Sensitivity of DiD estimate to parallel-trends violations — morie_did_sensitivity

For each \(\delta\), computes a bias-adjusted confidence set under the bound \(|\mathrm{bias}| \le \delta \hat\sigma\) (Rambachan & Roth, 2023, conservative version).

Usage

morie_did_sensitivity_analysis(
  data,
  outcome,
  treatment,
  post,
  covariates = NULL,
  delta_range = NULL,
  cluster = NULL,
  alpha = 0.05
)

Arguments

data: A data frame containing the outcome, treatment, post and any covariate columns.
outcome: Name of the outcome column.
treatment: Name of the binary (0/1) treatment-group column.
post: Name of the binary (0/1) post-period column.
covariates: Optional character vector of covariate column names.
delta_range: Numeric vector of \(\delta\) values to evaluate (default seq(0, 2, 0.25)).
cluster: Optional cluster ID column for CR1 standard errors.
alpha: Significance level for confidence intervals (default 0.05).

Value

A data frame with columns delta, ci_lower, ci_upper, covers_zero.

Details

For a relative-magnitudes bound anchored on observed event-study pre-trends (Rambachan & Roth's \(\bar M\) parameterization, conservative version) see morie_did_honest_sensitivity.

References

Rambachan, A., & Roth, J. (2023). A more credible approach to parallel trends. Review of Economic Studies, 90(5), 2555–2591.

Examples

set.seed(7)
n <- 300
d <- rbinom(n, 1, 0.5); p <- rbinom(n, 1, 0.5)
y <- 1 + 0.3 * d + 0.4 * p + 0.5 * d * p + rnorm(n, sd = 0.5)
df <- data.frame(y = y, d = d, post = p)
out <- morie_did_sensitivity_analysis(df, "y", "d", "post")
str(out, max.level = 1)
#> 'data.frame':	9 obs. of  4 variables:
#>  $ delta      : num  0 0.25 0.5 0.75 1 1.25 1.5 1.75 2
#>  $ ci_lower   : num  0.1952 0.165 0.1349 0.1047 0.0745 ...
#>  $ ci_upper   : num  0.668 0.699 0.729 0.759 0.789 ...
#>  $ covers_zero: logi  FALSE FALSE FALSE FALSE FALSE FALSE ...