Reconstructing Nonparametric Productivity Networks

Area of Science:

• Economics
• Network Science
• Statistical Inference

Background:

• Network models are widely used for economic productivity analysis, often estimated non-parametrically via Data Envelopment Analysis (DEA).
• Existing network DEA models lack a robust statistical framework for inference, hindering a deep understanding of complex network dynamics.
• The complexity of network processes in systems limits the inferential capabilities of current models.

Purpose of the Study:

• To develop a general Bayesian framework for inferring network structures in economic systems.
• To address the limitations of existing network DEA models by providing a statistical inference approach.
• To estimate unobserved network linkages and understand the relationships driving system performance.

Main Methods:

• Developed a general Bayesian framework integrating information science, machine learning, and statistical inference from complex systems physics.
• Applied the framework to bibliometric data for world countries to analyze knowledge production, including own and cross-disciplinary research.
• Estimated unobserved network linkages to infer the underlying network production technology.

Main Results:

• Identified significant interactions between related disciplinary research outputs in terms of both quantity and quality.
• Demonstrated the framework's ability to infer underlying network structures and linkages.
• Found that cross-disciplinary research linkages are significant drivers of knowledge production.

Conclusions:

• The developed Bayesian framework offers a systematic approach for inferring network structures and aids model selection when the network is unknown.
• Results on cross-disciplinary research linkages can inform targeted research funding and institutional strategies.
• The framework is broadly applicable to various settings with spillovers, including public and private sector decision-making and coordination.