Previous research has shown that individuals appear to systematically underinfer from strong signals and overinfer from weak signals. Less is known about how individuals aggregate information across signals of varying strengths. We study this question using a variation of the balls-and-urns task in which subjects observe a sequence of draws and report posterior beliefs after each signal. We experimentally manipulate the order and relative strength of signals across draws, allowing us to identify how individuals weigh each signal conditional on what they have already observed. In particular, we vary whether the signal strength increases or decreases across draws, enabling us to test whether updating behaviour depends on the order of evidence rather than its joint strength. Our preliminary results suggest that under- and overinference differ when observing a strong signal followed by a weak signal compared to the reverse order, despite the signals being jointly equivalent in Bayesian terms.
We study the role of complexity in how individuals form judgements over potential outcomes. To do so, we propose a framework in which individuals form sophisticated causal mental models to make sense of a complex system. Our framework suggests that these causal models would allow individuals to leverage extensive domain-specific knowledge when forming judgements, but would also leave them vulnerable to exploitation. The more complex the setting and the more domain-specific knowledge an individual has, the more vulnerable they are. We test this hypothesis using an experiment in which we show participants a sequence of causally-related fictional events, generated by AI using real-world data. We experimentally manipulate the complexity and narrative of the events and elicit beliefs over possible outcomes. We then elicit preferences over series of gambles that vary in their complexity and causal narrative. Our research contributes to an emerging literature studying the role of complexity and narratives in individual decision making.
This paper examines how sequential contrast effects, the tendency to judge stimuli relative to preceding experiences, influence online ratings. While laboratory studies have documented contrast effects, field evidence remains limited by non-random ordering. We overcome this challenge by exploiting settings where ordering is either random or plausibly orthogonal to preferences. We identify persistence in rating outcomes using data on hundreds of millions of reviews left in online settings, including Amazon product reviews and Google Maps reviews. Preliminary results suggest positive contrast effects: after rating reviews highly, individuals tend to rate reviews higher than if followed by a lower rating. These findings have important implications for the reliability of online rating systems and help reconcile conflicting evidence between laboratory and field studies of sequential judgments.
We study how firms update their reference points when information is revealed to a market in a setting where firms are set to lose millions of pounds for deviating from rational decision making: football player transfers. We collect data every football player transfer made by firms in the ‘Big Five’ football leagues over a ten-year period and control for variation in transfer fees with data on club finances, player performances, player histories, and player injuries. Our preliminary results suggest that superstar transfers do not shift reference points and do not distort prices in this market.