README.md

In bennysalo/predict-recidivism: Analyses for prediction of recidivism in Finnish data using supervised learning

Predictors of recidivism in Finland

This repository contains the results, and the scripts that generated those results, for a study comparing the power of dynamic and static risk factors to predict recidivism in Finland. We use elastic net logistic regression and random forest as supervised learning methods and allow for the possibility that different sets of predictors work better with either algorithm.

A preprint is available at https://psyarxiv.com/v5uq3/ . The current manuscript is a revised version that is currently being rewritten according to reviewer suggestions.

Files are organised in an R package structure (even though currently it is not designed to be installed in R). The following specification might help:

• Analyses that require data that we are not able to make public can be found in /data-raw. These generete R data objects found in /data.
• Analyses that can be run using the results we provide (in /data) can be found in /vignettes. These include analyses of discrimination and calibration plus descriptive statistics for all variables.

The manuscript specifically mentions the following to be found on these pages

• Descriptive statistics as an extension to Table 1 in the revised manuscript can be found in table_1_desciptive_statistics.md. Descriptive statistics can also be found in the vignette Descriptive statistics.
• The process for choosing tuning parameters and R packages used is documented here.

(Updated 2019-03-13)

bennysalo/predict-recidivism documentation built on May 29, 2019, 10:34 a.m.