When individuals are required to decide among many options, they must choose whether to examine all of those options carefully or to pick one that seems good and stick with it (i.e., explore-exploit dilemma). A particularly interesting version of this dilemma is in the context of vanishing options (i.e., options disappear if not selected). Diminishing options may uniquely interact with how people learn about reward and punishment to bias strategies in ways that are sub-optimal. Such an effect may be especially exaggerated among those with anxiety and/or depression as they are particularly susceptible to changes in reward/punishment and are known to exhibit loss aversion.
Research Questions / Hypotheses
This study aimed to investigate how symptoms of anxiety and depression relates to decision-making in the context of explore-exploit dilemmas with vanishing options. It was expected that people would initially exhibit exploration and shift towards exploitation as the task progressed. How variations in reinforcement schedule (fixed vs. changing) influence learning about the reward and punishment was also investigated. It was hypothesised that people would learn faster under rapidly changing outcome probabilities vs. fixed outcomes. It was predicted that people would be more sensitive to punishment than reward if they were more anxious/neurotic. Finally whether individual differences in anxiety/depression/neuroticism or attitude to uncertainty predicted the choices people make in the above tasks was investigated. It was expected that more anxious/neurotic people would be less likely to shift to exploitation; that anxious individuals would show decreased benefits from the volatility component and that depressed individuals would be less sensitive to reward overall and more sensitive to punishment.
160 REP participants completed the study in semester 2. Participants were required to meet the following conditions: 18-45 years of age – Have no history of severe mental illnesses (e.g., bipolar disorder, schizophrenia, psychosis, or severe substance use problems). Individuals with a current or past diagnosis of anxiety and depression (or other high prevalence conditions) are potentially eligible to participate. – Have no major neurological conditions or serious, chronic medical conditions
The study was conducted online using Qualtrics survey software for the presentation of questionnaires and basic instructions. A plain language statement and consent form were presented. Once consent was obtained, demographic questions pertaining to age, level of education and mental health history were answered. Task instructions for the Vanishing Bandit Task (VBT) were then presented and participants were given a 5-item multiple-choice comprehension test. Using a random number generator, successful participants were then allocated one of two versions of the VBT. Participants were directed via a link to the task, written in JsPsych and hosted on a secure, commercial web server. The participants were first presented two practice blocks of 20 trials each, before completing two runs of three blocks (stable, slow-changing and fast-changing), with the opportunity for a break between each. After completing the VBT participants received a unique identifier code and were asked to return to Qualtrics and enter the code to ensure their two datasets could be matched. They then answered questions about the strategies they used and gave feedback regarding the task and completed personality questionnaires. The study took approximately one hour to complete.
Results are yet to be analysed. The primary outcomes of interest were proportion of “good” choices and average viable options (cf. Navarro et al., 2018). A choice is considered to be “good” if it was one of the two options in that trial with the highest expected reward, including options that had expired. The proportion of “good” choices will be calculated as the percentage of times a “good” choice was selected. The proportion of “good” choices as a function of condition (i.e., stable, slow-change, fast-change) and across the duration of blocks to determine changes across time will be examined. Average viable options measure the number of options retained (i.e., prevented from expiring) and will be calculated within conditions and across the duration of blocks. Interactions between task performance and the questionnaires will be tested through a repeated-measures analysis of covariance (ANCOVA). Where significant interactions are observed, median splits will be implemented to provide a graphical interpretation of the interactions. For example, all participants who scored above the median on the STICSA (anxiety measure) make up the ‘high anxiety’ group while those below and including the median will be the ‘low anxiety’ group. Computational modelling of the data is also planned.
Once computational modelling is complete, it is anticipated that the findings of the study could contribute to more efficient diagnosis of mental health disorders. This may then lead to better treatment as per the Research Domain Criteria (RDoC) The study will be submitted for publication as a journal article where the REP data set will be compared with another data set consisting of Amazon Mechanical Turk workers.