuzzy, all distinctions arbitrary, and all comorbidity spurious. It should be possible to reap the benefits of network analysis without committing to a position that the symptom level is the only one that is real, and that latent categories could not occur within psychiatry.