do_combination: Workhorse function for creating a synthetic estimator by...
In denisagniel/synthate: Synthetic Causal Inference

Creates a synthetic estimator by minimizing the (estimated) mean squared error of a linear combination of multiple candidate estimators.

do_combination(
  ests,
  name_0,
  C,
  print = FALSE,
  exclude_t0 = FALSE,
  bias_type = "raw_diff",
  boot_mean = NULL,
  ate_0 = NULL,
  n = NULL
)

`ests`	one-row, p-column data frame of estimators.
`name_0`	character value of the name of the presumed unbiased estimator, θ_0. Default is NULL, which returns results for using each candidate estimator as θ_0, one synthetic estimator for each. If `ate_0` is given, then it is used as θ_0 in place of this.
`C`	p x p covariance matrix of `ests`.
`print`	logical indicating whether details should be printed. Default is FALSE.
`exclude_t0`	logical indicating whether θ_0 should be considered an external estimator (not a candidate for combining with others). Default is FALSE.
`bias_type`	method to compute the bias in the mean squard error. Default is `raw_diff`, which computes the bias as the raw difference between each of the candidate estimator and θ_0. Other options to compute the bias include: `bootstrap` which computes the bias as the difference between the mean of the bootstrap samples and the observed value of θ_0; `bootstrap_all` which computes the bias as the mean of the difference between the bootstrapped version of the candidate estimator and the bootstrapped version of θ_0; `none` which assumes no bias; `shrunk` which computes the bias as the raw difference divided by `n`.
`boot_mean`	mean of bootstrap samples for bootstrap-based bias estimation.
`ate_0`	external value of θ_0. Default is NULL, in which case θ_0 is taken to be `name_0`.
`n`	sample size. Default is NULL. Needed only if `bias_type` is `shrunk`.

list of three objects, including ate_res which gives results for the synthetic estimator, b_res which gives results for how the estimators were combined, and C which gives the covariance matrix of the estimators.

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