Abstract
Today, most endeavours require teamwork by people with diverse skills and characteristics. In managing teamwork, decisions are often made under uncertainty and resource constraints. The strategies and the effectiveness of the strategies different people adopt to manage teamwork under different situations have not yet been fully explored, partially due to a lack of detailed large-scale data. In this paper, we describe a multi-faceted large-scale dataset to bridge this gap. It is derived from a game simulating complex project management processes. It presents the participants with different conditions in terms of team members' capabilities and task characteristics for them to exhibit their decision-making strategies. The dataset contains detailed data reflecting the decision situations, decision strategies, decision outcomes, and the emotional responses of 1,144 participants from diverse backgrounds. To our knowledge, this is the first dataset simultaneously covering these four facets of decision-making. With repeated measurements, the dataset may help establish baseline variability of decision-making in teamwork management, leading to more realistic decision theoretic models and more effective decision support approaches.
Introduction: The Challenge of Teamwork Management
Managing a team is a constant exercise in decision-making under uncertainty. Managers must allocate tasks to members with diverse, often hidden, capabilities while juggling tight deadlines and limited resources. Theoretical models often fail to capture how real people behave in these high-pressure scenarios—specifically, how they fall into traps like "herding" (overloading a few reliable workers while others sit idle).
The primary bottleneck in improving teamwork management has been a lack of large-scale, granular data. Most existing studies rely on small samples or focus only on outcomes, ignoring the "why" (strategy) and the "feel" (emotion) behind the decisions. This paper introduces the Agile Manager dataset to bridge this gap.
The Agile Manager Simulation
The authors developed a custom game platform called "Agile Manager" to act as a data collection engine. The simulation places participants in the role of a project manager overseeing 10 "Worker Agents" (WAs). Each WA has different programmed traits for productivity and quality.
Participants must assign 20–30 tasks per round. They only have partial information about the WAs' capabilities, which they must learn over time through trial and error. The system logs every task assignment and its outcome (success, failure due to low quality, or failure due to tardiness).
After a session, participants report their strategy using a novel "User Strategy Index" (a binary bit-string representing different allocation logic) and record their emotional state using the "AffectButton"—an interactive widget that captures facial expressions—and Likert scales for six basic emotions.
Dataset Composition
The dataset, hosted on the Synapse platform, consists of several interconnected files covering 1,144 participants and over 495,000 individual decisions:
- Decisions.xlsx: The core "trajectory" data containing 495,533 records of specific decision points.
- Game Sessions.xlsx: Summary data for 9,854 completed sessions, including scores and emotional responses.
- Users.xlsx: De-identified demographic data for 1,144 participants (age, gender, education, and location).
- Psychometric Data: Optional personality (TIPI) and affective (PANAS) survey results for a subset of participants.
Key Insights and Results
This is the first dataset to link four critical facets of management: Situation (task difficulty/WA workload), Strategy (how the user thinks they are playing), Outcome (actual scores vs. an AI benchmark), and Emotion (how the user feels about their performance).
The game includes an AI competitor that uses a greedy, reputation-based algorithm, providing a consistent baseline to measure human efficiency. The researchers specifically manipulated the correlation between a worker's speed and their quality to study how humans adapt when their "fastest" worker is also their "sloppiest."
Key results from the initial analysis include:
- Engagement: Participants from Singapore played an average of 10.9 sessions, while those from China averaged 4.9.
- Performance Variability: As game levels increased in difficulty, the workload became more challenging, leading to higher rates of both "tardiness" (missed deadlines) and "low quality" failures.
- Strategy Mapping: The "User Strategy Index" revealed that many users combine multiple strategies, such as balancing workload while simultaneously favoring high-reputation workers.
Limitations and Future Potential
While the simulation captures resource constraints and uncertainty, it remains a game. Real-world management involves social nuances and interpersonal communication not captured here. Additionally, while demographic data is robust, the deeper personality surveys (TIPI/PANAS) were optional, resulting in a smaller subset of participants (approx. 150).
The authors suggest using this data to train neural networks that can predict a manager's emotional response or to develop more "persuasive" decision-support systems tailored to an individual's specific management style.
Summary updated