Diff-in-Diff in a Regression Framework

Overview

This is the first of three modules on regression approaches to difference-in-differences estimation. In this module, we review the basic two-by-two difference-in-differences regression specification, and discuss extensions that allow for continuous variation in treatent intensity and the inclusion of period-specific fixed effects.

Readings

Does a ban on informal health providers save lives? Evidence from Malawi by Susan Godlonton and Edward Okeke.

Review Questions

1. What are traditional birth attendants? Why did the government of Malawi decide to ban them?
2. What data set(s) do Professor Godlonton and Dr. Okeke use to estimate the impact of Malawi's ban on traditional birth attendants?
3. What is there identification strategy? What is the treatment group? Who or what is the treatment group compared to?
4. What alternatives to traditional birth attendants are available to expectant mothers in Malawi?
5. What are the impacts of Malawi's ban on traditional birth attendants? Does the ban lead to a decline in the use of TBAs? An increase in the use of skilled birth attendants? What other impacts do the authors report?
6. Did Malawi's ban on traditional birth attendants reduce neonatal mortality?

Video Lectures

Lecture 5.1: 2X2 Difference-in-Differences Specifications (19:34)

Review Questions

1. When you estimate a simple 2X2 difference-in-differences specification in a regression framework, what are the three independent variables? How do we interpret each of the coefficients?
2. What research design did Dr. David Hausman and Professor Dorothy Kronick use to estimate the impact of police reform in Chicago? What was there empirical (ie regression) specification? What did they find?
3. What specification did Professsor Favid Card and Professor Alan Krueger use to estimate the impact of New Jersey's increase in the minimum wage? What was the outcome variable of interest? What did they find?

Lecture 5.2: Diff-in-Diff with Continuous Treatment Variation (28:06)

Review Questions

1. When might we wish to use a continuous treatment variable in a difference-in-differences analysis?
2. What are the two approaches to difference-in-differences estimation when we observe continuous variation in treatment intensity?
3. When we dichotomize treatment (ie partition the sample into a more intensely treated group and a less intensely treated group), how should we interpret the estimated regression coefficient?
4. When we estimate a difference-in-differences specification with a continuous measure of treatment intensity, which observations/units are implicitly included in the comparison group?

Lecture 5.3: Diff-in-Diff with Time Period Fixed Effects (19:14)

Review Questions

1. What are fixed effects? How do they work? If you wanted to implement fixed effects by hand, what would you do?
2. When and why might we include fixed effects in a difference-in-differences specification?
3. How do Professor Susan Godlonton and Dr. Edward Okeke estimate the impact of Malawi's ban on traditional birth attendants on neonatal mortality? What do they find?
4. How does Professor Esther Duflo estimate the impact of Indonesia's school construction program on educational attainment? What doew she find?

Empirical Exercise

The empirical exercise for this module is not publicly available (as of March 2021). We hope to make the exercise available to the public in the future.

Acknowledgments

Special thanks to Patrick Blanchenay, Garret Christensen, Liz Cascio, Jess Goldberg, Guy Grossman, Kirabo Jackson, Michael Kevane, Scott Imberman, Manisha Shah, and Mike Urbancic for their diff-in-diff suggestions.