Still another factor that could underlie the description-experience gap, especially in the feedback paradigm, is the notion that people inform their decisions by recruiting recent and past experiences garnered in similar situations (for related notions see Gilboa & Schmeidler, 1995; Gonzalez, Lerch, & Lebiere, 2003). Such contingent sampling is likely to be ubiquitous in the wild (Klein, 1999). For example, when firefighters need to predict the behavior of a fire, they appear to retrieve from memory similar instances from the past. Contingent sampling implies recency and reliance on small sampling to the extent that similarity decreases with time. Furthermore, in dynamic environments (e.g., the restless bandit problem; Whittle, 1988), reliance on similar experiences is an efficient heuristic (Biele, Erev, & Ert, 2009). Below, we turn to the manner in which the process of contingent sampling can be modeled.
Spatial Search Policies
Like any organism, humans can sample information in at least two very different ways from payoff distributions (e.g., flowers, ponds, other people, and gambles). Figure 8.3 depicts two paradigmatic sequential-sampling strategies based on two assumed options. In piecewise sampling, the searcher oscillates between options, each time drawing, in the most extreme case, the smallest possible sample. In comprehensive sampling, by contrast, the searcher samples extensively from one option and then turns to the other option to explore it thoroughly.
Taking these two sampling strategies as a starting point, Hills and Hertwig (2010) suggested that this spatial way of sampling foreshadows how people make their final decision. Specifically, they proposed that a person who samples piecewise will tend to make decisions as would a judge who scores each round of a boxing match: She determines which option yields the better reward in each round of sampling and ultimately picks the one that wins the most rounds. By contrast, a person using a comprehensive-sampling strategy will tend to gauge the average reward for each
Fig. 8.3 (a) Representations of the sampling patterns associated with piecewise- and comprehensive-sampling strategies. Piecewise strategies repeatedly alternate back and forth between options. Comprehensive-sampling strategies take one large sample from each option. Following the sample phase, the participants make a decision about which option they prefer. (b) Representations of the comparison strategies associated with roundwise and summary strategies for a set of hypothetical outcomes. Roundwise strategies compare outcomes over repeated rounds and choose options that win the most rounds. Summary strategies compare final values (here, the overall expected value) and choose options with the better final value (Reprinted from Hills and Hertwig (2010, p. 1788) with permission from Associations for Psychological Science)
option and then choose the one promising the larger reward harvest. The reason for this dependency of the decision strategies on search is that the piecewise- and comprehensive-sampling strategy foster comparisons across different scales of information: rounds vs. summaries, respectively. Determining a winner who is ahead in most rounds and determining the one yielding the largest expected reward can lead to different choices even when both decisions-makers experience the same information. The reason is that the person using the former decision strategy weighs each round equally, ignores the magnitude of wins and losses, and thus acts as if it underweights rare, but consequential, outcomes. That link between sampling strategy and decision strategy is exactly what Hills and Hertwig (2010) found. Individuals who frequently oscillated between options were more likely to choose the round- wise winning options and to make choices as if they underweighted rare events than were individuals who switched options rarely.
In summary, modern behavioral decision research has been strongly focused on people’s responses to descriptions of events. In recent years three experiential paradigms have been used to study how experience affects risky choice. A consistent picture has emerged. When rare events are involved, description-based and experience-based decisions can drastically diverge. We now turn to different ways of modeling decisions from experience.