Rationality or irrationality of preferences? Quantitative tests of decision theories

Quantitative Testing, Decision Making, Hypothesis Testing, Bayes Factor, Cumulative Prospect Theory, Transitivity, Order-Constrained Statistics

Guo, Ying

关键词: Quantitative Testing, Decision Making, Hypothesis Testing, Bayes Factor, Cumulative Prospect Theory, Transitivity, Order-Constrained Statistics;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/102478/GUO-DISSERTATION-2018.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

To have transitive preferences, for any options x, y, and z, one who prefers x to y and y to z must prefer x to z. Transitivity of preferences is a very fundamental element of utility and plays an important role in many major contemporary theories of decision making under risk or uncertainty. One has to be very careful about claiming violations of transitivity of preferences. In my thesis, I present a comprehensive analysis of several decision heuristics that permit intransitive preferences: the lexicographic semiorder model (Tversky, 1969), the similarity model (Rubinstein, 1988), and perceived relative argument model (PRAM, Loomes, 2010a), as well as several transitive decision theories: the linear order model and 49 versions of Cumulative Prospect Theory (CPT, Tversky and Kahneman, 1992a). For each decision theory, I use two kinds of probabilistic specifications to explain choice variability: a distance-based probabilistic specification models preferences as deterministic and response processes as probabilistic, and a mixture specification models preferences as probabilistic and response processes as deterministic. I test these probabilistic models on data sets from different experiments, using both frequentist (Davis-Stober, 2009, Iverson and Falmagne, 1985, Silvapulle and Sen, 2005) and Bayesian (Myung et al., 2005) order-constrained, likelihood-based statistical inference methods. This thesis is one of the largest scale projects for a systematic evaluation of both transitive and intransitive decision theories. The quantitative analyses in this paper consumed about 822,000 CPU hours on Pittsburgh Supercomputer Center’s Blacklight, Greenfield, and Bridges supercomputers, as an Extreme Science and Engineering Discovery Environment project (see also, Towns et al., 2014). Individual model selection using Bayes factors shown extensive heterogeneity across participants and stimulus sets. In general, the overall conclusion is that Cumulative Prospective Theory and Perceived Relative Argument Model was systematically violated, and the intransitive heuristics performed reasonable well.