Overconfidence in Future Events: Cross-Cultural Variations

Future Events

As we argued previously, many of the decisions people make routinely rest on their beliefs about facts as they currently exist (e.g., decisions about cities to include in an itinerary predicated in part on assumptions about how expensive are accommodations in those cities right now). However, many other decisions are driven by people’s judgments about future occurrences (e.g., investment decisions contingent on anticipated sales of some product). Do the usual cross-cultural variations in overconfidence observed in judgments about currently knowable facts extend to judgments about the future?

Yates et al. (1989) addressed this question as it applied to quantity judgments. In one item, their Chinese participants were asked for distribution judgments about the low temperature the next day in the southern China city of Guangzhou. Their American counterparts were asked for the same kinds of judgments applying to the southern U.S. city of Miami. The overall 2% surprise index for the American respondents was 41.4%, which was greatly surpassed by the corresponding Chinese 2% surprise index of 53.2%. Zhang (1992) conducted a similar distribution judgment study with professional economists in Beijing. Their task was to make probabilistic distribution judgments of economic indexes for the Chinese economy. Their normal professional duties required them to make such forecasts routinely, although only in a deterministic, “single-value” way (i.e., making point forecasts equivalent to saying, “My best estimate is X,” termed striking numbers by the economists themselves). Unfortunately, Zhang did not have a comparison group of economists performing a similar task in some Western culture, so he used the results of Yates et al. (1989) as a point of reference. Zhang observed that the overconfidence of the Chinese economists was at least as great as that of the Chinese respondents in the Yates et al. study, and therefore much stronger than that of the Americans in that investigation.

Wright and Wisudha (1982) examined British and Indonesian judgments about future discrete events rather than quantified variables. An illustrative item (p. 221): “At least one national leader (president or prime minister, etc.) (a) will (b) will not die during the next 30 days.” The future-event judgments of the Indonesian respondents were far less overconfident than their judgments about currently knowable discrete facts (e.g., whether the Black Sea is larger than the Caspian Sea). Yet those judgments were still much more overconfident than the future-event judgments of the British respondents. The difference between these Indonesian and British samples extended the earlier findings to future-event judgments.

A significant feature of the future events used by Yates et al. (1989), Wright and Wisudha (1982), and, to some extent, those used by Zhang (1992) is their heterogeneity. Thus, in a quantity judgment study modeled after that of Yates et al., a given respondent could be asked to make judgments about temperatures, prices of goods and services, and average test scores. The very heterogeneity of the focal events limits our ability to learn much about the details of a person’s underlying judgment processes, and this constrains the potential for improving those processes. Considerations such as these are among those that motivate studies focusing on people’s judgments about homogeneous event classes. Another is that homogeneous events often have such great practical significance that acquiring accurate judgments about them is widely recognized as a priority. Therefore, the task of providing these judgments is often assigned to professionals who render them over and over again and are expected to develop great expertise at doing so. Weather forecasting provides familiar examples, such as forecasting precipitation or the kinds of severe storms that can create an airline tragedy in an instant. Medicine offers plenty of other illustrations (e.g., making cancer diagnoses). Will previously documented cross-cultural variations in overconfidence be observed in judgments about homogeneous event classes also?

Yates, Lee, Shinotsuka, Patalano, and Sieck (1998) pursued this issue in their studies within a simulated medical environment. Participants in Taiwan, Japan, and the United States each assumed the role of a physician who was confronted with two new (fictional) diseases in the community, Trebitis and Philiosis. On each trial, the “physician” must indicate whether it is more likely that the patient has Trebitis or Philiosis and then record a 50% to 100% probability judgment that that selection is indeed the patient’s true condition. After each trial, the “physician” is told of the actual illness. The “physician” makes each diagnosis on the basis of a profile of “symptoms” (technically, signs and symptoms in real-life medical parlance) displayed on a computer screen (e.g., the presence or absence of earache, nausea, and rash). The symptoms for a given patient represent a random sample from an underlying “artificial ecology” for the target diseases that, as in the real life being simulated, is initially completely unknown to the participant as physician. That ecology is comprised of variables corresponding to the symptoms as well as the diseases. It entails statistical associations between the symptoms on the one hand and Trebitis and Philiosis, on the other hand, ranging from moderately strong to none. At the start of the experiment, being ignorant of the ecology, the participant can achieve no more than chance-level diagnostic accuracy. However, after inducing the nature of the ecology through feedback, quite respectable diagnostic accuracy is attainable, in principle, at least.

In the first study, the computer automatically displayed to the “physician” all of the known symptoms for any given patient. In the second study, though, the participant had the discretion to display as many or as few of the cues as desired. However, uncovering cues “cost” the participant points against the accuracy bonus that the participant was eligible to receive. This distinction did not affect conclusions about overconfidence. In both studies, the same basic pattern emerged. Mimicking what had been found in earlier studies, the Chinese participants displayed substantially more overconfidence than the Americans or the Japanese. Furthermore, in each study, Japanese overconfidence was weaker than American overconfidence, but not significantly so in either instance. So, even for judgments about a homogeneous class of events under well-controlled conditions, we see the same, now-familiar pattern.

The “medical practice” for each participant in the Yates et al. (1998) studies extended for 120 patients, distributed over 2 days separated by 1 week. This feature of the design allowed for an assessment of learning effects. Bias, or overconfidence, was essentially unchanged from one week to the next for the Japanese and American participants. Importantly, however, overconfidence diminished significantly over that time for the Chinese in both studies. Focused opportunities to learn relative frequencies as well as contingencies, like those in the Yates et al. (1998) study, might be the foundations for one other result reported by Zhang (1992). Modeling his study after that of Keren (1987), Zhang had Chinese bridge-club members make probability judgments that final contracts would be made during bridge card games. Zhang found almost no difference between the calibration curve for the Chinese participants in his study and that for Keren’s Dutch bridge players. If anything, the Chinese players seemed to be slightly less extreme in their judgments than the Dutch. We return to the learning issues that might be implicated in findings like these when we discuss in earnest the “whys” of cross-cultural variations.