A Bayesian Mixture Cure Rate Model for Estimating Short-Term and Long-Term Recidivism

Area of Science:

• Statistics
• Survival Analysis
• Biostatistics

Background:

• Standard survival models are inadequate for data with a non-failing proportion.
• Mixture cure rate models address populations with susceptible and non-susceptible individuals.
• These models account for both the probability and timing of failure, influenced by covariates.

Purpose of the Study:

• To propose a Bayesian approach for estimating parametric mixture cure rate models with covariates.
• To investigate risk factors influencing long-term and short-term recidivism survival.
• To apply the methodology to real-world data on prison releases.

Main Methods:

• Utilized a Bayesian framework for inference.
• Employed Markov Chain Monte Carlo (MCMC) methods.
• Estimated probability of eventual failure via binary regression and failure timing via Weibull distribution.

Main Results:

• Developed a robust Bayesian method for mixture cure rate models.
• Demonstrated the model's applicability using England and Wales sexual offender recidivism data.
• Identified key covariates influencing recidivism patterns.

Conclusions:

• The proposed Bayesian approach provides a flexible framework for analyzing survival data with a cure fraction.
• This method enhances understanding of factors affecting recidivism, informing policy and interventions.
• Mixture cure rate models are valuable tools for studying phenomena with non-susceptible populations.