Evaluating cognitive sequential risk-taking models: Manipulations of the stochastic process

Thumbnail Image


umi-umd-1849.pdf (495.24 KB)
No. of downloads: 1381

Publication or External Link






This dissertation evaluates, refines, and extends to a new paradigm, a set of stochastic models that describe the cognitive processes of individuals while they complete multiple trials of the Balloon Analogue Risk Task (BART; Lejuez et al., 2002). Wallsten, Pleskac, and Lejuez (2004) designed the models using prospect theory and a Bayesian learning process to better understand why the BART correlates so well with self-reported risky behaviors. The models differed in terms of the individuals' beliefs of the task's probabilistic structure and when option evaluations occur. The models revealed that although respondents use a Bayesian learning process to understand the task, they misunderstand the BART's stochastic process as stationary. Results also indicated that individuals' attitudes toward outcomes are, in part, a source of the BART's success. From these conclusions a new task was developed that allows manipulations of both the actual stochastic structure and the individuals' level of knowledge regarding the structure. Participants (N = 71) completed four different conditions of the task. Fitting the various cognitive models to each individual's data revealed that only a subset of the models correctly distinguished between the stochastic processes underlying the different conditions. Incorporating prospect theory's weighting function and a trial-dependent bias component into the models accounted for performance differences between conditions. Of the assorted model parameters, only prospect theory's value function correlated with external self-reported risky drug use. The results also showed that the learning component of the original BART may cloud its association to risky behaviors. Implications in terms of gambling tasks and the cognitive models will be discussed.