Package 'rdstagger'

rdstagger: Staggered Regression Discontinuity with Network Interference

Description

Implements a unified framework combining staggered difference-in-differences with regression discontinuity designs and network interference. Extends Callaway and Sant'Anna (2021) to settings where:

1. Treatment assignment is determined by a running variable crossing a cutoff (RD)

2. Treatment adoption timing is heterogeneous across units (staggered DiD)

3. Spillover effects operate through a known network structure (interference)

Main functions

sim_rdstagger

Simulate a staggered RD panel dataset with interference

rdstagger_bw

Optimal bandwidth selection per cohort-time cell

rdstagger_attgt

Estimate ATT(g,t) — direct and spillover effects

rdstagger_spillover

Estimate spillover effects at network distance d

rdstagger_agg

Aggregate ATT(g,t) into event-study or overall ATT

rdstagger_pretest

Pre-treatment parallel trends falsification tests

References

Callaway, B., & Sant'Anna, P. H. C. (2021). Difference-in-differences with multiple time periods. Journal of Econometrics, 225(2), 200-230.

Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014). Robust nonparametric confidence intervals for regression-discontinuity designs. Econometrica, 82(6), 2295-2326.

Manski, C. F. (2013). Identification of treatment response with social interactions. The Econometrics Journal, 16(1), S1-S23.

Author(s)

Maintainer: Subir Hait [email protected]

See Also

Useful links:

• https://github.com/causalfragility-lab/rdstagger

• Report bugs at https://github.com/causalfragility-lab/rdstagger/issues

---

Plot Aggregated ATT Estimates

Description

Produces a ggplot2 event-study or aggregation plot from an "rdstagger_agg" object.

Usage

## S3 method for class 'rdstagger_agg'
plot(x, ...)

Arguments

x	An object of class "rdstagger_agg".
...	Additional arguments (currently unused).

Value

A ggplot2 object.

Examples

sim <- sim_rdstagger(n = 300, nperiods = 6, n_cohorts = 2,
                     true_direct = 0.3, seed = 42)
res <- rdstagger_attgt(data = sim$data, yname = "y", xname = "x",
                       gname = "g", tname = "period", idname = "id",
                       bw = 1.5, boot = FALSE)
agg <- rdstagger_agg(res, type = "dynamic")
plot(agg)

---

Aggregate ATT(g,t) Estimates

Description

Aggregates group-time average treatment effects ATT(g,t) from rdstagger_attgt into summary estimands: event-study (dynamic), cohort-level, calendar-time, or overall ATT.

Usage

rdstagger_agg(
  x,
  type = c("dynamic", "group", "calendar", "overall"),
  min_periods = 1L
)

Arguments

x	An object of class "rdstagger_attgt" from rdstagger_attgt.
type	Character. Aggregation type: "dynamic" (event-study), "group" (by cohort), "calendar" (by calendar period), or "overall". Default "dynamic".
min_periods	Integer. Minimum number of cohort-time cells required to include an event-time bin. Default 1.

Value

An object of class "rdstagger_agg", a list with:

agg

Data frame of aggregated estimates

type

Aggregation type used

overall_att

Simple overall ATT (post-treatment average)

attgt

The original ATT(g,t) data frame

Examples

sim <- sim_rdstagger(n = 300, nperiods = 6, n_cohorts = 2,
                     true_direct = 0.3, seed = 42)
res <- rdstagger_attgt(data = sim$data, yname = "y", xname = "x",
                       gname = "g", tname = "period", idname = "id",
                       bw = 1.5, boot = FALSE)

# Event study
agg_dyn <- rdstagger_agg(res, type = "dynamic")
print(agg_dyn)
plot(agg_dyn)

# Overall ATT
agg_ov <- rdstagger_agg(res, type = "overall")
print(agg_ov)

---

Estimate Group-Time Average Treatment Effects in Staggered RD

Description

Main estimation function for the staggered RD framework with network interference. Estimates ATT(g, t) — the average treatment effect for cohort $g$ at time $t$ — separately for direct effects on treated units and spillover effects on their network neighbors, within an RD bandwidth around the cutoff.

Usage

rdstagger_attgt(
  data,
  yname,
  xname,
  cutoff = 0,
  gname,
  tname,
  idname,
  network = NULL,
  bw = "optimal",
  control_group = c("nevertreated", "notyetreated"),
  xformla = NULL,
  doubly_robust = TRUE,
  boot = TRUE,
  nboot = 999L,
  alpha = 0.05,
  kernel = c("triangular", "epanechnikov", "uniform")
)

Arguments

data	A data.frame in long (panel) format.
yname	Character. Outcome variable name.
xname	Character. Running variable name.
cutoff	Numeric. RD cutoff. Default 0.
gname	Character. Cohort variable name (Inf/NA for never-treated).
tname	Character. Time period variable name.
idname	Character. Unit identifier variable name.
network	Matrix or NULL. $n \times n$ adjacency matrix for spillover estimation. If NULL, spillovers are not estimated.
bw	Numeric or "optimal". Bandwidth around the cutoff. If "optimal", calls rdstagger_bw internally. Default "optimal".
control_group	Character. Which units form the control group. "nevertreated" (default) or "notyetreated".
xformla	Formula or NULL. Covariate formula for outcome regression (e.g., ~ age + income). Default NULL.
doubly_robust	Logical. Use doubly-robust estimator. Default TRUE.
boot	Logical. Compute bootstrap standard errors. Default TRUE.
nboot	Integer. Number of bootstrap replications. Default 999.
alpha	Numeric. Significance level for confidence intervals. Default 0.05.
kernel	Character. RD kernel. Default "triangular".

Value

An object of class "rdstagger_attgt", a list with:

attgt

Data frame of ATT(g,t) estimates (direct effects)

spillgt

Data frame of spillover ATT(g,t) estimates (if network supplied)

args

List of call arguments

bandwidth

Bandwidth used

Examples

sim <- sim_rdstagger(n = 300, nperiods = 6, n_cohorts = 2,
                     true_direct = 0.3, true_spill = 0.1, seed = 42)

res <- rdstagger_attgt(
  data    = sim$data,
  yname   = "y",
  xname   = "x",
  cutoff  = 0,
  gname   = "g",
  tname   = "period",
  idname  = "id",
  network = sim$network,
  bw      = 1.5,
  boot    = FALSE
)
print(res)

---

Bandwidth Selection for Staggered RD

Description

Computes an optimal bandwidth for each cohort-time cell using the mean-squared-error-optimal bandwidth selector from rdrobust. Separate bandwidths are estimated for pre-treatment and post-treatment periods to ensure appropriate comparison groups.

Usage

rdstagger_bw(
  data,
  yname,
  xname,
  cutoff = 0,
  gname,
  tname,
  kernel = c("triangular", "epanechnikov", "uniform"),
  bw_common = FALSE
)

Arguments

data	A data.frame containing the panel data.
yname	Character. Name of the outcome variable column.
xname	Character. Name of the running variable column.
cutoff	Numeric. The RD cutoff value. Default 0.
gname	Character. Name of the cohort variable column (Inf or NA for never-treated units).
tname	Character. Name of the time period column.
kernel	Character. Kernel for RD estimation. One of "triangular" (default), "epanechnikov", or "uniform".
bw_common	Logical. If TRUE, returns a single common bandwidth across all cohort-time cells (the median of cell-specific bandwidths). Default FALSE.

Value

A list with elements:

bw_matrix

A matrix of bandwidths with rows = cohorts, columns = time periods

bw_common

Single common bandwidth (median across cells)

bw_summary

A data.frame summarising bandwidths by cohort and period

Examples

sim <- sim_rdstagger(n = 400, nperiods = 6, n_cohorts = 2, seed = 42)
bw  <- rdstagger_bw(data = sim$data, yname = "y", xname = "x",
                    cutoff = 0, gname = "g", tname = "period")
bw$bw_common
bw$bw_summary

---

Pre-Treatment Falsification Tests for Staggered RD

Description

Tests the pre-treatment parallel trends assumption within the RD bandwidth. Performs a joint test across all pre-treatment cohort-time cells and individual cell tests, analogous to pretest in the did package but adapted for the staggered RD setting.

Usage

rdstagger_pretest(x, method = c("joint", "individual", "both"))

Arguments

x	An object of class "rdstagger_attgt".
method	Character. Test method: "joint" (chi-squared joint test, default), "individual" (t-tests per cell), or "both".

Value

A list with elements:

joint

Joint test statistic, df, and p-value (if requested)

individual

Data frame of per-cell tests (if requested)

passes

Logical. TRUE if joint test p-value > 0.05

Examples

sim <- sim_rdstagger(n = 400, nperiods = 8, n_cohorts = 2,
                     true_direct = 0.3, seed = 42)
res <- rdstagger_attgt(data = sim$data, yname = "y", xname = "x",
                       gname = "g", tname = "period", idname = "id",
                       bw = 1.5, boot = FALSE)
pt  <- rdstagger_pretest(res)
print(pt)

---

Estimate Spillover Effects in Staggered RD

Description

Estimates the spillover (indirect) treatment effects on network neighbors of treated units within the RD bandwidth. Spillover effects are estimated separately for each cohort-time cell.

Usage

rdstagger_spillover(
  data,
  yname,
  xname,
  cutoff = 0,
  gname,
  tname,
  idname,
  network,
  bw,
  kernel = c("triangular", "epanechnikov", "uniform"),
  boot = TRUE,
  nboot = 999L,
  alpha = 0.05
)

Arguments

data	A data.frame in long panel format.
yname	Character. Outcome variable name.
xname	Character. Running variable name.
cutoff	Numeric. RD cutoff. Default 0.
gname	Character. Cohort variable name.
tname	Character. Time period variable name.
idname	Character. Unit identifier variable name.
network	Matrix. $n \times n$ adjacency matrix.
bw	Numeric. Bandwidth around the cutoff.
kernel	Character. RD kernel. Default "triangular".
boot	Logical. Bootstrap standard errors. Default TRUE.
nboot	Integer. Bootstrap replications. Default 999.
alpha	Numeric. Significance level. Default 0.05.

Value

A data.frame with columns:

cohort

Treatment cohort

period

Time period

spill_att

Spillover ATT estimate

se

Standard error

ci_lower, ci_upper

Confidence interval

pval

p-value

n_exposed

Number of exposed neighbors

Examples

sim <- sim_rdstagger(n = 300, nperiods = 6, n_cohorts = 2,
                     true_direct = 0.3, true_spill = 0.15, seed = 42)
sp  <- rdstagger_spillover(
  data    = sim$data,  yname = "y",   xname = "x",
  gname   = "g",       tname = "period", idname = "id",
  network = sim$network, bw = 1.5, boot = FALSE
)
head(sp)

---

Simulate a Staggered RD Panel Dataset with Network Interference

Description

Generates synthetic panel data suitable for testing and demonstrating the rdstagger estimators. The data generating process features a running variable with a cutoff-based treatment assignment, staggered adoption across cohorts, and network spillover effects.

Usage

sim_rdstagger(
  n = 500,
  nperiods = 8,
  n_cohorts = 3,
  cutoff = 0,
  bw = 1,
  network_density = 0.1,
  true_direct = 0.3,
  true_spill = 0.1,
  outcome_type = c("continuous", "binary", "count"),
  heterogeneous_te = FALSE,
  seed = NULL
)

Arguments

n	Integer. Number of units. Default 500.
nperiods	Integer. Number of time periods. Default 8.
n_cohorts	Integer. Number of treatment cohorts. Default 3.
cutoff	Numeric. RD cutoff value on the running variable. Default 0.
bw	Numeric. True bandwidth around the cutoff. Default 1.
network_density	Numeric. Probability of a network tie between any two units (Erdos-Renyi model). Must be in (0, 1). Default 0.1.
true_direct	Numeric. True direct average treatment effect. Default 0.3.
true_spill	Numeric. True spillover effect on network neighbors. Default 0.1.
outcome_type	Character. One of "continuous", "binary", or "count". Default "continuous".
heterogeneous_te	Logical. If TRUE, treatment effects vary by cohort. Default FALSE.
seed	Integer. Random seed for reproducibility. Default NULL.

Value

A list with three elements:

data

A data.frame with columns: id, period, y, x (running variable), g (cohort, Inf for never-treated), treated, neighbor_treated, spillover_share

network

An $n \times n$ adjacency matrix

true_params

A list of the true parameter values used to generate the data

Examples

# Basic continuous outcome
sim <- sim_rdstagger(n = 300, nperiods = 6, n_cohorts = 2,
                     true_direct = 0.3, true_spill = 0.1, seed = 42)
head(sim$data)
sim$true_params

# Binary outcome
sim_bin <- sim_rdstagger(n = 500, nperiods = 8, n_cohorts = 3,
                         outcome_type = "binary", seed = 123)
table(sim_bin$data$y)

# Count outcome
sim_cnt <- sim_rdstagger(n = 400, nperiods = 6, n_cohorts = 2,
                         outcome_type = "count", true_direct = 0.5,
                         seed = 999)

---