DoWhy-style robustness checks, all native: a placebo (permuted) treatment should give an effect near zero; adding a random common cause or re-estimating on subsets should leave the estimate stable.
Usage
morie_dag_refute(
dag,
data,
method = c("placebo_treatment", "random_common_cause", "data_subset"),
estimator = "backdoor.linear",
n_reps = 20L,
seed = 42L
)Arguments
- dag
A
morie_dag.- data
The data frame used for estimation.
- method
One of
"placebo_treatment","random_common_cause","data_subset".- estimator
Passed through to
morie_dag_estimate().- n_reps
Number of refutation replications (default 20).
- seed
Random seed.
Examples
set.seed(1)
z <- rnorm(400); x <- rbinom(400, 1, plogis(z))
y <- 0.8 * x + z + rnorm(400)
df <- data.frame(z = z, x = x, y = y)
g <- morie_dag(c("z -> x", "z -> y", "x -> y"), "x", "y")
morie_dag_refute(g, df, method = "placebo_treatment",
estimator = "backdoor.linear", n_reps = 5)
#> $original
#> [1] 0.8599143
#>
#> $refuted
#> [1] -0.03681865
#>
#> $reps
#> [1] 0.031018816 0.141159928 -0.100545748 -0.247965756 -0.007760502
#>
#> $passed
#> [1] TRUE
#>
#> $method
#> [1] "placebo_treatment"
#>