2016 Bulletin

Bulletin vol. 30 no. 1 | 19 Diffusion Modeling in ADHD: A Brief Introduction and Application for Clinical Practice Attention Deficit Hyperactivity Disorder (ADHD) is a neurodevelopmental disorder that affects 3-7% of children in the United States, and is characterized by extreme and age- inappropriate difficulties with attention, hyperactivity, and impulsivity (American Psychiatric Association, 2013). In addition to the psychological costs that ADHD places upon children, parents, teachers, and communities, the estimated aggregate annual treatment costs for the disorder is $8 billion dollars across the heath/mental health sectors, and is close to $14 billion dollars for the educational sector (Pelham, Foster, & Robb, 2007). As a field, it is therefore critical that we not only continue to identify the underlying causes and cognitive mechanisms that drive the disorder, but also work to provide researchers and clinicians with the most valid methods by which those mechanisms can be measured. Ultimately, the goal is to move diagnosis of the illness away from reliance on behavioral symptomology and towards a mechanism-based nosology and functional-deficit approach to assessment. Of the putative etiologic mechanisms, deficits in executive function (EF) are perhaps the most well-known, and are implicated in 30-50% of children diagnosed with ADHD (Nigg, Willcutt, Doyle, & Sonuga-Barke, 2005). However, originally designed to detect frank brain injury, traditional neuropsychological tests of EF tend to tap multiple component processes. While no task is process-pure, clinical science's more recent adoption of computer-based, empirically-supported models and paradigms of cognitive control from the cognitive sciences (e.g. go-no-go, continuous performance tasks, working memory span tasks, flanker tasks, etc) has allowed researchers to more easily identify the neural networks and areas of localization involved in the disorder. That being said, currently, the vast majority of the most well-validated measures EF in both research and clinical practice use mean reaction time (RT) or mean accuracy as a primary outcome variable. This is problematic for two reasons. First, as two descriptors of a single response, RT and accuracy are produced simultaneously and are non-independent. However, the standard in the field is that either one or the other is selected for analysis, even if this leads to important differences in interpretation depending on the variable selected. For example, one may as a rule, tend to emphasize accuracy in responding over speed. But, if RT is used as the dependent variable for that particular paradigm, then such performance would be interpreted as impaired. Or vice versa. So, considering RT and accuracy separately provides at best an incomplete understanding of performance, and at worse, erroneous interpretations of data. Second, RT is multiply influenced: by the efficiency with which information is accumulated to make a decision, which is typically the construct of interest (e.g. "do I go or not go?" "is this a word or non-word?"), but also by the amount of time needed to encode a stimulus; to prepare and execute a motor response; and whether one tends Galloway-Long, H., BS, Shapiro, Z., MS, Huang-Pollock, C. Ph.D. Penn State University (or has been instructed) to emphasize speed over accuracy, or vice versa) (Figure 1). These associated processes ultimately reduce the sensitivity of RT measures to reliably detect individual differences on the core construct of interest. Although clinicians often "eyeball," or informally take differences in speed/accuracy into some consideration during interpretation, it is clear that an empirically-supported method is needed to reliably and rationally incorporate both error rate and RT into a single set of performance indicators. A well-known and well- validated computational method called diffusion modeling may present a solution (See White, Ratcliff, Vasey, & McKoon, 2010 for an excellent introduction and argument for this technique). Commonly used in cognitive research in college aged adults, diffusion modeling has also more recently been validated in developmental, aging, and clinical populations, most notably ADHD and anxiety disorders. Diffusion Modeling The diffusion model can be used for any tasks requiring simple two-choice decisions. There are three primary parameters, or variables, that describe performance. First, "drift rate" represents the rate at which an individual is able to make a decision. Here, larger is better/faster. Second, "boundary separation" refers to the relative amount of information a person requires prior to making a decision. Requiring more information (i.e. having wider boundaries) will result in slower but more accurate choices, while requiring less information produces faster, but more error-prone responses. And third, "non-decision time" represents the time it takes to encode a stimulus and the time it takes to prepare the motor response once a decision is reached. These variables are obtained by fitting the observed RT distributions for correct and error responses, to the RT distributions that are predicted by the model. Using an initial set of parameter values, predictions are derived and then the values of the parameters are adjusted using an automatic routine that computes the goodness of fit, and then replaces the worst fitting parameter values with new ones. By this method, the model ultimately produces the best fit to the data. The best fitting set of parameter values are then used as indices of the latent psychological processes that each parameter represents (i.e. cognitive processing efficiency, speed-accuracy trade off settings, and the time to encode/prepare a motor response). Diffusion Modeling in Research Because the diffusion model makes full use of multiple dimensions of the data—the shape of the RT distribution for correct and incorrect responses—rather than relying on just mean RT or mean accuracy, it is able to yield a more complete picture of performance. As such, it is able to go beyond telling us if someone is slow, to explaining what step in the flow of information processing is the cause of slower RTs. For example, in aging populations, the diffusion model has been used to demonstrate

• Cover