MASSIVE: MASSIVE algorithm (Model Assessment and Stochastic Search for...

In igbucur/MASSIVE: MASSIVE: Tractable and Robust Bayesian Learning of Many-Dimensional Instrumental Variable Models

Description

MASSIVE algorithm (Model Assessment and Stochastic Search for Instrumental Variable Estimation)

Usage

MASSIVE(
  J,
  N,
  SS,
  sigma_G,
  sd_slab,
  sd_spike,
  max_iter = 1000,
  greedy_search = parallel_greedy_search,
  Laplace_approximation = safe_Laplace_approximation,
  pruning_threshold = 0.003,
  posterior_samples = 10000,
  ...
)

Arguments

J	Integer number of candidate instruments.
N	Integer number of observations.
SS	Numeric matrix containing first- and second-order statistics.
sigma_G	Numeric vector of instrument standard deviations.
sd_slab	Numeric scale parameter of slab component.
sd_spike	Numeric scale parameter of spike component.
max_iter	Maximum number of stochastic search steps.
greedy_search	Function for initial greedy search.
Laplace_approximation	Function used to compute Laplace approximation of IV model.
pruning_threshold	Numeric threshold for pruning approximated IV models. Models with probability less that threshold are pruned out.
posterior_samples	Integer number of posterior samples to generate.
...	Extra arguments for greedy_search, Laplace_approximations and find_causal_models.

Value

List of explored IV model approximations and their evidences

Examples

set.seed(2020)
J <- 10
N <- 10000
G <- matrix(rbinom(N * J, 2, 0.3), N, J)
U <- rnorm(N)
X <- G %*% runif(J, 0.3, 0.5) + U + rnorm(N)
Y <- G[, 1:5] %*% runif(5, 0.1, 0.3) + X + U + rnorm(N)

Z <- cbind(1, G, X, Y)
SS <- t(Z) %*% Z / N
sigma_G <- apply(G, 2, sd)

samples <- MASSIVE::MASSIVE(J, N, SS, sigma_G, sd_slab = 1, sd_spike = 0.01, max_iter = 1000)
plot(density(samples$betas))
median(samples$betas)

igbucur/MASSIVE documentation built on Oct. 26, 2020, 1:26 a.m.