[Paper Review] Probabilistic Estimation of Hidden Migrant Fatalities Along the Central Mediterranean Route

Gregor Zens, Zoe Sigman|arXiv (Cornell University)|Mar 20, 2026

Trauma and Emergency Care Studies0 citations

TL;DR

The paper develops a Bayesian latent-class, time-stratified model to estimate total migrant fatalities and deadly incidents along the Central Mediterranean route, accounting for unobserved incidents and fatalities.

ABSTRACT

Estimating the number of migrants who die or go missing along dangerous routes such as the Central Mediterranean remains challenging as available records are incomplete. Some incidents are never documented, and fatalities associated with such unobserved incidents are absent from observed totals. We propose a Bayesian approach for probabilistic estimation of total migrant fatalities in such settings. Building on recent developments in multiple-systems estimation, we develop a time-stratified latent-class framework that accommodates missing fatality counts for unobserved incidents. We apply the method to recoded incident-level data from the Missing Migrants Project for the Central Mediterranean route from 2014 to 2025, encompassing 25,712 fatalities across 1,562 incidents. Our model yields 95% credible intervals of 30,426-39,172 fatalities and 2,200-2,591 deadly incidents, indicating that approximately 66%-85% of fatalities and 60%-71% of incidents are reflected in the available data. We estimate that unreported fatalities were concentrated between 2014 and 2016. Furthermore, we document that reporting likelihood increases with incident severity, implying that smaller incidents are most likely to remain undetected. While contingent on modeling assumptions and incomplete data, our method provides a broadly applicable and principled alternative to naive data adjustment methods.

Motivation & Objective

• Quantify the total number of migrant fatalities and deadly incidents on the Central Mediterranean route from 2014–2025.
• Develop a probabilistic framework that accounts for unobserved incidents and associated fatality counts.
• Model dependencies between reporting sources and between fatalities and reporting patterns.
• Incorporate time stratification to reflect changing surveillance and incident dynamics.
• Assess the extent to which observed data capture the true mortality burden and identify periods with higher undercounting.

Proposed method

• Propose a time-stratified latent-class Bayesian model for marked incidents with unobserved fatalities.
• Use a four-category meta-observer coding (UN/IGO, Official, NGO/Humanitarian, Media) to model reporting patterns.
• Model log-fatality marks with cluster-specific Gaussian distributions within latent classes.
• Treat the total number of incidents per stratum as Poisson and augment data with unobserved (missed) incidents.
• Employ a sparse finite mixture prior to infer the number of active latent clusters and share cluster parameters across strata.
• Implement a Gibbs sampler with data augmentation to sample latent classifications and parameters.

(a) Posterior distribution of the total number of incidents.

Experimental results

Research questions

• RQ1 What is the total number of migrant fatalities and deadly incidents on the Central Mediterranean route between 2014 and 2025?
• RQ2 How do these totals evolve over time and across seasons?
• RQ3 How does incident severity relate to the probability of being captured by different surveillance systems?
• RQ4 How can latent-class structure and time stratification improve estimation relative to naive data adjustments?

Key findings

• The baseline model estimates 95% credible fatalities between 30,426 and 39,172 (roughly 7–9 per day) and 2,200–2,591 deadly incidents.
• Observed data from the Missing Migrants Project record 25,712 fatalities and 1,562 incidents, suggesting about 66%–85% of fatalities and 60%–71% of incidents are captured in the data.
• Unobserved fatalities were likely concentrated in 2014–2016.
• The average reporting likelihood increases with incident severity, so smaller incidents are more likely to go undetected.
• Reporting patterns are temporally variable but show overall stability in aggregate, with complex dependencies between reporters and between fatalities and reporting.

(b) Posterior distribution of the total number of fatalities.