Experiments on confidence calibration and decision making

This thesis reports on three experiments studying subjects' confidence about performance on a task and how it relates to decision-making under uncertainty. Chapter 1 introduces the thesis providing an overview of the common themes and methods underlying this research. Chapter 2 reports the first experiment, investigating the relationship between risk attitudes and confidence judgements. We measure confidence in two different ways, with an incentivized elicitation tool and with unincentivized self-reports. Using our incentivised tool we find that, in the absence of controls for risk attitudes, subjects tend to be underconfident about their own performance. When we filter out the effects of risk attitudes we find that underconfidence is reduced, but not eliminated. We also identify an interesting link between self-reported confidence and risk attitudes in that experimental subjects with less concave utility functions and more elevated probability weighting functions tend to report higher confidence levels. Chapter 3 reports the second experiment, investigating the role of information in experimental market entry games. We look at whether individual over-entry to simple and under-entry to difficult markets disappears when subjects make entry decisions in groups or are given statistical information about performance of previous subjects. We find that individuals and groups are both susceptible to the same type of biases in entry and both fail to learn from repetition and feedback. We find that individuals learn to de-bias their entry decisions in the second half of the experiment when given explicit information about the performance of others. Chapter 4 reports an experiment investigating "snowballing of confidence" in hierarchical tournaments. We analyse how high/low scorers of a group in one stage of the tournament change their confidence levels in the next stage when they are re-grouped with other high/low scorers. We find that all subjects start the tournament assigning an equal chance to being high or low scorers in their groups. As they proceed through the stages, low scorers become more underconfident whereas high scorers become more overconfident about their relative performances. We also identify an interesting difference in the perceptions of the task between high and low scorers that is linked to self-serving causal attribution biases previously found in the psychology literature. Chapter 5 summarizes the findings of this dissertation and concludes