"Shy Trump" list experiment design declaration

list_experiment_declaration.r

library(DeclareDesign)
proportion_shy <- .06

population <- declare_population(
  N = 5000,
  # true trump vote (unobservable)
  truthful_trump_vote = draw_binary(.45, N),
  
  # shy voter (unobservable)  
  shy = draw_binary(proportion_shy, N),
  
  # Direct question response (1 if Trump supporter and not shy, 0 otherwise)
  Y_direct = as.numeric(truthful_trump_vote == 1 & shy == 0),
  
  # Nonsensitive list experiment items
  raise_minimum_wage = draw_binary(.8, N),
  repeal_obamacare = draw_binary(.6, N),
  ban_assault_weapons = draw_binary(.5, N)
)

potential_outcomes <- declare_potential_outcomes(
  Y_list_Z_0 = raise_minimum_wage + repeal_obamacare + ban_assault_weapons,
  Y_list_Z_1 = Y_list_Z_0 + truthful_trump_vote
)

# Inquiry -----------------------------------------------------------------
estimand <- declare_estimand(
  proportion_truthful_trump_vote = mean(truthful_trump_vote))

# Data Strategy -----------------------------------------------------------
sampling <- declare_sampling(n = 500)
assignment <- declare_assignment(prob = .5)

# Answer Strategy ---------------------------------------------------------
estimator_direct <- declare_estimator(
  Y_direct ~ 1,
  model = lm_robust,
  term = "(Intercept)",
  estimand = estimand,
  label = "direct"
)

estimator_list <- declare_estimator(Y_list ~ Z,
                                    model = difference_in_means,
                                    estimand = estimand,
                                    label = "list")

# Design ------------------------------------------------------------------
design <- 
  population + 
  potential_outcomes + 
  sampling + 
  estimand + 
  assignment + 
  declare_reveal(outcome_variables = Y_list) + 
  estimator_direct + 
  estimator_list

draw_data(design)

diagnosis <- diagnose_design(design, sims = 500, bootstrap_sims = FALSE)
diagnosis