// [[Rcpp::export]]
NumericVector attribs() {
NumericVector out = NumericVector::create(1, 2, 3);
out.names() = CharacterVector::create("a", "b", "c");
out.attr("my-attr") = "my-value";
out.attr("class") = "my-class";
return out;
}
// [[Rcpp::export]]
std::unordered_map<std::string, Node*> internal_huffman_map(NumericVector probabilities) {
std::unordered_map<std::string, Node*> map(probabilities.size());
CharacterVector items = probabilities.names();
visitNodes(internal_huffman_encode(probabilities), "", [&items, &map] (Node *n, string s) {
if (n->index >= 0) {
map[(const string) items[n->index]] = n;
}
});
return map;
}
// [[Rcpp::export]]
Environment huffman_encode(NumericVector probabilities) {
CharacterVector items = probabilities.names();
Environment env = Environment::empty_env().new_child(true);
visitNodes(internal_huffman_encode(probabilities), "", [&items, &env] (Node *n, string s) {
if (n->index >= 0) {
env.assign((const string) items[n->index], s);
}
delete n;
});
return env;
}
// [[Rcpp::export]]
NumericVector mse_compare_cpp(double b, colvec p, mat Ve,
mat Vz, int n, int n_reps){
//Function to run n_reps of the simulation study and calculate the MSE
//of various estimators
colvec ols(n_reps);
colvec tsls(n_reps);
colvec fmsc(n_reps);
colvec DHW90(n_reps);
colvec DHW95(n_reps);
colvec AVG(n_reps);
for(int i = 0; i < n_reps; i++){
dgp_OLS_IV sim_data(b, p, Ve, Vz, n);
fmsc_OLS_IV sim_results(sim_data.x, sim_data.y, sim_data.z);
ols(i) = sim_results.b_ols();
tsls(i) = sim_results.b_tsls();
fmsc(i) = sim_results.b_fmsc();
DHW90(i) = sim_results.b_DHW(0.1);
DHW95(i) = sim_results.b_DHW(0.05);
AVG(i) = sim_results.b_AVG();
}
double const trim_frac = 0; //Change this if you want trimmed MSE
double MSE_ols = MSE_trim(ols, b, trim_frac);
double MSE_tsls = MSE_trim(tsls, b, trim_frac);
double MSE_fmsc = MSE_trim(fmsc, b, trim_frac);
double MSE_DHW90 = MSE_trim(DHW90, b, trim_frac);
double MSE_DHW95 = MSE_trim(DHW95, b, trim_frac);
double MSE_star = MSE_trim(AVG, b, trim_frac);
//Create and return vector of results
NumericVector out = NumericVector::create(MSE_ols, MSE_tsls, MSE_fmsc,
MSE_star, MSE_DHW90, MSE_DHW95);
out.names() = CharacterVector::create("OLS", "TSLS", "FMSC",
"AVG", "DHW90", "DHW95");
return(out);
}
// [[Rcpp::export]]
List CIs_compare_default_cpp(double p , double r, int n,
int n_reps){
//Function to test the confidence interval code with the same parameter
//values as in mse_compare_default_cpp
//In this specification there are three instruments (z1, z2, z3)
//which ensures that the finite-sample RMSE of the 2SLS estimator exists.
//Further, the design sets
// var(x) = 1
// corr(e,v) = r
// cor(x, z1 + z2 + z3) = p
double b = 0.5;
colvec p_vec = p * ones(3);
mat Vz = eye(3, 3) / 3;
mat Ve;
Ve << 1 << r * sqrt(1 - pow(p,2)) << endr
<< r * sqrt(1 - pow(p,2)) << 1 - pow(p, 2) << endr;
//Default to nominal 90% CIs
double nominal_coverage = 0.9;
double alpha_1step = 1 - nominal_coverage;
//Same level for each interval in the 2-step procedures
double alpha_2step = alpha_1step / 2;
//Number of simulations for sim-based CIs
int n_CI_sims = 1000;
//Grid for tau
int n_tau_grid = 100;
mat OLS(n_reps, 2, fill::zeros);
mat TSLS(n_reps, 2, fill::zeros);
mat FMSC_naive(n_reps, 2, fill::zeros);
mat FMSC_1step(n_reps, 2, fill::zeros);
mat FMSC_correct(n_reps, 2, fill::zeros);
mat AVG_1step(n_reps, 2, fill::zeros);
mat AVG_correct(n_reps, 2, fill::zeros);
for(int i = 0; i < n_reps; i++){
//Simulate data and calculate estimators
dgp_OLS_IV data(b, p_vec, Ve, Vz, n);
fmsc_OLS_IV results(data.x, data.y, data.z);
//CIs that do not require draw_CI_sims()
OLS.row(i) = results.CI_ols(alpha_1step);
TSLS.row(i) = results.CI_tsls(alpha_1step);
FMSC_naive.row(i) = results.CI_fmsc_naive(alpha_1step);
//Initialization for CIs constructed using simulations
results.draw_CI_sims(n_CI_sims);
//One-step simulation-based CIs
FMSC_1step.row(i) = results.CI_fmsc_1step(alpha_1step);
AVG_1step.row(i) = results.CI_AVG_1step(alpha_1step);
//Two-step simulation-based CIs
FMSC_correct.row(i) = results.CI_fmsc_correct(alpha_2step,
alpha_2step,
n_tau_grid);
AVG_correct.row(i) = results.CI_AVG_correct(alpha_2step,
alpha_2step,
n_tau_grid);
}
CharacterVector col_names = CharacterVector::create("OLS",
"TSLS",
"FMSC_naive",
"FMSC_1step",
"FMSC_correct",
"AVG_1step",
"AVG_correct");
NumericVector cover = NumericVector::create(coverage_prob(OLS, b),
coverage_prob(TSLS, b),
coverage_prob(FMSC_naive, b),
coverage_prob(FMSC_1step, b),
coverage_prob(FMSC_correct, b),
coverage_prob(AVG_1step, b),
coverage_prob(AVG_correct, b));
cover.names() = col_names;
NumericVector width = NumericVector::create(median_width(OLS),
median_width(TSLS),
median_width(FMSC_naive),
median_width(FMSC_1step),
median_width(FMSC_correct),
median_width(AVG_1step),
median_width(AVG_correct));
width.names() = col_names;
return List::create(Named("coverage.prob") = cover,
Named("median.width") = width);
}
