Normative models of perception, action and cognition share the assumption that humans base their decisions on accurate representations of probability and relative frequency. Experimental results indicate that they often do not. In decision under risk, for example, participants act as if their choices were based on probability values systematically different from those that are objectively correct. Similar systematic distortion is found in tasks involving relative frequency judgments. These distortions are dynamic: the same participant can have different distortions in different tasks. We propose a model of probability distortion as a form of bounded rationality (Simon, 1982). The "bound" in this Bounded Log-Odds Model (BLO) is the plausible assumption that the neural representation of probability has a fixed, limited dynamic range (Thurstone Case V). We used factorial model comparison to compare BLO to versions of BLO with each of its assumptions weakened on altered. BLO accounted for human performance better than any of the variant models. We also compared BLO to all previous models of decision under risk and found that it provided superior fits to existing data. We demonstrate that, subject to these BLO constraints, people maximize their information transmission or nearly so, a form of rational behavior constrained by immutable bounds.
- Laurence T Maloney, Psychology and Neural Science, New York University
- Hosted by
- School of Psychology
- University of Aberdeen
Dr Chu or Ms Carolyn Porter (01224 272227)