Description-Dependence

Description-Dependence

Many examples of description-dependence arise throughout the behavioral sciences. For instance, failures in Piagetian conservation tasks (Flavell, 1963) can be seen as illustrations of description-dependence (e.g., the same amount of liquid appears greater when in a tall narrow container than a short wide container). As another example, the influence on memory of variations in the wording of seemingly equivalent questions (e.g., Loftus, 1975) can be viewed as a form of description-dependence.

Decision framing effects offer what is perhaps the closest parallel to description dependence in probability judgment. The same medical procedure can be described in terms of either its survival or its mortality rates, leading to systematic changes in decisions made by experienced physicians (McNeil, Pauker, Sox, & Tversky, 1982). The same object can be valued differently by those who possess it versus those who do not, even when the object is distributed arbitrarily (Kahneman, Knetsch, & Thaler, 1990). In general, the way a choice option is described or the perspective from which it is viewed influences the attractiveness of that option.

A key element of description-dependence is the notion that people typically accept and treat problems as they are given without actively transforming those problems to some standard canonical representation (cf. Tversky & Kahneman, 1986). In the case of risky choice, people do not typically transform an offered gamble into a distribution of final wealth states. Analogously, in the case of judgment under uncertainty, people do not typically map a hypothesis onto some invariant sample space.

According to this view, probability judgment operates directly on the hypotheses and is not mediated through some representation of “event” (in the set-theoretic sense of the word). Descriptive models that invoke a canonical representation, such as event, may be inappropriately generalizing an important aspect of a normative model to the prediction and explanation of behavior (cf. Gigerenzer, 1991). Indeed, the notion of an event in probability theory is a subtle and nonintuitive concept. Instructors of probability and statistics classes often find that the fundamentals of probability theory are among the most difficult topics for their students. For example, recognizing that the hypothesis “A baseball batter will get at least one hit in four at-bats” is equivalent to “A baseball batter will not go hitless in each of four at-bats” is a difficult skill to acquire. Although a person may recognize the equivalence of two descriptions when presented with both and prompted to compare them, it is unlikely that when presented with one description the person will spontaneously recognize its equivalence to the many other possible descriptions and maintain consistency of judgment across those descriptions. The abstraction of events in probability theory is valuable precisely because people do not naturally engage in such abstraction themselves.

Demonstrations of description-dependence do not merely imply that people can be misled from their “true” beliefs, attitudes, or preferences. When there is not a single canonical description of some object, we cannot assume a single corresponding true belief, attitude, or preference toward that object. Descriptive theories of belief, attitude, or preference thus need to address the variability in responses due to changes in description, without simply appealing to the attractive but perhaps simplistic notion of an underlying canonical representation or “true score” (cf. Brenner, 2000).

We can distinguish between negative and positive analyses of description-dependence (Kahneman & Tversky, 1982). A negative analysis focuses on the failure to maintain invariant judgments across differences in description that (in a normative sense) should not matter. In contrast, a positive analysis provides some account (e.g., a causal explanation, processing model, or mathematical representation) that can be used to predict the occurrence, direction, and magnitude of description dependence. A positive analysis of conservation failures, for example, might posit that children focus primarily on the height of a container in judging its volume.

Support theory offers a partial positive analysis of description dependence in probability judgment: unpacking a hypothesis into components yields larger probability judgments, as does separate evaluation of the component hypotheses. However, the unpacking principle emphasized in the original formulation of support theory is just one of many possible forms of description-dependence. Clearly, other features of hypotheses besides their level of detail also influence support. Identification of the various determinants of description-dependence constitutes an important agenda for future research.

Two general points are worth stressing. First, the emphasis on subadditivity in previous research led researchers to focus almost exclusively on coextensional hypotheses. It may be liberating to abandon this constraint and investigate variations in description even if they entail changes in the referent event. Doing so would expand the scope of the theory by modeling judgment as a function of any identifiable features of hypotheses – not necessarily holding constant the referent event, as in studies of unpacking and repacking effects.

Second, it is worth noting that the evidence on which judgments are made, as well as the hypotheses under evaluation, is subject to description-dependence. Just as two hypotheses that refer to the same event can produce different judgments, so too can different descriptions of equivalent evidence.

Two broad research questions arise from this discussion. First: What features of hypotheses influence support, and how? Second: What features of evidence influence support, and how? In the remainder of this section, we offer some examples of features of hypotheses and evidence that may systematically influence support.