synthetic_split_subset: Compute synthetic estimator on a subset of candidate...

In denisagniel/synthate: Synthetic Causal Inference

Description

This function allows you to compute a set of split-sample candidate estimators and bootstrap estimates once and then compute on it many times. It's designed to work well with map2.

Usage

synthetic_split_subset(thetahat_a, thetahat_b, boot, estimators = NULL,
  ...)

Arguments

thetahat_a	one-row data frame of candidate estimators fit on one half of the sample.
thetahat_b	one-row data frame of candidate estimators fit on the other half of the sample.
boot	data frame of bootstrapped versions of candidate estimators.
estimators	vector of estimator names to be included in synthetic estimator; if NULL (the default) all estimators are used.
...	additional arguments to be passed to combine_estimators.

Value

vector of synthetic estimators

Examples

ate_list <- list(
ipw1 = ipw1_ate,
ipw2 = ipw2_ate,
ipw3 = ipw3_ate,
strat = strat_ate,
strat_regr = strat_regr_ate,
match_ps = match_ps_ate,
match_prog = match_prog_ate,
match_both = match_both_ate,
cal_ps = caliper_ps_ate,
cal_both = caliper_both_ate,
regr = regr_ate,
dr = dr_ate,
bal = bal_ate,
hdbal = highdim_bal_ate
)
gen_mod <- generate_data(n = 100, 
                         dgp = 'ks', 
                         correct_outcome = FALSE,
                         correct_ps = TRUE)
this_data <- gen_mod$data
this_data <- with(gen_mod, estimate_scores(this_data, outcome_fm = outcome_fm,
                             ps_fm = ps_fm,
                             ps_fam = ps_fam,
                             outcome_fam = outcome_fam))
thetahat <- with(gen_mod,
                 estimate_ates(this_data,
                          ate_list,
                          cov_ids = cov_ids,
                          outcome_fm = stringr::str_c('d + ', outcome_fm),
                          outcome_fam = outcome_fam))
                          predict_delta <- function(d) {
                          gen_mod$true_ate
                          }
 resample_thetas <- with(gen_mod, resample_fn(dat = this_data,
                                dpredfn = predict_delta,
                                B = 50,
                                ate_list = ate_list,
                                outcome_fm = outcome_fm,
                                ps_fm = ps_fm,
                                ps_fam = ps_fam,
                                outcome_fam = outcome_fam,
                                cov_ids = cov_ids))
 boot_theta <- resample_thetas[[1]]
 
 synthetic_subset(thetahat, boot_theta,
                  estimators = c('ate_regr', 'ate_ipw_2', 'ate_dr', 'ate_bal'),
                  name_0 = 'ate_dr')
 synthetic_subset(thetahat, boot_theta,
                  estimators = c('ate_regr', 'ate_ipw_2', 'ate_dr', 'match_ps'),
                  ate_0 = thetahat$ate_bal,
                  name_0 = 'ate_dr')
 
 out_df <- tibble::tibble(
 thetahat = list(thetahat),
 boot_theta = list(boot_theta))
 
 
 ## 
 
 dplyr::mutate(out_df, 
 theta_s = unlist(purrr::map2(
    thetahat, 
    boot_theta, 
    synthetic_subset, 
    estimators = c('ate_regr', 'ate_ipw_2', 'ate_dr', 'ate_bal'),
    name_0 = 'ate_dr')))
    
    

denisagniel/synthate documentation built on April 16, 2020, 12:45 a.m.