Schedule - Parallel Session 1 - Risk Attitudes 1

Over the course of the last decades it turned out that risk preferences are only captured partially by the concept of risk aversion. Higher-order risk preferences like prudence (Kimball, 1990) and temperance (Kimball, 1992) also impact decisions made by individuals when facing uncertainty. Higher-order risk preferences have been studied in a small number of experiments so far (for an overview see Noussair et al., 2014). These studies reveal that a majority of people is not only risk-averse, but also prudent and temperate. In a recent laboratory experiment, Deck and Schlesinger (2014) “hereafter D&S “test the hy-pothesis that risky choices can be explained either by a lottery preference for combining “good” with “bad” or for combining “good” with “good” (see Crainich et al., 2013). The former implies mixed risk-averse behavior, the latter mixed risk-loving behavior. Both types differ in their pref-erences for lotteries of even orders but coincide for odd orders. In line with their hypothesis, D&S in fact observe a consistent pattern of behavior with US-American subjects: Risk averters tend to be temperate (order 4) and risk-apportionate of order 6 while risk lovers tend to choose in the opposite way. Furthermore, both types exhibit prudent (order 3) and edgy (order 5) behavior. We build on the analysis by D&S and explore the consistency of higher-order risk preferences with regards to (A) cross-country differences, (B) differences in stake size, and (C) differences through displaying reduced rather than compound lotteries. We use the elicitation method intro-duced by D&S and conducted a series of economic laboratory experiments in China, Germany, and the US with a total of 605 subjects. In short, subjects faced a total of 38 tasks in randomized order one of which was randomly selected for payment. In each task subjects had to choose be-tween two lotteries which revealed there risk preference up to the 6th order. In order to investigate the effects of the stake size, we increased the payoff by a factor of ten for some of the Chinese subjects. In order to investigate the effect of reducing compound lotteries on choices, we ran ad-ditional sessions in Germany where the participants faced order 1 and order 2 lotteries in original (compound) form plus two additional orders in compound as well as in reduced form. We replicate the finding of mixed risk-averse and mixed risk-loving behavior by D&S in the US and identify a similar pattern in Germany and in China. Moreover, we observe an increase in risk aversion when stakes are increased tenfold. We also observe the pattern of mixed risk-averse and mixed risk-loving behavior with high stakes. Finally, we observe that subjects choose the prudent and temperate options less often when they are displayed in a reduced rather than compound form. In the reduced lotteries there is weak evidence that subjects behave generally prudent and no evidence that they are generally temperate.

Communicating risk of uncertain outcomes (e.g., court sentence) often involves probabilities. People prefer to receive probabilistic information numerically (e.g., there is 70% chance that your case will not end in your favour), but they prefer to express it verbally (e.g., it is likely that your case will not end in your favour) (Erev & Cohen, 1990). Consequential decisions (e.g., should I take the plea bargain?) thus require an adequate interpretation of verbal probabilities. Some research suggests that severe outcomes trigger higher probability estimates (Harris & Corner, 2011) whereas other line of research suggests that higher base rates inflate probability (Fisher & Jungermann, 1996). A critical challenge in risk perception research is that severity and base rate are natural confounds. To draw any robust conclusions about the severity bias we need to test the effect of outcome severity while also controlling for its base rate.

In three experiments we tested how severity and base rate of a court sentence may influence the perceived probability of suspect’s conviction. In a 2(base rate: low vs. high) × 2(severity: low vs. high) between-subjects design, we manipulated base rate by describing the sentencing in similar cases as low or high and severity by describing the corresponding sentences as mild or severe. Participants then provided a numerical estimate for the lawyer’s statement: “it is possible that you will be convicted”.

The effect of severity on probability was not moderated by base rate in Experiment 1 (β3 = -.53, p = .94, 95% CI [-13.62, 12.57]) nor in Experiment 2 (β3 = -5.85, p = .25, 95% CI [-15.94, 4.24]), but it was moderated by base rate in Experiment 3 (β3 = -13.03, p = .03, 95% CI [-24.83, -1.24]). Conditional effects confirmed that the severity effect occurred at low levels of base rate (β = 10.66, p = .01, 95% CI [2.13, 19.19]). When meta-analysed, we found a significant severity effect in a low base rate condition (g = 0.29, p = .03, 95% CI [.04, .53]) and non-significant severity effect in a high base rate condition (g = -.03, p = .77, 95% CI [-.26, .19]). Moderated meta-analysis of the three experiments showed that the overall moderation effect of base rate was not significant (QM (df = 1) = 3.75, p = .05).

In three experiments we showed that participants were not susceptible to the severity bias when estimating risk of uncertain events. Instead they used base rate as a primary tool for their judgment. Experiments 1 and 2 showed that the severity effect on probability perception was independent of base rate, whereas Experiment 3 showed that the severity effect occurred at low levels of base rate. Moderated meta-analysis model of the three experiments reached marginal significance, suggesting that the severity bias may emerge at low levels of base rate. These findings lend support to the asymmetric loss function account (Weber, 1994), but offer to re-examine the assumption that the severity bias occurs independent of its base rate.