Brown University - Economics 1660The spread of information technology has lead to the generation of vast amounts of data on human behavior. This course explores ways to use this data to better understand the societies in which we live. The course weaves together methods from machine learning (OLS, LASSO, trees) and economics (reduced form causal inference, economic theory, structural modeling) to work on real world problems. We will use these problems as a backdrop to weigh the importance of causality, precision, and computational efficiency.
Class 1: Introduction and ClusteringThe promise of 'big data', and the theme of this course: what do you need to add to data in order to learn from it?
References: Hastie, Tibshirani, and Friedman section 14.7
Class 2: VisualizationA bottleneck in understanding and communicating complex ideas is their representation. How can effective visual representations reveal relationships in data?
Class 3: TreesWhile our visual system can pick out patterns in low dimensional data, it can be less effective with high dimensional data. How can algorithms help us identify patterns in data?
Class 4: MeasurementIncreasingly, everyday behaviors are captured through our interactions with information technologies. What behaviors can we measure?
Class 5: FitHow should we weigh detail against confidence and comprehensibility?
References: Hastie, Tibshirani, and Friedman section 7
Class 6: Linear Models and RegularizationWhat are the benefits of different representations? How can we control fit in an elegant manner?
References: Hastie, Tibshirani, and Friedman section 3
Class 7: Structure and CausalityHow can we decide what actions to take based on relationships we find in data? How can formal thinking help?
Problem Set 7: Structural Models and Identification, and Supporting Files (for both PS6 and PS7)
References: Pearl, Glymour, and Jewell chapters 3 and 4
Class 8: Structure and Strategic InteractionWhat if agents react to our actions--and to each other's reactions? How can formal modeling help predict the results of major changes?
Thanks to Simon Freyaldenhoven, Nicholas Hartmann, and Burak Karaca for excellent research assistance.