What Causes the Description-Experience Gap?
Several causes may be contributing to the description-experience gap.
Two classes of factors have been identified as shaping the search process in the sampling paradigm: properties of the decision-making problems (e.g., the magnitude of the incentives, see Hau, Pleskac, Kiefer, & Hertwig, 2008; and whether the outcomes are gains or losses, see Lejarraga, Hertwig, & Gonzalez, 2012) and individual characteristics, such as people’s emotional state (Frey, Hertwig, & Rieskamp, 2014) or age (Frey et al. 2015) . However, across numerous studies (reviewed in Hau et al., 2010) , respondents typically proved restrained in their information search, with a median number of samples per choice problem typically ranging between 11 and 19. These results suggest that reliance on small samples is one factor that contributes to the attenuated impact of rare events (Hertwig et al., 2004). For small samples the chances are that a person does not even experience the rare events. More generally, one is more likely to undersample than oversample the rare event, for the binomial distribution of the number of times a particular outcome will be observed in n independent trials is markedly skewed when p is small (i.e., the event is rare) and n is small (i.e., few outcomes are sampled). Interestingly, reliance on small samples has also been discussed as a potential explanation for bumblebees’ underweighting of rare events: Studying foraging decisions by bees in a spatial arrangement of flowers that promise with varying probabilities different amounts of nectar, Real (1991) concluded that “bumblebees underperceive rare events and overperceive common events” (p. 985). He explained this distortion in bees’ probability perception as a consequence of their sampling behavior—“bees frame their decisions on the basis of only a few visits” (Real, 1992, p. 133)—and suggested that such reliance on small samples can be adaptive.
Short-term optimization may be adaptive when there is a high degree of spatial autocorrelation in the distribution of floral rewards. In most field situations, there is intense local competition among pollinators for floral resources. When “hot” and “cold” spots in fields of flowers are created through pollinator activity, then such activity will generate a high degree of spatial autocorrelation in nectar rewards. If information about individual flowers is pooled, then the spatial structure of reward distributions will be lost, and foraging over the entire field will be less efficient. In spatially autocorrelated environments (“rugged landscapes”), averaging obscures the true nature of the environment. (p. 135)
Could there be any advantage to frugal sampling in experience-based decisions by humans? Hertwig and Pleskac (2008, 2010) proposed one possible advantage that rests on the notion of amplification. Unlike Real (1992), however, they argued that amplification proffers a cognitive rather than an evolutionary benefit. Through mathematical analysis and computer simulation, Hertwig and Pleskac (2010) showed that small samples amplify the difference between the options’ average rewards. That is, drawing small samples from payoff distributions results in experienced differences of sample means that are larger than the objective difference. Such amplified absolute differences simplify the choice between gambles and thereby explain the frugal sampling behavior observed in investigations of decisions from experience—a conjecture for which Hertwig and Pleskac (2010) found empirical evidence.
The explanation of the description-experience gap in terms of small samples has prompted a critical response (Fox & Hadar, 2006) and has led to an ongoing debate. What appears to be underweighting of rare events in decisions from experience could be consistent with overweighting of low probabilities as assumed in cumulative prospect theory. When the probability experienced in a sample is smaller than the event’s objective probability, people may still overweight this sample probability.
Despite this overweighting, the erroneous impression of underweighting would emerge if the overweighting did not fully compensate for the underestimation that results from the skew in small samples. In this view the description-experience gap is statistical (sampling error) rather than psychological in nature.
Several approaches have been taken to examine whether the gap observed in the sampling paradigm can indeed be reduced to sampling error. If sampling error was the sole culprit, then reducing the error by extending the sample should attenuate and eventually eliminate the gap. Increasing sample sizes substantially (up to 50 and 100 draws per choice problem) reduced but did not eliminate the gap (Hau et al., 2008, 2010). If sampling error caused the gap, then removing the error by aligning the sample’s experienced probabilities to the objective probabilities should eliminate it. It did not (Ungemach, Chater, & Stewart, 2009). If sampling error was the sole root of the gap, then presenting respondents in the description condition the same information that others experienced (yoking) should eliminate the gap. In one study it did (Rakow, Demes, & Newell, 2008); in another it did for small samples but not for large ones (Hau et al., 2010 ; see these authors’ discussion of trivial choices as one possible explanation for the mixed results obtained). The gap persisted even when people were presented both descriptions and experience rather than descriptions only (Jessup, Bishara, & Busemeyer, 2008).
In summary, the reality of the description-experience gap across the three experiential paradigms is unchallenged—its cause, however, is disputed. Some researchers have argued that the gap in the sampling paradigm is statistical in nature (Fox & Hadar, 2006; Hadar & Fox, 2009; Rakow et al., 2008); others have proposed that the sampling error is not the sole cause (Hau et al., 2008, 2010; Hertwig et al., 2004; Ungemach et al., 2009). Regardless of how this debate will advance, it is informative to go beyond the sampling paradigm. Reliance on small samples, for example, cannot be the reason behind the description-experience gap in the full-feedback paradigm (Fig. 8.1d) paradigm, in which the impact of rare events is attenuated even after a hundred trials with perfect feedback. Beyond sampling error, what psychological factors may be in play?