Week 4 - Video 3: Tidy Patchwork

In this episode we walk through the modeling of a burning question: what about the Seychelles? Does terrain ruggedness impact economic development? And how do clear outliers like Austria and The Seychelles influence the impact? We build a straight=forward probabilistic regression model of gdp (real-per capita) on ruggedness (terrain ruggedness index) across two regions: Africa and the rest of the world as in Nunn and Puga's 2012 RESTAT article. We discuss a outlier information gain and loss observation by observation using Pareto Smoothed Importance Sampling with leave-one-out cross validation, uncertainty measured by the shape parameter in the Generalized Pareto Distribution, and volatility measured by a penalty variance a la the Watanabe approach to non-smooth parameter spaces. We then review parameter by parameter, region by region uncertainties. After that we build a posterior predictive study of gdp to ruggedness. Last, hardly least, we compare our regional indexing results with traditional OLS dummy variable methods to reveal some, perhaps, surprising insights. Much fun for all. Enjoy!