To Trust or to Stockpile: Modeling Human-Simulation Interaction in Supply Chain Shortages

Title of the Paper

To Trust or to Stockpile: Modeling Human-Simulation Interaction in Supply Chain Shortages

Paper Information

• Research Area: Human-Computer Interaction and Supply Chain Management
• Keywords: Human-Computer Interaction, Supply Chain Shortages, Hidden Markov Model (HMM), Principal Component Analysis (PCA), Sequence Analysis, Simulation Games, Behavior Modeling

Research Background and Problem

• Issues and Challenges:

  • Understanding human decision-making processes in dynamic and complex environments, particularly in the context of supply chain shortages (e.g., drug shortages), is challenging.
  • There is a lack of research on how to extract useful information from decision-making processes, especially in the field of modeling human behavior in simulation or game systems.
  • The diversity of human behavior and responses to system dynamics complicates prediction and control. For instance, participants' reactions to shortages can significantly impact the state of the supply chain.

• Significance:

  • Supply chain shortages affect the availability of critical products (e.g., pharmaceuticals) and the normal operation of medical distribution chains, especially during global emergencies like COVID-19.
  • Understanding human decision-making patterns in supply chain management contexts can help develop more effective intervention strategies and enhance supply chain resilience.

• Relevant Background:

  • Simulation games and gamification methods are increasingly used to study human decision-making behaviors in complex systems, including approaches like "gamettes."
  • This study leverages a supply chain simulation framework combined with psychological behavior modeling techniques to explore human interaction with a simulated environment.

Solution

Methods and Framework

• Research Methodology:

  • A three-step approach was proposed to characterize human-simulation interaction:
    1. System State Representation: Use Principal Component Analysis (PCA) for dimensionality reduction to extract key states from the high-dimensional supply chain system, combined with hierarchical clustering to identify "system states."
    2. Behavioral Response Modeling: Use Hidden Markov Models (HMM) to quantify player behavior patterns and classify behaviors through sequence analysis.
    3. Behavior-System Interaction Analysis: Jointly analyze the relationship between behavioral patterns and system changes.

• Experimental Design:

  • The experiment was based on a pharmaceutical shortage supply chain simulation, creating an online "gamette" game where participants acted as "wholesalers" in the supply chain.
  • The system simulated two experimental conditions:
    1. No-Info: Players could not view manufacturer inventory information.
    2. Info: Players had access to manufacturer inventory information.
  • The game process was divided into five phases, including "system stability, supply disruption, and recovery" states.

• Key Techniques:

  • PCA was used to extract dynamic system states, identifying three system states: stable, disrupted, and recovering.
  • HMM analyzed participants' ordering behavior patterns and identified sequence patterns through probability transition matrices.
  • Sequence Analysis classified player behavior types into three main categories: hoarders, reactors, and followers.

Research Findings

Key Discoveries

1. System Representation:

   • Identified three primary states of the dynamic system:
     • Stable State: Players received consistent shipments and met downstream demand.
     • Disruption State: Supply chain was obstructed due to manufacturer shutdowns.
     • Recovery State: Supply chain gradually normalized after manufacturers resumed production.

2. Behavioral Modeling:

   • HMM identified eight response patterns among players, which were classified into three player types through sequence analysis:
     • Hoarders: Frequently ordered above recommended quantities, tending to stockpile or maintain safety stock.
     • Reactors: Adjusted behavior after receiving information about supply shortages, previously following system recommendations.
     • Followers: Consistently adhered to order recommendations with minimal deviation.

3. Impact of Information Sharing:

   • During the stable phase, information sharing significantly reduced hoarders' order deviations, mitigating over-ordering caused by uncertainty.
   • However, during the recovery phase, hoarders exhibited higher order deviations (especially under the information-sharing condition), potentially due to psychological concerns about future shortages.

4. Interaction Between System and Behavior:

   • Player behavior was influenced by system states but also had reciprocal impacts on system states. For example, different decision-making behaviors could stabilize or further destabilize the supply chain.
   • The effects of information sharing on certain player types (e.g., hoarders) were complex and might depend on their psychological responses to uncertainty.

Comparative Advantages Over Existing Methods

• Emphasized the impact of "human dynamic behavior" on exacerbating or alleviating supply chain shortages, addressing the shortcomings of previous studies that focused solely on system optimization.
• Provided a reproducible behavior modeling framework, incorporating simulation dimensionality reduction (PCA), behavior pattern modeling (HMM), and player classification.

Limitations and Future Directions

1. The study's participants were primarily online users who may lack expertise in supply chain management.
   • Future Directions: Include professional supply chain managers or data scientists to validate the generalizability of the findings.
2. Limited analysis of extreme behaviors (e.g., highly irrational ordering behaviors were treated as outliers).
3. Gamification may have led some players (e.g., followers) to oversimplify their decision-making.
   • Improvement Suggestions: Introduce more complex decision-making scenarios to observe multidimensional behaviors.
4. The study focused solely on one form of information optimization (manufacturer inventory sharing).
   • Expansion Directions: Explore additional types of supply chain information sharing, such as upstream and downstream data visibility.

Conclusion

This paper proposed an integrated framework to study complex dynamic human-computer interactions within a gamified supply chain simulation environment, combining system state modeling, human decision-making behavior analysis, and interaction impact assessment. The findings reveal the potential heterogeneity in human behavior patterns and their dynamic impact on supply chain shortages, particularly the possible negative incentives (e.g., hoarding) under information-sharing scenarios. These insights provide critical decision-making support for optimizing supply chain strategies, especially for disaster management and emergency supply chain planning.