Comparison of Experiences and Descriptions of Basic Economic Preferences in 11 Countries

Background and Motivation

Recent studies have shown that humans exhibit a high degree of context-dependency in encoding the value of rewards, which in some cases leads to suboptimal decisions. However, it remains unclear whether this computational limitation is a universal feature of human cognition. In this study, the authors examine the behavior of 561 individuals from 11 countries (Argentina, Iran, Russia, Japan, China, India, Israel, Chile, Morocco, France, and the United States) to explore whether the context-dependency of reward value encoding is a consistent feature of human cognition.

Research Sources

The study was conducted by Hernán Anlló, Sophie Bavard, Fatimaezzahra Benmarrakchi, Darla Bonagura, and several other scholars from multiple internationally renowned academic institutions, including the École Normale Supérieure in Paris and Waseda University. The research results were published in the journal Nature Human Behaviour. The paper was accepted on April 19, 2024.

Research Process

Behavioral Experiment Design

1. Reinforcement Learning Task (RL Task): This task includes two phases—the learning phase and the transfer phase. In the learning phase, participants choose between two options to obtain the highest reward. Each option has different reward probabilities, and participants learn the value of each option through multiple trials. In the transfer phase, previously learned options are recombined into new decision contexts without providing feedback.

2. Descriptive Decision Task (Lottery Task): This task involves choosing options with known reward probabilities and values, matching the transfer phase of the RL task. Participants choose between high-value and low-value rewards under different probabilities.

3. Questionnaire: After completing the tasks, participants fill out a series of questionnaires to collect data on their socioeconomic, cultural, and cognitive characteristics.

Data Analysis Methods

The study uses a generalized linear mixed-effects model for data analysis, focusing on the correct choice rate as the dependent variable. To quantify the observed decision strategies, a model based on subjective outcome scaling was used to formalize choice behavior.

Participant Characteristics and Selection Criteria

Participants were selected from various socioeconomic and cultural backgrounds to ensure the sample’s broadness and representativeness. Participants had to meet the following criteria: hold nationality of the target country, reside in the target country, have completed at least basic education, and speak the official language of the target country as their mother tongue.

Main Findings

Experiential RL Task

• Learning Phase: Participants from different countries had an average correct choice rate significantly higher than random levels in each decision context, indicating effective learning. However, no significant magnitude effect was found, meaning decision contexts with different expected value differences had no statistically significant impact on learning performance.
• Transfer Phase: In the transfer phase, participants from different countries exhibited similar or even consistent suboptimal choice behavior in certain contexts, further proving that context-dependency is a universal feature of human cognition.

Descriptive Lottery Task

• Risk Preference: Participants’ risk aversion behavior significantly differed among countries. As risk increased, the proportion of participants choosing high-value options decreased in all countries, but the extent of this effect varied significantly across countries.
• Comparison of Descriptive Choices and RL Choices: In similar decision contexts, risk aversion could not explain the preference reversal phenomenon in the RL task, indicating significant cross-cultural differences between descriptive and experiential decision-making.

Computational Results

• Scaling Parameter (ν): The scaling parameter for the RL task showed no significant differences across countries, whereas the scaling parameter for the Lottery task showed significant cross-country differences.

Conclusion

The key conclusion of this study is that context-dependency in reward value encoding is a consistent and cross-culturally stable feature of human cognition, in stark contrast to the cultural sensitivity seen in descriptive decision-making. This finding provides new evidence for reinforcement learning theory and has important implications for designing behavior science-based public policies and interventions.

Research Highlights

1. Cross-Cultural Consistency: Regardless of socioeconomic status, educational level, or cultural background, participants from all countries exhibited highly consistent context-dependent behavior in the RL task.
2. Distinction Between Context-Dependency and Risk Aversion: Risk aversion could not explain the suboptimal choices in the RL task, indicating that these two decision-making modes are cognitively independent.
3. Validation of Computational Models: Using a simple subjective outcome scaling model, the study validated the cross-cultural differences in experiential and descriptive decision-making, demonstrating the varying impact of different cultural backgrounds on descriptive decision-making.

Significance and Value

This study contributes to understanding universal features and cross-cultural stability of human cognition, making important contributions to fields such as psychology and behavioral economics. In particular, the stability of experiential decision-making models suggests that such models can be relied upon for broader applications and promotion when designing international public policies.

This research provides an important reference for verifying cross-cultural consistency in human decision-making behavior and lays a foundation for future cross-cultural studies.