standardize_treatment_hybrid: Re-weight groups to target population means

In ebenmichael/balancer: Matrix constraints for covariate balance

Re-weight groups to target population means

Description

Re-weight groups to target population means

Usage

standardize_treatment_hybrid(
  X0,
  Xtau,
  target,
  S,
  Z,
  pscores,
  kernel0 = kernlab::vanilladot(),
  lambda = 0,
  lowlim = 0,
  uplim = nrow(X0),
  scale_sample_size = F,
  data_in = NULL,
  verbose = TRUE,
  return_program = TRUE,
  init_uniform = F,
  eps_abs = 1e-05,
  eps_rel = 1e-05,
  gc,
  ...
)

Arguments

X0	n x d0 matrix of untransformed covariates defining the mean control response function
Xtau	n x dtau matrix of transformed covariates defining the mean treatment effect function
target	Vector of population means to re-weight to
S	Numeric vector of site indicators with J levels
Z	Numeric vector of treatment indicators with 2 levels
pscores	Numeric vector of propensity scores
kernel0	Kernel for control outcome covariates, default is the inner product
lambda	Regularization hyper parameter, default 0
lowlim	Lower limit on weights, default 0
uplim	Upper limit on weights, default nrow(X0)
scale_sample_size	Whether to scale the dispersion penalty by the sample size of each group, default F
data_in	Optional list containing pre-computed objective matrix/vector and constraints (without regularization term)
verbose	Whether to show messages, default T
return_program	Whether to return the objective matrix and vector and constraints, default T
init_uniform	Wheter to initialize solver with uniform weights, default F
eps_abs	Absolute error tolerance for solver
eps_rel	Relative error tolerance for solver
gc	boolean indicating whether to garbage collect between operations
...	Extra arguments for osqp solver

Value

• weights Estimated primal weights as an n x J matrix

• imbalance_0 Imbalance in covariates defining mean control response function as a d0 x J matrix

• imbalance_tau Imbalance in covariates defining mean treatment effect function as a dtau x J matrix

• data_out List containing elements of QP min 0.5 x'Px + q'x st l <= Ax <= u

  • P, q

  • constraints A, l , u

ebenmichael/balancer documentation built on Jan. 17, 2024, 2:56 p.m.