Theory-Based Clinical Representations, 1 Running head: THEORY-BASED CLINICAL REPRESENTATIONS Clinical Psychologists' Theory-Based Representations of Mental Disorders Predict their Diagnostic Reasoning and Memory

The theory-based model of categorization posits that concepts are represented as theories rather than as feature lists. Thus, it is particularly interesting that the DSM-IV (American Psychiatric Association, 1994), establishes a set of atheoretical guidelines for diagnosis in the domain of mental disorders. Five experiments investigated how clinicians handle an atheoretical nosology. Clinical psychologists' causal theories for DSM-IV disorders and their responses on diagnostic and memory tasks were measured. Participants were more likely to diagnose a hypothetical patient with a disorder if that patient had causally central rather than causally peripheral symptoms according to their theory of the disorder. They also showed biased memory for the causally central symptoms. Clinicians are cognitively driven to form and apply theories despite decades of training and practice with the DSM’s atheoretical guidelines. Theory-Based Clinical Representations, 3 Clinical Psychologists' Theory-Based Representations of Mental Disorders Predict their Diagnostic Reasoning and Memory The theory-based view of categorization proposes that concepts are represented as theories or causal explanations. Murphy and Medin (1985) suggested that our naïve theories about the world hold the features of a concept together in a cohesive package. For instance, a layperson's concept of anorexia not only contains the features "fear of becoming fat" and "refuses to maintain minimal body weight," but also the notion that the fear of becoming fat helps cause the refusal to maintain minimal body weight (Kim & Ahn, 2002). Indeed, a growing body of evidence supports the notion that the human mind constantly seeks out rules and explanations that make sense of incoming data concerning its surroundings, and forms concepts based on its theories about the world (Carey, 1985; Gelman, 2000; Keil, 1989; Murphy & Medin, 1985). A considerable number of studies has demonstrated theory-based categorization to date (e.g., Ahn, Kim, Lassaline, & Dennis, 2000; Gelman & Kalish, 1993; Medin & Shoben, 1988; Ross, 1997; Wisniewski & Medin, 1994), mostly in artificial and common everyday categories. The aim of the current paper is to examine what kind of reasoning occurs for a real-life domain in which official guidelines for categorization deliberately attempt to minimize a prescribed theoretical structure. Specifically, the current study investigates how clinical psychologists operate in the domain of mental disorders. This population and domain are highly unique in that clinical psychologists have been guided since 1980 by atheoretical manuals for the diagnosis of mental disorders (Follette & Houts, 1996). Most mental disorders lack a single universally Theory-Based Clinical Representations, 4 acknowledged pathogenesis, which in the past led to unreliability between clinicians in diagnosis. The Diagnostic and Statistical Manual of Mental Disorders’ (American Psychiatric Association, 1994) widely acclaimed solution was to ferret out syndromal clusters of symptoms that clinicians, regardless of theoretical orientation, could agree upon. The manual itself represents disorders as checklists of symptoms, and does not attempt to supply “an organizing theory that describes the fundamental principles underlying the taxonomy” (Follette & Houts, 1996, p. 1120). The introduction of the DSM-IV states that “DSM-III introduced a number of important methodological innovations, including explicit diagnostic criteria...and a descriptive approach that attempted to be neutral with respect to theories of etiology” (APA, 1994, pp. xvii-xviii). This approach was also adopted in the 4 version of the DSM. Indeed, the DSM casebook, used in training, encourages clinical psychologists to search for symptoms in their patients that match up with DSM-IV (APA, 1994) diagnostic criteria, without explicitly instructing them to incorporate any additional notions they may have of how these symptoms may affect each other (Spitzer, Gibbon, Skodol, Williams, & First, 1994). Furthermore, the DSM-IV (APA, 1994) states that if a subset of the diagnostic criteria list is present in the patient, that is sufficient for a diagnosis regardless of which combination of symptoms appears. The manual thereby assumes that all symptoms in the list are equally central to the disorder. For example, any 2 of the following 5 symptoms should warrant a diagnosis of schizophrenia, according to the DSM-IV (APA, 1994): hallucinations, delusions, disorganized speech, grossly disorganized or catatonic behavior, and negative symptoms. Over the past 20 years beginning with the DSM-III (APA, 1980), the DSM system has become widely accepted in the U.S., forming the core of research, clinical assessment, diagnosis, Theory-Based Clinical Representations, 5 and treatment in psychopathology. Research funding, journal titles, and health care reimbursements are all organized by, and dependent on, use of the categories defined by the DSM-IV (APA, 1994). Clearly, use of the categories laid out by the DSM-IV (APA, 1994) is widespread. The question for the current paper, then, is what kind of clinical reasoning emerges under use of the DSM. Do clinicians actually adhere to these guidelines outside formal diagnosis situations? If so, we would expect clinicians to give equal credence to all symptoms of a disorder. On the other hand, the theory-based approach argues that people treat features that are central to their domain theories as central in categorization as well. For instance, “being round” is treated as more central in categorizing basketballs than in categorizing cantaloupes because roundness is central in the naive physics underlying basketball concepts, but not central in the naive biology underlying cantaloupe concepts. Thus, clinicians reasoning in a theory-based manner might weigh symptoms differently depending on the theories they hold about the disorder. In the remainder of this introduction, previous work on clinical decision making from the perspective of categorization research will first be outlined. Then, it will be suggested that clinicians' representations of mental disorders include notions of how the symptoms affect one another, and that these representations can account for the relative importance clinicians assign to different symptoms in diagnosis and clinical reasoning. The possibility of expertise effects will also be explored, followed by a brief overview of the current experiments. Categorization in clinical decision making research The categorization literature has undergone several major shifts over the last few decades. The classical rule-based approach was followed by similarity-based models, which include prototype models (e.g., Rosch & Mervis, 1975; Posner & Keele, 1968) and exemplar Theory-Based Clinical Representations, 6 models (e.g., Medin & Schaffer, 1978). The theory-based view (e.g., Carey, 1985; Keil, 1989; Medin, 1989; Murphy & Medin, 1985) was also added to account for the role of explanation in categorization. This section discusses how these different approaches have been applied to mental disorder representations, as well as how these representations have been thought to be used in diagnostic reasoning. It is certainly not the intention here to suggest that only one of these views must be correct. Indeed, some, and perhaps all, of these approaches may account for some aspect of categorization, and are furthermore likely to interact in some way with each other (Keil, Smith, Simons, & Levin, 1998; Wisniewski & Medin, 1994). Rule-based approach. The rule-based view of categorization postulated that each category has individually necessary, and collectively sufficient, defining features (Medin, 1989; Smith & Medin, 1981). For example, the category "bachelor" may be defined by the conjunction of the features "adult,” “male," and "unmarried." The earliest known classification systems of mental disorders, espoused by Kraepelin (1913) and later by the DSM-I (APA, 1958) and DSM-II (APA, 1968), also required necessary and sufficient features for diagnosis. For instance, Kraepelin (1893) postulated that the defining feature of schizophrenia was an early-life onset of dementia (Hill, 1983). However, it proved difficult or impossible to come up with satisfactory defining features for most natural categories (Wittgenstein, 1953). For instance, “priest” fits the defining features for “bachelor,” but few people would actually refer to a priest as a bachelor. In addition, the classical approach cannot account for typicality effects in which some exemplars are rated as better members of a category than others. Such effects would not occur if categories were truly represented as defining features. Similarity-to-prototype approach. One alternative to the rule-based approach is the Theory-Based Clinical Representations, 7 prototype view (e.g., Rosch, 1978; Rosch & Mervis, 1975), which states that a category is represented as a prototype, an averaged, abstract representation of category members. Category membership is determined by an instance's similarity to that prototype. This approach can solve problems with the classical view. For instance, it can account for typicality effects in that the more similar the instance is to the prototype, the more likely it is to be categorized quickly, rated as more typical of the category, and so on. Similarly, the DSM-III (APA, 1980) task force adopted a format more like the prototype approach. That format was retained in the DSM-III-R (APA, 1988) and current DSM-IV (APA, 1994). This prototype-based nosology allows for more flexibility than did previous versions of the manual (Barlow & Durand, 1999). For example, the prototypical patient with Schizophrenia has five symptoms, but a presenting patient need only have two of those five symptoms for a diagnosis of the disorder. The validity of the prototype approach as the model of mental disorder diagnosis gained ground with additional studies. Cantor and her colleagues (Cantor, Smith, French, & Mezzich, 1980; Genero & Cantor, 1987), for instance, found that clinicians do make graded typicality ratings of patients, such that some are considered to be better examples of a disorder than others. They also found that clinicians were less accurate and confident when diagnosing atypical patients than when they were diagnosing patients highly or moderately typical of a disorder. (See also Clarkin, Widiger, Frances, Hurt, & Gilmore, 1983; Horowitz, Post, French, Wallis, & Siegelman, 1981; Horowitz, Wright, Lowenstein, & Parad, 1981; Russell, 1991; Widiger, 1982). Theory-based approach. The theory-based approach posits that the human mind forms categories and concepts based on its theories about the world (Carey, 1985). The current study investigates what role theory-based reasoning might also play in diagnosis by asking whether Theory-Based Clinical Representations, 8 clinicians have internalized the atheoretical reasoning of the DSM system. One possibility is that experienced clinicians, after many years of following the prescribed DSM system (Spitzer et al., 1994), reason about mental disorder categories without being affected by their theories about how the symptoms affect each other. The alternative is that clinicians, despite being given atheoretical guidelines for diagnosis, are still influenced by their own idiosyncratic theories about disorders when reasoning about them. Indeed, Medin (1989) has suggested that the DSM system "provides only a skeletal outline that is brought to life by theories and causal scenarios underlying and intertwined with the symptoms that comprise the diagnostic criteria (p. 1479).” The current research differentiates the two possibilities by examining two specific issues. First, we examined whether clinicians represent mental disorders as a list of independent symptoms, as represented in DSM-IV (APA, 1994) guidelines, or as a rich structure of symptoms that are highly interrelated, as assumed by the theory-based approach. Second, we examined whether clinicians give equal weights to symptoms as prescribed by the DSM system, or alternatively, whether symptoms central to clinicians’ theories about mental disorders determine their weights, as claimed by the theory-based approach. The next section describes two known specific mechanisms by which domain theories determine feature weighting. Ways in which theories determine feature weighting In this paper, the focus will be on the internal structure of concepts (i.e., Murphy & Medin, 1985), or how features are structured like theories, rather than on how concepts are connected to each other in domain theories. One specific mechanism by which the internal structure of concepts affects reasoning is the causal status effect (Ahn, 1998; Ahn et al., 2000). The causal status effect occurs when features causally central to an individual’s theory of that category are treated as more important in categorization than less causally central features. For Theory-Based Clinical Representations, 9 instance, if symptom A causes symptom B in a clinician’s theory, then A is more causally central than B, and A is thereby predicted to have greater diagnostic importance than B. This effect has been shown in lay people with both DSM-IV (APA, 1994) disorders and artificial mental disorders (Kim & Ahn, 2002). To derive the causal centralities of individual symptoms embedded in a complex theory, the following formula can be used: ci,t+1 = Sj dij cj,t (1) where dij is a positive number that represents how strongly symptom j depends on symptom i, and cj,t is the conceptual centrality of feature j, at time t (Sloman, Love, & Ahn, 1998). This model states that the centrality of feature i is determined at each time step by summing across the centrality of every other feature multiplied by that feature's degree of dependence upon feature i. Thus, in the current studies, the theory-based view was operationalized as a systematic effect of relational structures on conceptual representation and use. Another way in which theories influence feature weighting is that features relationally connected to other features are treated as more important than isolated features in reasoning (Kim & Ahn, 2002). Gentner’s (1983, 1989) structure-mapping theory, for example, argues that relational features (statements taking at least two arguments; for instance, x is smaller than y) are more important than attributes (statements taking only one argument; for instance, x is blue) in analogical inference. For instance, attributes such as “yellow,” “hot,” and “massive” are not particularly useful in making the analogy “an atom is like the solar system.” In contrast, relational features such as “more massive than,” and “revolves around” can be used to draw the analogy that electrons revolve around the nucleus in an atom as planets revolve around the sun in a solar system. (See also Lassaline, 1996.) Theory-Based Clinical Representations, 10 To summarize, the first hypothesis is that clinicians will show a causal status effect when reasoning about symptoms, such that symptoms central to their theory of a disorder will be treated as more important in diagnosis. The second hypothesis is that symptoms causally related to one another in a clinician's theory of a disorder will be treated as more important than symptoms not thought to be part of the causal theory. Putting these two hypotheses together, we predicted that if symptom A causes symptom B, but symptom C is isolated (it does not and is not caused by any other symptoms in a clinician’s theory), C would be the least central symptom of the three. The presence of these effects will be taken as evidence that domain theories influence clinical reasoning. Alternative explanations of such effects will be considered in the General Discussion. It must be noted that the ecological validity achieved by using real-life mental disorders comes at the price of not being able to control for all other factors (see General Discussion). In our previous work with lay people, we were able to create artificial mental disorders and manipulate which symptoms were causally connected. In the present study, however, we were more interested in how clinicians deal with real disorders than in how they might learn to deal with hypothetical disorders for which they are fed the causal explanations (which, as we have pointed out, the DSM does not do). Thus, the studies reported here must be read with that caveat. Expertise and the use of theories in diagnosis We also examined whether or not experts and trainees both adhere to the same strategy of mental disorder diagnosis. One alternative is that experts are more likely to show theory-based reasoning than trainees because experts have developed more theories after decades of clinical work. A second alternative is that whereas trainees may behave like lay people, making theorybased diagnoses (Kim & Ahn, 2002), experts might diagnose atheoretically because of years of Theory-Based Clinical Representations, 11 experience with the atheoretical DSM system. The third possibility is that both trainees and experts use their theories to weight features in diagnosis because theory-based reasoning is too cognitively compelling to be diminished even with a widely accepted set of atheoretical guidelines. Finally, it is possible that neither group is affected by theories in diagnosis, adhering strictly to DSM-IV (APA, 1994) guidelines. These last two alternatives, albeit for a different aspect of clinical reasoning, would be consistent with previous findings that experts and novices differ little in the treatment outcome aspect of clinical work (i.e., Durlak, 1979; Faust & Zlotnick, 1995), an issue that we will return to in the General Discussion. Overview of experiments Three major experiments (Experiments 1, 2, and 4) are reported in this paper. For purposes of comparison across studies, Figure 1 summarizes the methods and critical results for these experiments. In each of the experiments, the general methodology involves measuring each individual’s theories (Task I in Figure 1), and seeing whether these theories can predict which symptoms are treated as more central in that individual's representation of mental disorders. Two sets of centrality measures were taken in each experiment: (1) importance of a symptom when diagnosing or thinking of a mental disorder (Tasks II and III in Figure 1), and (2) memory for patients' symptoms (Task IV in Figure 1). As described earlier, the main prediction is that a symptom on which another symptom depends in an individual’s theory is treated as more important in diagnosis and is more likely to be remembered than dependent and isolated symptoms. Experiment 1: Use of Causal Theories in Clinical Reasoning Our question for this first experiment, then, was whether clinical psychologists conceptualize familiar mental disorders as the atheoretical, unweighted lists of the DSM-IV Theory-Based Clinical Representations, 12 (APA, 1994), or whether they represent them as theories that affect their diagnostic reasoning. The focus was on explanatory, or causal, relations because these have been pegged as critical to the background knowledge used in categorization (Ahn, Marsh, Luhmann, & Lee, 2002; Carey, 1985; Wellman, 1990). In the current experiment, we operationalized clinicians' and clinical trainees' use of theories in diagnosis with three different measures. First, participants were asked to rate the likelihood that hypothetical patients actually have the associated disorder. These "patients" had only either causally central, causally peripheral, or isolated symptoms, created according to each participant's theory of the disorder. Each type of hypothetical patient was given the same number of diagnostic criteria according to the DSM-IV (APA, 1994). Therefore, if participants do follow the guidelines of the DSM-IV (APA, 1994), then no systematic differences in diagnosis likelihood ratings would be expected. However, if the causal status model accurately reflects how they use theory in diagnosis, patients with causally central symptoms would be expected to have a higher likelihood of diagnosis than patients with causally peripheral symptoms. Furthermore, if the elevated importance for relational versus isolated features in analogical reasoning extends to diagnosis, then patients with causally central or peripheral symptoms would be expected to have a higher likelihood of diagnosis than patients with isolated symptoms. Second, participants were asked to recall the symptoms of those hypothetical patients. If clinicians are biased to attend to symptoms causally central to their theories about the disorder, then they would be expected to recall more causally central symptoms than causally peripheral or isolated symptoms. Alternatively, if they have internalized the DSM's guidelines for diagnosis, then they would be expected to recall equal proportions of all types of symptoms. Finally, we measured the conceptual centrality of each symptom and examined whether this measure Theory-Based Clinical Representations, 13 correlated with causal centrality in each clinician’s theory.