Democracy Dataset
In many projects, I am working on the usage of machine learning techniques for economic applications. A particular advantage of ML-techniques is that it allows for non-linear classifications and data analyses without the need of ex ante functional assumptions. In Gründler and Krieger (2016, 2020, 2021) we develop a new technique for data aggregation that we use to construct the ML Democracy Index. A visualization of our index, data and descriptions can be found in Gründler and Krieger (2021) and on the official page at https://ml-democracy-index.net/. A discussion of the underlying machine-learning aggregation versus other types of data aggregation cabn be found in Gründler and Krieger (2021): Should we care more about data aggregation, European Economic Review, forthcoming.
Democracy in the World
Continuous Machine Learning Indicator (CSVMDI), post-2010 period
Source: Gründler and Krieger (2019).
This pages provides access to the SVMDI Democracy Database that I have compiled together with Tommy Krieger (ZEW Mannheim). The indicator is constructed based on machine learning techniques for pattern recognition. Specifically, we provide a new aggregation technique based on Support Vector Machines (SVMs), which transfers the problem of data aggregation into a non-linear optimization problem. The advantage of our machine learning algorithm is that it strongly reduces assumptions about the functional form between the input variables and the composite measure and allows for any non-linear relationship between the input variables and the outcome. Also, our method allows us to derive distributions to account for measurement errors in the data and enables the computation of dichotomous and continuous indicators based on the identical set of assumptions.
The SVMDI Dataset is available for 186 countries between 1919 and 2019. For methodical details, see the latest version Gründler and Krieger (2021).