Expands a grid over DGP designs, sample sizes, covariate dimensions, and
instrument counts, then runs sim_iv_one_rep() for each replicate. Each
cell uses the same replicate seed vector so methods are compared on aligned
draws.
Usage
sim_iv_run(
designs = c("dgp1_iv_justid_linear", "dgp2_iv_homog_overid", "dgp3_iv_heterog_overid",
"dgp4_iv_weak_local", "dgp5_iv_correlated_instruments",
"dgp6_iv_nonlinear_firststage"),
n_vals = c(500L, 1000L, 2000L),
kx_vals = 2L,
q_vals = c(1L, 2L, 3L),
n_reps = 500L,
master_seed = 42L,
k = 1L,
s_fn = function(n, kx) floor(0.4 * n^((kx + 1L)/(kx + 2L))),
tuning = NULL,
methods = "dnn_iv_oracle",
alpha = 0.05,
sigma = 0.5,
clip = 0.001,
n_folds = 5L,
verbose = TRUE
)Arguments
- designs
Character vector of IV DGP names.
- n_vals
Integer vector of sample sizes.
- kx_vals
Integer vector of covariate dimensions.
- q_vals
Integer vector of requested instrument counts. The just-identified design is run once with realized
q = 1.- n_reps
Number of Monte Carlo replications per cell.
- master_seed
Seed used to generate replicate seeds.
- k
Integer DNN neighbor count.
- s_fn
Function
f(n, kx)returning kernel order. The finite-sample default is \(\lfloor 0.4 n^{(k_x+1)/(k_x+2)} \rfloor\).- tuning
Optional tuning specification. When supplied, omit
s_fn.- methods
Character vector of methods; see
sim_iv_methods().- alpha
Nominal test size.
- sigma
Noise SD forwarded to
sim_iv_data().- clip
Clipping bound forwarded to DGPs and nuisance fits.
- n_folds
Number of cross-fitting folds.
- verbose
Print one progress line per cell.
Examples
if (FALSE) { # \dontrun{
raw <- sim_iv_run(
designs = "dgp2_iv_homog_overid",
n_vals = 200,
q_vals = 2,
n_reps = 10,
methods = c("dnn_iv_oracle", "iv_oracle")
)
summarise_iv_results(raw)
} # }