Abstract
As studies of the neural circuits underlying choice expand to include more complicated behaviors, analysis of behaviors elicited in laboratory paradigms has grown increasingly difficult. Social behaviors present a particular challenge, since inter- and intra-individual variation are expected to play key roles. However, due to limitations on data collection, studies must often choose between pooling data across all subjects or using individual subjects' data in isolation. Hierarchical models mediate between these two extremes by modeling individual subjects as drawn from a population distribution, allowing the population at large to serve as prior information about individuals' behavior. Here, we apply this method to data collected across multiple experimental sessions from a set of rhesus macaques performing a social information valuation task. We show that, while the values of social images vary markedly between individuals and between experimental sessions for the same individual, individuals also differentially value particular categories of social images. Furthermore, we demonstrate covariance between values for image categories within individuals and find evidence suggesting that magnitudes of stimulus values tend to diminish over time.
Highlights
Over the last decade, neuroscientists have made increasing use of model-based analysis methods to capture the dynamics of neural signals, in studies of choice (Schultz et al, 1997; Montague et al, 2004; Daw et al, 2006; Kennerley et al, 2006; Behrens et al, 2007; Quilodran et al, 2008; Krajbich et al, 2009; Pearson et al, 2009)
We show that hierarchical models allow us to make valuable statements about individual differences in choice behavior, but to tame ill-behaved fits via partial pooling, leading to better-behaved models and more reliable characterizations of behavior
Subjects repeatedly chose between two options, a visual target resulting in juice delivery and a visual target resulting in juice delivery plus the display of a social image from one of four categories (Neutral, Female, Dominant Male, Subordinate Male)
Summary
Neuroscientists have made increasing use of model-based analysis methods to capture the dynamics of neural signals, in studies of choice (Schultz et al, 1997; Montague et al, 2004; Daw et al, 2006; Kennerley et al, 2006; Behrens et al, 2007; Quilodran et al, 2008; Krajbich et al, 2009; Pearson et al, 2009). Parameters derived from models fitted to subjects’ behavior are used as regressors in models of neural dynamics, and studies test the hypothesis that these inferred parameters are encoded in experimental measures such as neuronal firing rates, EEG, or the BOLD signal (Friston et al, 2003). Choice behavior in both humans and non-human animals has proven notoriously variable within and between experimental sessions, resulting in highly variable estimates of subjects’ individual model parameters. The correctness of correlations between neural measures and model-derived parameters depends crucially on obtaining accurate and robust estimates of the latter. Constraints in data collection have limited the ability of researchers to draw statistically robust conclusions, in experimental designs where the amount of data per subject is necessarily large, for instance, when a behavioral model must be fit to each subject’s data
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
