SUBJECTIVE INTEGRATION OF PROBABILISTIC INFORMATION FROM DESCRIPTION AND FROM EXPERIENCE

Abstract

Subjective integration of probabilistic information obtained via description and experience underlies potentially consequential judgments and choices. However, little is known about the quality of the integration and the underlying processes. I contribute to filling this gap by investigating judgments informed by integrating probabilistic information from the two sources. Building on existing information integration frameworks (e.g., N. Anderson, 1971), I develop and subsequently test computational models that represent the integration process. Participants in three experiments estimated the percentage of red balls in a bag containing red and blue balls based on two samples drawn from the bag. They experienced one sample by observing a sequence of draws and received a description of the other sample in terms of summary statistics. Subjective integration was more sensitive to information obtained via experience than via description in a manner that depended on the extremity of the experienced sample relative to the described one. Experiment 1 showed that experience preceding description leads to integration that is less biased towards experience than the reverse presentation sequence. Following this result, Experiment 2 examined the effect of memory-retrieval demands on the quality of the integration. Specifically, we manipulated the presence or absence of description- and experience- based decision aids that eliminate the need to retrieve source-specific information. The results show that the experience aid increased the bias, while the description aid had no interpretable effect. Experiment 3 investigated the effect of the numerical format of the description (percentage vs. frequency). When description was provided in the frequency format, the judgments were unbiased and the leading model suggested that the two sources are psychologically equivalent. However, when the description was provided in the percentage format, the leading model implied a tradeoff between the two sources. Finally, participants in Experiment 3 also rated how much they trusted the source of the description. The participants' ratings were correlated with how they used the description and with the quality of their judgments. The findings have implications for interpreting the description-experience gap in risky choice, for information integration models, and for understanding the role of format on the use of information from external sources. In addition, the methods developed here can be applied broadly to study how people integrate information from different sources or in different formats.