Untangling serially dependent underreported count data for gender-based violence.

Area of Science:

• Statistics
• Public Health
• Criminology

Background:

• Underreporting of gender-based violence (GBV) is a global issue, leading to underestimated data and a flawed understanding of domestic violence prevalence.
• Poor data quality from underreporting biases results, hindering effective societal responses and interventions.
• Accurate data is crucial for addressing the public health and social dimensions of domestic violence.

Purpose of the Study:

• To develop and validate novel time series models capable of accounting for underreported count data.
• To estimate the degree of underreporting in domestic violence statistics.
• To explore the relationship between underreporting patterns and socioeconomic/health indicators.

Main Methods:

• Proposed time series models incorporate a latent integer autoregressive process (order 1) with Poisson innovations.
• A latent binary state, modeled as a first-order Markov chain, accounts for the underreporting mechanism.
• Parameter estimation utilized moment-based and maximum-based methods.

Main Results:

• The developed models were applied to quarterly domestic violence complaints against women in Galicia, Spain (2007-2017).
• The models successfully quantified the extent of underreporting in the observed data.
• Analysis revealed correlations between underreporting frequency/intensity and regional socioeconomic and health indicators.

Conclusions:

• The proposed time series models provide a robust framework for analyzing underreported count data in public health contexts.
• Quantifying underreporting is essential for a realistic assessment of domestic violence magnitude.
• Understanding factors influencing underreporting can inform targeted interventions and improve data collection strategies.