DfE-DB: A systematic database of 3.8 million human decisions across experience-based tasks

The paper introduces the Decisions From Experience Database (DfE-DB), a comprehensive open-access resource harmonizing 3.8 million trial-level decisions from 11,921 participants across 168 studies to enable systematic comparisons of how design features shape human decision-making tendencies and foster reproducible research.

The Gist

Imagine that for years, scientists trying to understand how humans make choices based on their past experiences have been like detectives working in separate, isolated rooms. Each detective has a pile of clues (data), but they are written in different languages, stored in different notebooks, and organized in confusing ways. Because of this, it's been nearly impossible to compare one detective's findings with another's to see the big picture.

This paper introduces a massive new tool called DfE-DB (Decisions From Experience Database) that acts like a giant, universal translator and filing cabinet for all these scattered clues.

Here is what they did, broken down simply:

1. The Great Cleanup
The researchers gathered a staggering 3.8 million individual decisions made by nearly 12,000 people across 168 different studies. Before, this data was messy and fragmented. The team took all these raw, unorganized bits of information and "harmonized" them. Think of it as taking thousands of different jigsaw puzzles from different boxes, sorting every single piece, and putting them all into one giant, organized box where every piece fits together perfectly.

2. The Universal Map
They didn't just dump the data in; they created a detailed map. They categorized every study based on 13 specific design features. Imagine if you were trying to understand why people choose different flavors of ice cream. Instead of just looking at the flavors, you'd also note the temperature of the room, the color of the spoon, and whether the person was hungry. The DfE-DB does exactly this for decision-making, tagging every experiment with details like:

• Did the person get immediate feedback?
• Were the outcomes predictable or random?
• Did the rules of the game stay the same or change?

3. What They Discovered
Once they had this unified map, they could finally see patterns that were previously hidden. They found that how people choose—whether they lean toward taking risks, playing it safe, or going for the average result—changes drastically depending on the "rules of the game" (the design features).

For example, the way a task is set up (like whether you get told the result immediately or have to wait) strongly shapes whether a person acts boldly or cautiously. The database showed that these specific design features explain a huge amount of the differences seen between studies. It's like realizing that the reason people behave differently in different experiments isn't because humans are inconsistent, but because the "games" they are playing have subtle but powerful differences.

4. The Result
The main goal of this paper is to provide a unified, open-access foundation for science. By giving researchers a single, reliable place to compare data, it allows them to test if the rules of human behavior they discover in one study hold true in another. It turns a chaotic collection of isolated experiments into a cohesive, integrated science of how we learn from our experiences.

In short, the paper didn't just build a bigger library; it built a system that lets us finally read all the books in that library at the same time to understand the full story of human decision-making.

Technical Summary

Technical Summary: DfE-DB

Problem Statement
Research on learning from experience in human decision making has been constrained by a lack of systematic comparability across studies. Progress in this field is hindered by two primary issues: heterogeneous task designs and fragmented data formats. These inconsistencies prevent researchers from conducting unified quantitative comparisons, obscuring the generalizability of behavioral phenomena and computational models across different experimental contexts.

Methodology
To address these fragmentation issues, the authors constructed the Decisions From Experience Database (DfE-DB). This resource involves the systematic aggregation and harmonization of raw behavioral data from 168 distinct studies involving 11,921 participants. The database encompasses approximately 3.8 million trial-level decisions spanning 13 different experimental paradigms.

The core methodological contribution lies in the standardization process. The authors did not merely aggregate data; they harmonized raw behavioral records and classified each study according to 13 key design features. This classification framework allows for the systematic mapping of experimental variables, such as feedback types, outcome structures, stationarity, and sampling methods, enabling quantitative comparisons that were previously impossible due to structural heterogeneity.

Key Contributions

1. Unified Infrastructure: The DfE-DB serves as a structured, open-access resource that unifies disparate datasets into a single, extensible data infrastructure.
2. Design Feature Taxonomy: By classifying studies along 13 specific design features, the database provides a mechanism to analyze how specific experimental parameters influence decision-making outcomes.
3. Reproducibility: The resource is designed to facilitate reproducible research, allowing for the rigorous testing of the generality of behavioral phenomena and computational models across a broad spectrum of experimental conditions.

Results
Leveraging the DfE-DB, the authors conducted an analysis of choice tendencies, specifically examining preferences toward higher risk, expected value, or experienced mean. The results indicate that these tendencies vary substantially across different paradigms. Crucially, the analysis demonstrates that this variability is strongly shaped by core design features, including:

• Feedback type
• Outcome structure
• Stationarity
• Sampling methods

The database reveals that these design features explain a substantial portion of the cross-study variability in decision-making behavior. Furthermore, the analysis highlights the existence of underexplored paradigm variants within the existing literature.

Significance
The paper positions the DfE-DB as a foundational tool for fostering a more integrated science of decisions from experience. By providing a unified platform, the database enables researchers to move beyond isolated study comparisons to systematic, quantitative evaluations of how experimental design shapes human behavior. The authors claim that this infrastructure is essential for testing the robustness and generality of existing behavioral theories and computational models, thereby addressing the limitations imposed by fragmented data and heterogeneous task designs.