mouthwash_second_step: The second step of MOUTHWASH.
In dcgerard/vicar: Various Ideas for Confounder Adjustment in Regression

Description Usage Arguments Author(s) References

The second step of MOUTHWASH.

mouthwash_second_step(
  betahat_ols,
  S_diag,
  alpha_tilde,
  lambda_seq,
  tau2_seq = NULL,
  a_seq = NULL,
  b_seq = NULL,
  mixing_dist = c("normal", "uniform", "+uniform", "sym_uniform"),
  likelihood = c("normal", "t"),
  pi_init_type = c("zero_conc", "uniform", "random"),
  scale_var = TRUE,
  degrees_freedom = NULL,
  plot_update = FALSE,
  sprop = 0,
  var_inflate_pen = 0,
  verbose = TRUE
)

`betahat_ols`	A vector of numerics. The OLS estimates of the coefficients of interest.
`S_diag`	A vector of positive numerics. The estimated standard errors.
`alpha_tilde`	A matrix. The number of rows should be equal the length of betahat_ols. The number of columns should equal the number of hidden confounders.
`lambda_seq`	A numeric vector with elements all greater than or equal to 1. These are the tuning parameters for the mixing proportions.
`tau2_seq`	The grid of variances of the mixing distributions if `mixing_dist = "normal"`. Only one of `tau2_seq` or `a_seq` and `b_seq` need be specified.
`a_seq`	The grid of lower bounds for the uniforms if `mixing_dist` is one of the uniforms.
`b_seq`	The grid of upper bounds for the uniforms if `mixing_dist` is one of the uniforms.
`mixing_dist`	A character. Should we use a mixture of uniforms (`"uniform"`), a mixture of uniforms with minimum at 0 (`"+uniform"`), a mixture of uniforms symmetric at 0 (`"sym_uniform"`), or a mixture of normals (`"normal"`)?
`likelihood`	Either `"normal"` or `"t"`. If `likelihood = "t"`, then the user may provide the degrees of freedom via `degrees_freedom`.
`pi_init_type`	How should we initialize the mixing proportions? By concentrating on zero (`"zero_conc"`), by equal weights on all mixing distributions (`"uniform"`), or by sampling uniformly on the simplex (`"random"`)?
`scale_var`	A logical. Should we estimate a variance inflation parameter (`TRUE`) or not (`FALSE`)?
`degrees_freedom`	The degrees of freedom of the t-distribution if `likelihood = "t"`.
`plot_update`	A logical. Should I plot the the path of the log-likelihood (`TRUE`) or not (`FALSE`)? Only applicable when `mixing_dist` is not `"normal"`.
`sprop`	If b is an effect and s is an estimated standard error, then we model b/s^{sprop} as exchangeable. The default is 0. When `sprop = 1`, for identifiability reasons it must be the case that `scale_var = FALSE`.
`var_inflate_pen`	The penalty to apply on the variance inflation parameter. Defaults to 0, but should be something non-zero when `alpha = 1` and `scale_var = TRUE`.
`verbose`	If `verbose = TRUE`, print progress of the algorithm to the console.

David Gerard

Gerard, D., and Stephens, M. 2020. "Empirical Bayes shrinkage and false discovery rate estimation, allowing for unwanted variation", Biostatistics, 21(1), 15-32 doi: 10.1093/biostatistics/kxy029

dcgerard/vicar documentation built on July 7, 2021, 1:08 p.m.