Application of Double/Debiased Machine Learning

Available data grows fast, for example characteristics of individuals. When estimating the average treatment effect by controlling for characteristics using Ordinary Least Squares, this method will become inconsistent. The new method Double/Debiased Machine Learning delivers unbiased estimators and is able to construct confidence intervals that is root-N consistent, the sample size is N, and approximately normally distributed. This method is applied to the data sets of the two papers “The Long-Term Effects of Africa’s Slave Trades” and “The Slave Trade and the Origins of Mistrust in Africa”. These papers have drawn conclusions about the effect of Africa’s Slave Trade using OLS and Instrumental Variables, based on a large data set. The first paper uses a specific measure for the slave trade in every table. This thesis shows that it is better to use another more important measure. Moreover, this thesis shows more reliable estimations and confidence intervals for the treatment variable. Besides applying DML, the most important results of these two papers are replicate using OLS. For almost every result the estimations are the same, however, when estimating for channels of causality, the estimations differ.