11 Conclusion

Causal inference is an important and fun area. It’s fun because the potential outcomes model is both an intuitive and a philosophically stimulating way to think about causal effects. The model has also proved quite powerful at helping us better understand the assumptions needed to identify causal effects using exotic quasi-experimental research designs beyond the randomized control trial. Pearl’s directed acyclic graphical models are also helpful for moving between a theoretical model and an understanding of some phenomenon and a strategy to identify the causal effect you care about. From those DAGs, you can learn whether it’s even possible to design such an identification strategy with your data set. And although that can be disappointing, it is a disciplined and truthful approach to estimation. These DAGs are, in my experience, empowering, extremely useful for the design phase of a project, and adored by students.

The methods I’ve outlined are merely some of the most common research designs currently employed in applied microeconomics. I have tried to selectively navigate the research to bring readers as close to the frontier as possible. But I had to leave some things out. For instance, there is nothing on bounding or partial identification. But perhaps if you love this book enough, there can be a second edition that includes that important topic.

Even for the topics I did cover, these areas are constantly changing, and I encourage you to read many of the articles provided in the bibliography to learn more. I also encourage you just to use the links provided in the software code throughout this book and download the data files yourself. Play around with the programs, explore the data, and improve your own intuition on how to use R and Stata to tackle causal inference problems using these designs. I hope you found this book valuable. Good luck in your research. I wish you all the best.