Forecasting Disappeared and Missing Persons Cases in Mexico Using the ARIMA Model and Implications for Violence Prevention

Background/Purpose: Since 1964, more than 286,000 disappeared and missing persons have been recorded in the Mexican Registry of Disappeared or Missing Persons. Despite increasing rates of disappearances and the Mexican government’s efforts to mitigate this crisis, significant analytical gaps remain in understanding past trends and predicting future patterns, which are crucial for developing proactive and targeted violence prevention strategies. Previous studies have focused on socio-economic, political, and cultural factors contributing to disappearances, yet the temporal dynamics of this phenomenon are poorly understood. Our study aims to bridge this gap by leveraging historical data on disappeared and missing persons in Mexico to develop a robust predictive model that can forecast future trends and provide insights for informed decision-making on resource allocation to stem the predicted tides of disappearances.

Methods: We analyzed data from the Mexican Registry of Disappeared or Missing Persons from 1964 to 2022 (N = 298,759). Utilizing time series analysis, we used the Autoregressive Integrated Moving Average (ARIMA) model to forecast disappearances over the next three years. Integration was employed to achieve data stationarity, which was confirmed by the Augmented Dickey Fuller test. Forecasts were then generated using an ARMA model, which considered both past values and prediction errors, and were evaluated using the Akaike Information Criterion and the Bayesian Information Criterion. Additionally, a linear regression model was tested to identify potential hotspots for increased disappearances.

Results: The time series analysis using the ARIMA model provided a robust forecast for disappearances in Mexico. Despite initial non-stationarity, differencing effectively transformed the data to achieve stationarity (p-value = 0.015), allowing for reliable predictions. The results suggest an increasing pattern in disappearances for the next 3 years. The linear regression analysis revealed a strong and statistically significant association ( 0.94, p < .01) between the annual disappearances in 15 municipalities and the national trend in Mexico from 2006 to 2022. These municipalities are predictive of the national figures, accounting for approximately 24% of the total recorded cases.

Discussion: Our findings underscore the effectiveness of ARIMA and linear regression models in utilizing historical data to predict future disappearances in Mexico. Results suggest that the identified municipalities should be prioritized in resource allocation strategies to enable targeted interventions. By shifting from reactive to proactive measures, policymakers can more efficiently allocate limited resources to address and reduce the incidence of disappearances. This strategic focus not only aims to curtail the current rates but also serves as a preventative framework against potential escalations in these regions. Future research should involve collaborating with affected communities and local stakeholders to explore the specific socio-political and cultural factors contributing to the high rates of disappearances in the identified hotspots. A collaborative approach will enable the development of more targeted and context sensitive prevention strategies that are informed by the lived experiences and insights of those directly impacted by this crisis.