Suboptimality and efficiency in human decision-making

Thesis

A good decision in isolation may be a bad decision in other conditions. Existing normative theories have been used to benchmark human performance. This is a compelling feature as they provide an understanding of how our decision unfolds in an orderly and rational way. These theories often assumed that decisions are made independent of contexts. However, decisions we have to make in the real-world depends on the context in which they occur. In the past few decades, psychologists and economists notably reveal that human decisions deviate from normative principles when the decisions are made in a context, and protest against one another about whether humans are suboptimal under different considerations of suboptimality. The goal of this thesis is to take these “suboptimal” decisions as a tool to understand what contributes to these suboptimalities. In this thesis, I refine these normative models to show that under reasonable assumptions, biases by decision context can be optimal; and (ii) describe a model embodying an efficiency principle that captures these biases by context across an unusually wide range of settings. In particular, I will focus on how a system that maximises computational efficiency with respect to behavioural needs and environmental statistics can explain: i) why do humans discard useful information during summary judgements, ii) take into account irrelevant information in perceptual judgement tasks, cognitive control, and value-based judgement tasks, and lastly iii) which efficient algorithm can approximate human performance in a trinary choice task.

Authors: Li, CL

• DOI: 10.5287/ora-44mee616e
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: suboptimality; decision making; efficient coding; gain control; adaptive gain
• Subjects: Psychology, Experimental; Cognitive neuroscience