Abstract
Risk attitudes, including higher-order risk attitudes, are typically measured using transparent presentations of payoff distributions. Yet many real choices require people to evaluate risks that are visually or cognitively complex. We conduct experiments that vary visual complexity in choices over second-, third-, and fourth-order risks while holding objective payoff distributions fixed. Complexity changes measured risk attitudes in all three domains. In the main second-order task, complexity shifts subjects toward the safer lottery even when the safer lottery itself is complex, a pattern inconsistent with simple avoidance of complex options. In third-and fourth-order tasks, subjects are generally less likely to select visually complex options, so complexity can make individuals appear more or less prudent or temperate depending on which option is complex. A follow-up second-order diagnostic experiment shows that subjects also choose the safer lottery more often when both options are complex, ruling out a pure option-specific cost of complexity. Structural estimates using the pooled second-order data favor an ambiguity-like pessimism model in which visually complex lotteries are evaluated partly as if weight were shifted toward their worst outcomes. Elicited risk attitudes therefore reflect not only preferences over payoff distributions, but also the complexity of how those distributions are represented.