Evaluating Spatial Interaction Models for Regional Mobility in Sub-Saharan Africa

Area of Science:

• Epidemiology
• Mobility Science
• Computational Social Science

Background:

• Spatial interaction models, like gravity and radiation models, are widely used in infectious disease epidemiology to describe human mobility.
• These models, primarily developed and validated in high-income countries, are increasingly applied to diverse populations, including those in Sub-Saharan Africa.
• Unique factors in Sub-Saharan Africa, such as transport constraints, informal economies, and rapid urbanization, may limit the generalizability of existing mobility models.

Purpose of the Study:

• To evaluate the accuracy of standard gravity and radiation models in representing human mobility in Kenya.
• To compare model performance against empirical mobility data derived from mobile phone usage.
• To assess the implications of model inaccuracies for infectious disease spread modeling in African contexts.

Main Methods:

• Review of infectious disease frameworks and their incorporation of human mobility at various spatial scales.
• Analysis of anonymized mobile phone data from approximately 15 million individuals in Kenya to capture spatiotemporal mobility patterns.
• Systematic comparison of observed mobility with predictions from gravity and radiation models.

Main Results:

• Gravity and radiation models systematically fail to accurately capture human mobility patterns in Kenya as measured by mobile phone data.
• Both models significantly overestimate the spatial extent of travel and demonstrate poor performance in rural areas.
• Distinct patterns of failure were observed for each model concerning travel routes and volumes.

Conclusions:

• Standard spatial interaction models are not directly generalizable to populations in Sub-Saharan Africa due to differing mobility characteristics.
• The overestimation of travel spread and poor rural performance by these models can lead to misrepresentations of disease transmission dynamics.
• Infectious disease modeling frameworks relying on these conventional models may produce inaccurate predictions for disease spread in many African populations.