We will

• Understand "the credibility revolution in economics" and draw parallels to the current and coming credibility crisis in data science
• Look at how we can avoid it altogether, by learning from decades of econometric history
• Review the ideal randomized control trial and talk about why they can be difficult in the context of data science
• Look at methods from the social science toolbox for quasi-experimental design such as
• Instrumental variables
• Matching
• Difference-in-differences
• Regression discontinuity design
• Survey experiments
• Finally, review a few good papers from the literature that illustrate the application of these techniques