A Materiality Standard for Prioritizing Information

Table 41.2. A Materiality Standard for Prioritizing Information

Carotid endarterectomy involves scraping out the artery leading to the brain, reducing the risk of stroke for patients with arteriosclerosis. However, having a sharp object in one’s neck can also cause problems, as can major surgery itself. Consider a population of patients, for all of whom the surgery would be a rational choice, if there were no risks (and money were no object). They vary in their physical condition, represented by probability distributions over possible consequences, and in their values, represented by distributions of tradeoffs across those consequences. Monte Carlo sampling produces a set of hypothetical patients, each characterized by a combination of physical states and personal values. Calculate the expected utility of the surgery decision for each such patient, ignoring all risks. Then recalculate it, including knowledge of the probability of each possible consequence. Information about a risk is material if it changes the expected utility of surgery from positive to negative.

In the reported simulation, about 15% of the hypothetical patients would prefer to decline surgery after including the probability of immediate death. Another 5% would decline after including iatrogenic stroke, and an additional 3% after including facial paralysis. These are proportions whose modeled choices change, not the probabilities of those complications. In this example, a small subset of the possible risk information affects many of the decisions.

Source: Merz et al., 1993.

The example assumes patients who know little, as might be the case with such a surgery. For patients with sharper (and potentially biased) priors, the analysis would consider the impact of each fact on updating them. Those priors could be elicited directly or derived theoretically. For example, patients relying on heuristics should underestimate risks that are either hard to imagine or not “representative” of the procedure (e.g., facial paralysis). Looking at potential messages from a cognitive perspective might also identify facts that are just too hard to communicate comprehensibly. Deemphasizing them saves time, while acknowledging that less overall understanding is possible. Priorities would also change if some facts proved redundant, in the sense that people who know one can predict others. With artificial tasks, any two features can occur together (e.g., the sizes and probabilities of outcomes in lab gambles), allowing no such efficiencies (although subjects may independently assume dependencies). However, people facing real-world prospects may already have many beliefs and quickly infer others.

The search for such cognitive connections has prompted the study of mental models in many domains (e.g., Bartlett, 1932; Furnham, 1988; Johnson-Laird, 1983; Tolman, 1937). A typical finding is that people organize diverse facts into highly flexible templates, in effect, substantive heuristics (e.g., how the circulatory system works, how the body processes toxins, how hazardous chemicals flow through groundwater, how risks mount up through repeated exposure). These studies typically contrast beliefs with a normative standard. For decisions regarding complex, loosely bounded problems, that standard can be captured in an influence diagram, showing the factors that determine the probabilities appearing in the associated decision tree (Clemen, 1991; Fischhoff et al., 1998; Howard, 1989). These diagrams show what is worth knowing about the problem, for individuals who otherwise know nothing. The contrast between them and lay beliefs shows what is worth saying, in order to make best use of individuals’ limited attention. Such contrasts have guided the development of communications on many topics, including radon, climate change, industrial chemicals, electromagnetic fields, mammography, HIV/AIDS, other sexually transmitted diseases, and nuclear energy in space (Bostrom et al., 1992; Fischhoff, 1998, 1999; Morgan et al., 2001).

Determining what is worth knowing and saying seems like the obvious communication strategy. Yet it is often hard to see its impact on many public health brochures, drug package inserts, home appliance instructions, or the like. Riley et al. (2001) showed that individuals who did everything listed on the labels of some brands of paint stripper could still face very high exposures. For other brands, users’ exposure depended on what they read (e.g., just first five items, the instructions, warnings, bolded items). Casman et al. (2000) argued that the outbreak warnings they studied for the waterborne parasite Cryptosporidium provided little useful information. Their concern was that confirmation could arrive after consumers had already been exposed. They proposed communication addressing ongoing exposure for immunocompromised individuals, rather than relying only on instructions issued after confirmation. Determining the content of communications and rendering it comprehensible requires collaborations between psychologists and subject-matter specialists.