"Dribia are very professional and competent: they have understood the complexity of the RisCanvi protocol and have done a good audit of the algorithmic part, with many interesting recommendations."
FEATURES
Audit of the recidivism risk estimation algorithm: evaluation of operation and implementation.
IMPACT
Analysis of the performance of the current algorithm and possible biases. Exploration of new algorithms that improve performance.
ACHIEVEMENTS
Complete report on the current status of the algorithm and compilation of recommendations for possible improvements to be implemented.
BUSINESS PROBLEM
RisCanvi is a tool and set of professional action protocols for the assessment and management of the risk of violent behavior and sentence breaches in the penitentiary and community settings of people interned in ordinary centers and on parole in the Catalan penitentiary system.
The tool was designed to improve individualized predictions of the risk of committing general recidivism, violent recidivism or intra-institutional violence, engaging in self-directed violent behavior, or breaking permits. The RisCanvi algorithm estimates the risk of recidivism based on assessments made by professionals. The first version was implemented in 2009 and has undergone several periodic reviews, the last in 2017.
The General Directorate of Penitentiary Affairs wants a new external review of the algorithm to be carried out in order to detect strengths and points to adapt to continue improving the tool.
DRIBIA’S SOLUTION
Dribia has conducted an audit of the RisCanvi algorithm that included a review of current operation (performance and detection of potential biases), an exploration of new algorithms and the compilation of recommendations for possible improvements to be implemented. The compilation of data on the operation of RisCanvi and the recidivism that occurred has allowed the new evaluation to be carried out taking into account the five types of recidivism that the algorithm evaluates. To consult the executive summary in Catalan, click here.
