Reward-seeking is a cross-cultural hallmark of adolescence across typically developing youth (Duell et al., 2018). Receipt of reward results in strong engagement of brain regions, such as the striatum, involved in processing motivationally salient stimuli (Galván, 2010). Cues that reliably predict later rewards activate similar regions of the brain during this anticipation period (Silverman, Jedd, & Luciana, 2015). As youth develop, reward anticipation elicits even greater striatal activation than receipt of reward (Hoogendam, Kahn, Hillegers, van Buuren, & Vink, 2013). This shift of increased activation to anticipation relative to reward outcome suggests developmental refinement of reward-based learning. Cues that precede higher rewards compared to lower rewards result in even greater anticipatory activation of reward processing regions, indicating that the expected value of rewards plays an important role in reward learning (Cohen et al., 2010; Lahat, Benson, Pine, Fox, & Ernst, 2016).
In contrast, youth at-risk for or diagnosed with anxiety often exhibit risk-aversion (Charpentier, Aylward, Roiser, & Robinson, 2017) potentially given high levels of intolerance toward uncertain, or risky, situations (Osmanağaoğlu, Creswell, & Dodd, 2018) and error monitoring (Buzzell et al., 2017). These features of youth anxiety may make the anticipatory period particularly distressing and arousing. Indeed, relative to typically developing youth, anxious and at-risk youth display greater striatal engagement during reward anticipation (Guyer et al., 2014, 2012, 2006) which scales as a function of anxiety severity (Lahat et al., 2016), especially when rewards are earned based on task performance (Bar-Haim et al., 2009; Benson, Guyer, Nelson, Pine, & Ernst, 2015). In stark contrast to their typically developing counterparts, anxious and at-risk youth demonstrate heightened reward network activation during loss of reward relative to gain of reward (Helfinstein et al., 2011). This seemingly paradoxical finding may represent enhanced negative prediction error signaling in response to aversive outcomes (Helfinstein, Fox, & Pine, 2012), which is in line with the predisposition of anxious youth to exhibit heightened awareness of their task performance.
This study utilizes behavioral and fMRI data from a probabilistic reward learning task as examined in a large sample of children and early adolescents across a continuum of anxiety severity. Applying a simple reinforcement learning model (e.g., Rescorla-Wagner) to these task data can enable investigation of both expected value and prediction error, as well as the learning signals that shape them including α (alpha; learning rate) which approximates how behavior changes following feedback and β (beta; inverse temperate parameter) which approximates stochasticity in choice behavior. Parsing these components of reward learning can provide a targeted approach to understanding how anxiety relates to reward anticipation and receipt of reward separately. Further, both expected value and prediction error can be used as parametric regressors in fMRI analyses to examine how activation in reward networks scales as a function of these metrics. Concomitant brain development and changes in youth’s socio-emotional environment render the life stage of adolescence especially prime for reward-driven approach behaviors, risk-averse avoidance behaviors, and complex interactions across approach-avoidance systems (Baker & Galván, 2020). Applying reinforcement learning models to reward learning can thus further our understanding of the nuanced relationship between anxiety and risk-taking in adolescence.