Causal Analysis using R The Difference in Difference Method

Readme.txt

This R script provides a step-by-step walkthrough of how to use R to apply the Difference-in-Differences (DiD) method for causal analysis. It explains the purpose, key assumptions, and interpretation of the DiD approach using the DiD package in R.
A step by step video demonstration can be found at: https://youtu.be/BNBv0pE5owc 

R Script

### Causal Analysis using R  The Difference in Difference Method

##DiD estimates causal effects by comparing changes over time between treatment and control groups.

##It uses before-and-after differences in outcomes to isolate the treatment effect.

##The critical assumption is parallel trends, meaning both groups would follow similar outcome paths 
#if untreated.

##Suitable for observational data where randomization is not possible.

##Controls for unobserved fixed differences between groups but is sensitive to violations of assumptions.

##Widely applied for policy evaluation and program impact analysis.

# Install and load the did package if not already installed
if (!requireNamespace("did", quietly = TRUE)) {
  install.packages("did")
}
library(did)

# Example: Create a panel dataset with multiple groups and time periods
set.seed(11092025)
n <- 100  # individuals per group
time_periods <- 4  # multiple time periods
groups <- 2  # treatment and control groups

id <- rep(1:n, groups * time_periods)
group <- rep(rep(0:1, each = n), time_periods)  # control = 0, treated = 1 (treated starting after period 2)
period <- rep(1:time_periods, each = n * groups)

# Create treatment timing variable: treated group gets treatment starting period 3
G <- ifelse(group == 1, 3, 0)  

# Simulate outcome with group, time, and treatment effects
Y <- 5 + 2*group + 1.5*period + 4*(period >= G & G > 0) + rnorm(n * groups * time_periods)

data_did <- data.frame(id = id, period = period, group = group, G = G, Y = Y)
head(data_did)
# Run DID estimation using att_gt (group-time average treatment effects)
#att_gt() stands for "average treatment effect on the treated." More specifically, 
#within the att_gt() output, "att" refers to the group-time average treatment effect
#—i.e., the estimated causal impact of the treatment for each group gat each time period t
#  after the treatment begins for that group.
att_gt_res <- att_gt(
  yname = "Y",
  tname = "period",
  idname = "id",
  gname = "G",
  data = data_did,
  control_group = "notyettreated",  # uses not-yet treated as control for staggered adoption
  est_method = "dr",  # doubly robust estimation
  panel = FALSE
)

summary(att_gt_res)

##HoestDiD https://cran.r-project.org/web/packages/HonestDiD/index.html 

# Aggregate results by type: dynamic effects over time since treatment
# "dynamic" aggregation computes the average treatment effects 
# relative to the length of exposure to treatment (event time).
agg_dynamic <- aggte(att_gt_res, type = "dynamic")
summary(agg_dynamic)

# Aggregate results by group
# aggregates the detailed group-time average treatment effects (ATTs) 
# estimated by att_gt() by groups defined by their treatment timing.
agg_group <- aggte(att_gt_res, type = "group")
summary(agg_group)

# Aggregate results by calendar time 
# Calendar time is real world clock time in the calendar such as year, meonth or day
# Event time is relative to a specific event or intervention such as 0 period after intervention
# 1 period after the intervention, etc.

agg_calendar <- aggte(att_gt_res, type = "calendar")
summary(agg_calendar)

##All ATT estimates are the same in this example.
## They may be different if there are heterogenous treatment effects across group and time
## or different group sizes, etc.

VideoDemo

A step by step video demonstration can be found at: https://youtu.be/BNBv0pE5owc 

A step by step video demonstration can be found at: https://youtu.be/BNBv0pE5owc