In the target article Borsboom et al. assert that network models have the potential to highlight the important role of cultural and contextual variables in psychopathology, and that they allow for the modeling of such variables so that their roles can be properly elucidated. As the authors correctly note, these variables have been understudied, in part because of past emphasis on universalist, biological explanations in mainstream psychopathology research. A growing body of research suggests that cultural and contextual variables within individuals’ local social worlds play key roles in the development of psychopathology (see Hinton & Good Reference Hinton, Good, Good and Hinton2016; Kirmayer & Ryder Reference Kirmayer and Ryder2016).
As psychologists who conduct research on the impact of culture and context on psychopathology, we agree with Borsboom et al.’s critique that the use of latent variable models in cross-cultural research has supported a fruitless search for universal, biological origins of psychopathology (see Littlewood Reference Littlewood2002). However, we caution that this may not be an inevitable outcome of the latent variable approach per se. There are several instructive examples of studies in cultural-clinical psychology that use latent variable models to explore how cultural constructs of distress covary (e.g., Rasmussen et al. Reference Rasmussen, Katoni, Keller and Wilkinson2011) and test their construct validity (e.g., Chhim Reference Chhim2012). Furthermore, as both network analytic approaches and latent variable models are based on covariance, both may lead researchers to the discovery of similar patterns. Studies using network analysis to examine patterns of daily stressors, traumas, and symptoms (e.g., De Schryver et al. Reference De Schryver, Vindevogel, Rasmussen and Cramer2015; Jayawickreme et al. Reference Jayawickreme, Mootoo, Fountain, Rasmussen, Jayawickreme and Bertuccio2017) have reported similar results to studies examining similar variables but using a latent variable approach (e.g., Jordans et al. Reference Jordans, Semrau, Thornicroft and van Ommeren2012; Rasmussen et al. Reference Rasmussen, Nguyen, Wilkinson, Vundla, Raghavan, Miller and Keller2010). This similarity in findings makes even more sense if one conceptualizes latent variable models as a pragmatic way to specify distributions that summarize the associations among a set of variables (Skrondal & Rabe-Hesketh Reference Skrondal and Rabe-Hesketh2004), rather than purely as reflections of underlying variables.
Another area in which we urge caution concerns the sources of data that have been used in dynamic network analysis thus far. Most studies to date have used data from responses to pre-existing (and often well known) psychological measures. But the vast majority of psychological measures have been constructed based on the assumption that items should reflect an underlying latent variable. Thus, these measures, which are constructed using methods such as factor analysis or item response theory, consist of items that have been selected to correlate highly with each other. One of the touted strengths of network analysis is its ability to identify symptoms that are key, or central, to the disorder in question (Borsboom & Cramer Reference Borsboom and Cramer2013), yet it is unlikely that this will be accomplished if one is using a measure in which all of the items are developed by design to correlate with one another. Studies that use expanded symptom pools or other theoretically relevant variables will be more valuable.
Overall, we believe that network analytical approaches as currently used have yet to deliver on the promise of network models (Wright Reference Wright and Widiger2017). Further methodological work is needed to determine the degree to which network analysis is able to elucidate the role of cultural and contextual variables in the development of psychopathology beyond latent variable models. Furthermore, network researchers should avoid including individual items from measures constructed to identify latent variables (e.g., through the use of factor analysis) and should consider using the average scores or total score instead. It should be noted that a number of these limitations of current network analytical approaches have already been observed by at least one of the authors of the target article (i.e., Cramer in Fried & Cramer Reference Fried and Cramer2017).
That said, the history of psychology suggests that research methods and analytic strategies often shape research questions and even epistemologies. The authors’ critique of the latent variable model, especially to the degree to which it led to a fruitless (at least so far) attempt to discover the “underlying” biological substrate of mental disorder, is thus refreshing. Network models may lead more researchers towards an appreciation of the impact that cultural and contextual variables have on the development of person-level psychopathology. Such models may have the potential to move cultural-clinical research beyond a conceptualization of culture as a static, monolithic variable and towards a dynamic model in which multiple cultural, contextual as well as biological variables interact with one another (Morris et al. Reference Morris, Chiu and Liu2015). Recognition of these important causal factors also has implications for the reorganization of our current psychiatric nosological system. It has long been argued by cultural-clinical researchers that nosological systems such as the DSM are cultural products (e.g., Gone & Kirmayer Reference Gone, Kirmayer, Millon, Krueger and Simonsen2010; Ryder et al. Reference Ryder, Ban and Chentsova-Dutton2011) that, if rigidly applied, conceal culturally specific expressions of psychopathology (Kleinman Reference Kleinman1988). Network models have the potential to help us develop more cross-culturally valid diagnostic systems that are flexible enough to take into account cultural and contextual variables. Ryder et al. (Reference Ryder, Ban and Chentsova-Dutton2011) have proposed that, rather than be restricted by diagnostic classifications, cultural-clinical researchers should consider both “lumping” syndromes together (e.g., examine anxiety disorders broadly rather than focus specifically on panic disorder, generalized anxiety disorder) and focusing on cross-cultural variability in specific symptoms. Network models lend themselves to such a conceptual approach.
In the target article Borsboom et al. assert that network models have the potential to highlight the important role of cultural and contextual variables in psychopathology, and that they allow for the modeling of such variables so that their roles can be properly elucidated. As the authors correctly note, these variables have been understudied, in part because of past emphasis on universalist, biological explanations in mainstream psychopathology research. A growing body of research suggests that cultural and contextual variables within individuals’ local social worlds play key roles in the development of psychopathology (see Hinton & Good Reference Hinton, Good, Good and Hinton2016; Kirmayer & Ryder Reference Kirmayer and Ryder2016).
As psychologists who conduct research on the impact of culture and context on psychopathology, we agree with Borsboom et al.’s critique that the use of latent variable models in cross-cultural research has supported a fruitless search for universal, biological origins of psychopathology (see Littlewood Reference Littlewood2002). However, we caution that this may not be an inevitable outcome of the latent variable approach per se. There are several instructive examples of studies in cultural-clinical psychology that use latent variable models to explore how cultural constructs of distress covary (e.g., Rasmussen et al. Reference Rasmussen, Katoni, Keller and Wilkinson2011) and test their construct validity (e.g., Chhim Reference Chhim2012). Furthermore, as both network analytic approaches and latent variable models are based on covariance, both may lead researchers to the discovery of similar patterns. Studies using network analysis to examine patterns of daily stressors, traumas, and symptoms (e.g., De Schryver et al. Reference De Schryver, Vindevogel, Rasmussen and Cramer2015; Jayawickreme et al. Reference Jayawickreme, Mootoo, Fountain, Rasmussen, Jayawickreme and Bertuccio2017) have reported similar results to studies examining similar variables but using a latent variable approach (e.g., Jordans et al. Reference Jordans, Semrau, Thornicroft and van Ommeren2012; Rasmussen et al. Reference Rasmussen, Nguyen, Wilkinson, Vundla, Raghavan, Miller and Keller2010). This similarity in findings makes even more sense if one conceptualizes latent variable models as a pragmatic way to specify distributions that summarize the associations among a set of variables (Skrondal & Rabe-Hesketh Reference Skrondal and Rabe-Hesketh2004), rather than purely as reflections of underlying variables.
Another area in which we urge caution concerns the sources of data that have been used in dynamic network analysis thus far. Most studies to date have used data from responses to pre-existing (and often well known) psychological measures. But the vast majority of psychological measures have been constructed based on the assumption that items should reflect an underlying latent variable. Thus, these measures, which are constructed using methods such as factor analysis or item response theory, consist of items that have been selected to correlate highly with each other. One of the touted strengths of network analysis is its ability to identify symptoms that are key, or central, to the disorder in question (Borsboom & Cramer Reference Borsboom and Cramer2013), yet it is unlikely that this will be accomplished if one is using a measure in which all of the items are developed by design to correlate with one another. Studies that use expanded symptom pools or other theoretically relevant variables will be more valuable.
Overall, we believe that network analytical approaches as currently used have yet to deliver on the promise of network models (Wright Reference Wright and Widiger2017). Further methodological work is needed to determine the degree to which network analysis is able to elucidate the role of cultural and contextual variables in the development of psychopathology beyond latent variable models. Furthermore, network researchers should avoid including individual items from measures constructed to identify latent variables (e.g., through the use of factor analysis) and should consider using the average scores or total score instead. It should be noted that a number of these limitations of current network analytical approaches have already been observed by at least one of the authors of the target article (i.e., Cramer in Fried & Cramer Reference Fried and Cramer2017).
That said, the history of psychology suggests that research methods and analytic strategies often shape research questions and even epistemologies. The authors’ critique of the latent variable model, especially to the degree to which it led to a fruitless (at least so far) attempt to discover the “underlying” biological substrate of mental disorder, is thus refreshing. Network models may lead more researchers towards an appreciation of the impact that cultural and contextual variables have on the development of person-level psychopathology. Such models may have the potential to move cultural-clinical research beyond a conceptualization of culture as a static, monolithic variable and towards a dynamic model in which multiple cultural, contextual as well as biological variables interact with one another (Morris et al. Reference Morris, Chiu and Liu2015). Recognition of these important causal factors also has implications for the reorganization of our current psychiatric nosological system. It has long been argued by cultural-clinical researchers that nosological systems such as the DSM are cultural products (e.g., Gone & Kirmayer Reference Gone, Kirmayer, Millon, Krueger and Simonsen2010; Ryder et al. Reference Ryder, Ban and Chentsova-Dutton2011) that, if rigidly applied, conceal culturally specific expressions of psychopathology (Kleinman Reference Kleinman1988). Network models have the potential to help us develop more cross-culturally valid diagnostic systems that are flexible enough to take into account cultural and contextual variables. Ryder et al. (Reference Ryder, Ban and Chentsova-Dutton2011) have proposed that, rather than be restricted by diagnostic classifications, cultural-clinical researchers should consider both “lumping” syndromes together (e.g., examine anxiety disorders broadly rather than focus specifically on panic disorder, generalized anxiety disorder) and focusing on cross-cultural variability in specific symptoms. Network models lend themselves to such a conceptual approach.