Cramer et al.'s target article shows a fruitful application of a dynamic network perspective – or complex dynamic systems (CDS) perspective – to clinical psychology, in particular the comorbidity of depression and anxiety. This commentary addresses three issues that can place the article in a broader framework of CDS-oriented theory and research in developmental psychology and psychopathology.

A complex dynamic system (CDS) is as a collection of interconnected components – the network – which change one another's properties through the interconnections and out of which emerge collective properties, such as patterns or correlations (Van Geert Reference Van Geert and Meyers2009). These patterns correspond with macroscopic phenomena (e.g., depression, attention-deficit hyperactivity disorder [ADHD], developmental levels). We believe that CDS theory provides a coherent conceptual and methodological tool-kit for understanding virtually all interesting phenomena of emergence in the behavioral sciences.

Cramer et al.'s network of causal influences is based on actual phenomenal appearances – symptoms – and not on underlying entities, such as depression, that correspond with latent variables. Assigning the causality to the actual phenomena instead of underlying entities is typical of the dynamic systems approach (e.g., Thelen & Smith Reference Thelen and Smith1994; van der Maas et al. Reference Van der Maas, Dolan, Grasman, Wicherts, Huizenga and Raijmakers2006). Our own work provides examples of a CDS approach to long-term cognitive and language development, based on the interaction of components in a network of variables (e.g., Bassano & van Geert Reference Bassano and Van Geert2007; Fischer & Bidell Reference Fischer, Bidell, Lerner and Damon2006; Steenbeek & Van Geert Reference Steenbeek and Van Geert2007; Reference Steenbeek and Van Geert2008; Van Geert Reference Van Geert1991; Reference Van Geert1994; Reference Van Geert1998; Van Geert & Steenbeek Reference Van Geert and Steenbeek2005). In an ongoing project, we have formulated a dynamic work model for explaining science and technology talents in young children (see Van Geert & Steenbeek, submitted).

The first question we address concerns the chosen level of aggregation for defining the dynamic network. Cramer et al.'s network is a structure of intra-personal symptoms, leading to comorbidity of depression and generalized anxiety. Our own approach to problematic learning trajectories in children with ADHD and PDD-NOS (pervasive developmental disorder, not otherwise specified) is based on the concept of a distributed network; that is, variables distributed over a child, an educator, and a material context. We view disorders such as ADHD as attractor states, consisting of fuzzy and variable ensembles of symptoms, actions, emotions, contexts, actions of the educators, and so on. These components affect one another in that they call forth one another's appearance in particular contexts. The interactions between components result in the consolidation of a particular pattern – for instance, one that educators or psychologists identify as the underlying condition ADHD, which they see as the “deep” causal mechanism of the symptoms. Our view is highly comparable to Cramer et al.'s view on clinical symptoms as attractor states; that is, self-sustaining patterns of mutually attracting symptoms. Our main goal is to use the short-term dynamics of relationships between variables (e.g., what happens during a concrete child-educator interaction) to explain the long-term trajectories through the state space of variables, which leads to the kind of self-sustaining states diagnosed as ADHD and the like.

Our distributed network shows a nested structure, with the child, the educator, and the educational materials as the main nodes, each comprising lower-level networks of variables specific to each of the main nodes (for an example of such a nested dynamic system, see our work on dyadic play interaction in children; Steenbeek & Van Geert Reference Steenbeek and Van Geert2007; Reference Steenbeek and Van Geert2008). The distributed network can be described from the perspective of each of its main nodes. For instance, from the perspective of the child, the educator and the educational materials constitute what is commonly called the context of the child's actions. Such contexts are not to be treated as independent variables. They are dynamically co-determined as part of the network (see Steenbeek & Van Geert Reference Steenbeek and Van Geert2007; Reference Steenbeek and Van Geert2008). They can change abruptly (e.g., if the child changes classes or activities), which may lead to relatively abrupt changes in the temporal attractor state (e.g., the actual expression of the child's ADHD), thus helping to explain the characteristic intrapersonal variability of a particular child's ADHD condition (see, e.g., Van Geert & van Dijk Reference Van Geert and van Dijk2002).

Cramer et al. emphasize the variations in connection strength between nodes, defining the nodes' degree of centrality, that is, importance in a particular network. Superficially similar constellations of symptoms under the umbrella term of ADHD, for instance, can be based on networks of different composition; that is, with different variables being more or less dynamically central, dependent on the particular individual in a particular context.

This brings us to our second issue, which is that of empirical design and methodology. In line with Cramer et al., we argue that networks emerge in the form of individual trajectories, and should therefore be studied with a time-serial case methodology. We study dynamic networks by directly observing the child's problematic learning in the class. We combine the observation of the short-term dynamics in a particular child – what happens during a single math lesson – with an account of the child's long-term dynamics, that is, changes in the patterning of the relationship between the components over the course of months or years (Steenbeek & Van Geert, submitted). This approach is ideographic, that is, it requires individual case studies. Generalization pertains primarily to the relationship between individual cases and the underlying CDS theory and only secondarily to sample generalization (see also Cramer et al's reference to Molenaar Reference Molenaar2004; cf. Van Geert, in press).

Our third issue concerns the clinical or educational application of complex dynamic network models. We believe that a practitioner's qualitative insight into a dynamic network explanation of clinical phenomena – for example, child psychopathologies – will lead to a different practice than one based on a theory of “deep” causes (e.g., the latent variable ADHD as a causal explanation of observable ADHD symptoms). An approach based on an understanding of the fuzzy and dynamic boundaries between disorder and normality may help practitioners to view their actions as part of an interconnected dynamic network, and may help them to redefine the kind of control problem they have to solve in order to help children overcome their clinical and behavioral problems.