Due to the specificities of the thematic areas, the unique set of variables has been defined in each case. 1.3 Data Analysis Accuracy of the built models diverged between the areas due to the limitations in the data availability or accessibility. Nevertheless, the information collected allowed for the designing of a dataset that is not only the most comprehensive of the attempts made so far in the area of deprivation of liberty, but above all – tailored to the unique needs of this Global Study. The number of children deprived of liberty has been estimated using regression-based models. These methods allowed for assessing the magnitude of the phenomenon despite the limitations in data availability. During the data analysis, a variety of approaches have been tested, including in particular: • Multiple linear regression – several independent variables are used to predict the value of a target variable (the number of children deprived of liberty in certain settings); • multiple imputation by chained equations (performed with predictive mean matching) – missing values are imputed in the iterative process that starts with estimating the variable with the least missing values using complete data (first iteration), followed by estimating the variable with the second least missing values using complete data and, additionally, values imputed in the first step (second iteration) etc.; • random forest regression – estimation is done by subdividing the dataset based on the values of predictors; typically, one subdivision tree includes three to five variables, meaning that the dataset is divided at three to five stages to create smaller clusters of similar cases (in addition, the same dataset is subdivided many times with various combinations of variables). The two latter methods have been recognised as the most efficient in predicting unknown values, confirming recent developments in the field of statistics.4 Separate models have been designed for the following sections: 1) administration of justice, 2) children in migrationrelated detention, 3) children living in prison with their primary caregivers, 4) children in 4 Anoop D. Shah et al., ‘Comparison of Random Forest and Parametric Imputation Models for Imputing Missing Data Using MICE: A CALIBER Study’, American Journal of Epidemiology, Vol. 179(6), 2014, pp. 764–774. Daniel J. Stekhoven et al., ‘MissForest – non-parametric missing value imputation for mixed-type data’, Bioinformatics, Vol. 28(1), 2012, pp. 112–118. Akbar K. Waljee et al., ‘Comparison of imputation methods for missing laboratory data in medicine’, BMJ open, Vol. 3, 2013, pp. 1–7. Jesper N. Wulff et al., ‘Multiple Imputation by Chained Equations in Praxis: Guidelines and Review’, The Electronic Journal of Business Research Methods, Vol. 15(1), 2017, pp. 41–56. 39