The Impact of Policy X: A Difference-in-Differences
Approach
Introduction
This research examines the effect of Policy X on Outcome Y using a Difference-in-Differences
(DiD) approach. The treatment group consists of … while the control group is composed of …
Study: Something
Data Description
Table PLACEHOLDER presents summary statistics for key variables before and after Policy X.
Pre-Treatment Post-Treatment Pre-Control Post-Control
Variable Mean Mean Mean Mean
Outcome Y 10.5 12.1 10.8 11.0
Covariate A 5.2 5.1 5.3 5.2
Table PLACEHOLDER: Summary Statistics
Empirical Strategy
We estimate the following Difference-in-Differences regression model:
′
𝑌𝑖𝑡 = 𝛽0 + 𝛽1 Post𝑡 + 𝛽2 Treatment𝑖 + 𝛽3 (Post𝑡 × Treatment𝑖 ) + 𝑋𝑖𝑡 𝛾 + 𝜀𝑖𝑡
where 𝑌𝑖𝑡 is the outcome for unit 𝑖 at time 𝑡, Post𝑡 is an indicator for post-treatment period,
Treatment𝑖 is an indicator for treatment group, and 𝑋𝑖𝑡 includes control variables.
We justify the use of covariates PLACEHOLDER based on logic (as follows) or domain-
specific precedent (references).
Results
Table PLACEHOLDER presents the results of our Difference-in-Differences estimation.
1
(1) Baseline (2) Controls (3) Robust
Post × Treatment 2.1*** 1.8** 1.9**
(0.5) (0.6) (0.6)
Controls No Yes Yes
Observations 2000 2000 2000
R-squared 0.15 0.18 0.19
Table PLACEHOLDER: Difference-in-Differences Regression Results
Parallel Trends Assumption
The key identifying assumption in a DiD framework is the parallel trends assumption. Figure
PLACEHOLDER plots the trends in Outcome Y for both groups before and after the policy.
[FIGURE HERE]
Our theory of change (described) also grounds our assumption that trends would develope simi-
larly absent treatment.
Robustness Checks
To test the robustness of our results, we did some things:
1. Alternative Control Groups: Using a different set of control observations.
2. Placebo Tests: Estimating DiD on a pre-policy period.
3. Event Study Approach: Estimating treatment effects over multiple time periods.
Discussion
Our findings suggest that Policy X influenced Outcome Y (in some way). The results are robust
to (whatever robustness checks).
