From the force fields and potential energy of physicists to the traits, temperaments, and abilities of psychologists, making dispositions into something real is deeply entrenched in science. What is radical about re-thinking the scientific validity of latent variables is that they are contemporary proxies for dispositions. Latent variables are omnipresent in contemporary psychology, and dispositions even more so.
I agree with Cramer et al. that psychiatric comorbidity refers to a real, important phenomenon that should not be dismissed as an artifact of the classification system (Zachar Reference Zachar2009). For example, epidemiological research in the United States indicates that in any given year a majority of psychiatric disorders occur in only 14% of the people (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994). This 14% likely represents a vulnerable population. How should we model this vulnerability? One current view is that neuroticism, a historically important latent variable in scientific psychology, is a primary risk factor for psychopathology (Clark Reference Clark2005; Kahn et al. Reference Kahn, Jacobson, Gardner, Prescott and Kendler2005; Mineka et al. Reference Mineka, Watson and Clark1998; Rothbart & Ahadi Reference Rothbart and Ahadi1994).
According to Cramer et al., however, the reality of a symptom pattern is not found hidden behind the symptoms, but in the symptoms themselves. They argue that symptoms are not effects of disorders, but rather, that relations between symptoms constitute disorders. This claim resembles the classical empiricists' contention that we have no knowledge of an underlying substance called matter, that is, what we know are observable properties only (color, hardness, etc.).
Given that latent variables are proxies for unobservable dispositions, the conceptual model described by Cramer et al. can be usefully applied to many hypothetical constructs in psychology, whether or not they are confirmable as latent variables. Exploring this a bit further can shed some light on what is being claimed about psychiatric disorders. Let me, therefore, briefly consider what the model might say about basic emotions theory, specifically, the non-essentialist model of James Russell (Reference Russell2003; Reference Russell2008).
Russell is an anti-realist about affect programs in the same way that Cramer et al. appear to be anti-realists about latent variables. According to Russell, an emotion such as fear represents a family of states that have varying degrees of similarity to one another. Russell claims that different tokens of fear share overlapping components, but there is no set of components that all episodes of fear must share. Two or more episodes classified as fear could potentially have very few components in common. In philosophical terms, fear components such as raised eyebrows and elevated blood pressure are not manifestations of a natural kind; rather, they are individual events that happen to co-occur in ways that we have learned to notice. If scientists could account for all the mechanisms that explain the different parts of an emotional episode, says Russell, there would be no need to posit an additional mechanism called the affect program to explain the emotional episode itself.
Affect programs, like latent variables, are constructs that are attributed ontological significance. Why? Because it is important to account for the patterning that occurs, and affect programs fill that role. They are similar to what Medin and Ortony (Reference Medin, Ortony, Vosniadou and Ortony1989) call essence placeholders. The key point for Russell is that the affect program model commits us to a false ontology. If we continue to use this ontology, we are not going to discover adequate scientific explanations of emotional phenomena. We need a better ontology. Similar to Cramer et al.'s claims about latent variables, Russell would prefer to eliminate the construct of affect programs from our scientific ontology, and with it the notion of basic emotions as legitimate scientific kinds.
One of the difficulties readers have with accepting Russell's model is that they expect the familiar categories of emotion to have some validity. They are uncomfortable with the possibility there is nothing more substantial about the patterns we do notice than those we do not notice. As a type of causal non-essentialism, Cramer et al.'s network model would allow that the patterning of emotion components occurs in a non-arbitrary way, but the patterning is best explained with reference to direct causal connections between those components rather than with reference to unobserved affect programs. It could support the elimination of affect programs, but unlike for Russell, Occam's razor would not be applied to basic emotions as well. Perhaps basic emotions can refer to clusters that have a greater probability of spontaneously emerging from the dynamic interaction of causally connected components – in the same way that amino acids are reliably folded into proteins without benefit of a latent protein-folding mechanism that makes it occur.
The difference between emotions and psychiatric disorders is that we do not typically think that psychiatric disorders emerge spontaneously like everyday psychological states emerge – disorders entail a failure of a normal mechanism (Wakefield Reference Wakefield1992). This is an important point because Cramer et al.'s argument is broader than the claim that shared symptoms, not latent variables, are the important causal factors in comorbidity; they also use relations among symptoms to explain psychiatric syndromes in general.
Perhaps the most important development in psychiatric classification in the last 20 years is the effort by many clinical psychologists and some psychiatrists to replace Robins and Guze's (Reference Robins and Guze1970) categorical approach to the medical model with a psychometrically based dimensional model (Livesley Reference Livesley, Phillips, First and Pincus2003; McCrae Reference McCrae, Costa and Widiger1994; Watson Reference Watson2005). What are the implications for the bold claim that latent variables (categorical or dimensional) are an inadequate basis for a science of psychopathology? For better or worse, success would mean that the network model might be a candidate for a new general model of psychopathology.
How does such a model fare as a general model of psychopathology? As currently formulated, the network model demarcates psychopathology as the collection of symptoms that are manifest in whatever psychiatrists decide to treat. Cases are defined when enough symptoms of the right type are present. This may be a bit too nominalistic. For example, what if psychiatrists start conceptualizing liberalism as a mental disorder – as some have suggested (Rossiter 2008; Savage Reference Savage2005)? What resources would exist for not including the symptoms of liberalism in the total psychiatric symptom space?
In the latent variable model, the pathology is supposedly located in the hidden reality behind the symptom patterns. If there is no reality behind the symptom patterns, would “pathologies” be relocated in the symptoms themselves? Could psychopathology be an emergent property? Such a radical relocation project raises a list of interesting complications. Latent variables can be taken as proxies for the underlying pathological processes of the medical model, but justifying the reality of emergent pathologies would require a different kind of thinking.
You have
Access
From the force fields and potential energy of physicists to the traits, temperaments, and abilities of psychologists, making dispositions into something real is deeply entrenched in science. What is radical about re-thinking the scientific validity of latent variables is that they are contemporary proxies for dispositions. Latent variables are omnipresent in contemporary psychology, and dispositions even more so.
I agree with Cramer et al. that psychiatric comorbidity refers to a real, important phenomenon that should not be dismissed as an artifact of the classification system (Zachar Reference Zachar2009). For example, epidemiological research in the United States indicates that in any given year a majority of psychiatric disorders occur in only 14% of the people (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994). This 14% likely represents a vulnerable population. How should we model this vulnerability? One current view is that neuroticism, a historically important latent variable in scientific psychology, is a primary risk factor for psychopathology (Clark Reference Clark2005; Kahn et al. Reference Kahn, Jacobson, Gardner, Prescott and Kendler2005; Mineka et al. Reference Mineka, Watson and Clark1998; Rothbart & Ahadi Reference Rothbart and Ahadi1994).
According to Cramer et al., however, the reality of a symptom pattern is not found hidden behind the symptoms, but in the symptoms themselves. They argue that symptoms are not effects of disorders, but rather, that relations between symptoms constitute disorders. This claim resembles the classical empiricists' contention that we have no knowledge of an underlying substance called matter, that is, what we know are observable properties only (color, hardness, etc.).
Given that latent variables are proxies for unobservable dispositions, the conceptual model described by Cramer et al. can be usefully applied to many hypothetical constructs in psychology, whether or not they are confirmable as latent variables. Exploring this a bit further can shed some light on what is being claimed about psychiatric disorders. Let me, therefore, briefly consider what the model might say about basic emotions theory, specifically, the non-essentialist model of James Russell (Reference Russell2003; Reference Russell2008).
Russell is an anti-realist about affect programs in the same way that Cramer et al. appear to be anti-realists about latent variables. According to Russell, an emotion such as fear represents a family of states that have varying degrees of similarity to one another. Russell claims that different tokens of fear share overlapping components, but there is no set of components that all episodes of fear must share. Two or more episodes classified as fear could potentially have very few components in common. In philosophical terms, fear components such as raised eyebrows and elevated blood pressure are not manifestations of a natural kind; rather, they are individual events that happen to co-occur in ways that we have learned to notice. If scientists could account for all the mechanisms that explain the different parts of an emotional episode, says Russell, there would be no need to posit an additional mechanism called the affect program to explain the emotional episode itself.
Affect programs, like latent variables, are constructs that are attributed ontological significance. Why? Because it is important to account for the patterning that occurs, and affect programs fill that role. They are similar to what Medin and Ortony (Reference Medin, Ortony, Vosniadou and Ortony1989) call essence placeholders. The key point for Russell is that the affect program model commits us to a false ontology. If we continue to use this ontology, we are not going to discover adequate scientific explanations of emotional phenomena. We need a better ontology. Similar to Cramer et al.'s claims about latent variables, Russell would prefer to eliminate the construct of affect programs from our scientific ontology, and with it the notion of basic emotions as legitimate scientific kinds.
One of the difficulties readers have with accepting Russell's model is that they expect the familiar categories of emotion to have some validity. They are uncomfortable with the possibility there is nothing more substantial about the patterns we do notice than those we do not notice. As a type of causal non-essentialism, Cramer et al.'s network model would allow that the patterning of emotion components occurs in a non-arbitrary way, but the patterning is best explained with reference to direct causal connections between those components rather than with reference to unobserved affect programs. It could support the elimination of affect programs, but unlike for Russell, Occam's razor would not be applied to basic emotions as well. Perhaps basic emotions can refer to clusters that have a greater probability of spontaneously emerging from the dynamic interaction of causally connected components – in the same way that amino acids are reliably folded into proteins without benefit of a latent protein-folding mechanism that makes it occur.
The difference between emotions and psychiatric disorders is that we do not typically think that psychiatric disorders emerge spontaneously like everyday psychological states emerge – disorders entail a failure of a normal mechanism (Wakefield Reference Wakefield1992). This is an important point because Cramer et al.'s argument is broader than the claim that shared symptoms, not latent variables, are the important causal factors in comorbidity; they also use relations among symptoms to explain psychiatric syndromes in general.
Perhaps the most important development in psychiatric classification in the last 20 years is the effort by many clinical psychologists and some psychiatrists to replace Robins and Guze's (Reference Robins and Guze1970) categorical approach to the medical model with a psychometrically based dimensional model (Livesley Reference Livesley, Phillips, First and Pincus2003; McCrae Reference McCrae, Costa and Widiger1994; Watson Reference Watson2005). What are the implications for the bold claim that latent variables (categorical or dimensional) are an inadequate basis for a science of psychopathology? For better or worse, success would mean that the network model might be a candidate for a new general model of psychopathology.
How does such a model fare as a general model of psychopathology? As currently formulated, the network model demarcates psychopathology as the collection of symptoms that are manifest in whatever psychiatrists decide to treat. Cases are defined when enough symptoms of the right type are present. This may be a bit too nominalistic. For example, what if psychiatrists start conceptualizing liberalism as a mental disorder – as some have suggested (Rossiter 2008; Savage Reference Savage2005)? What resources would exist for not including the symptoms of liberalism in the total psychiatric symptom space?
In the latent variable model, the pathology is supposedly located in the hidden reality behind the symptom patterns. If there is no reality behind the symptom patterns, would “pathologies” be relocated in the symptoms themselves? Could psychopathology be an emergent property? Such a radical relocation project raises a list of interesting complications. Latent variables can be taken as proxies for the underlying pathological processes of the medical model, but justifying the reality of emergent pathologies would require a different kind of thinking.