create_projection: Projection matrix (or list of projection matrices)

In umich-biostatistics/AdaptiveBayesianUpdates: Adaptive Bayesian Updates

Projection matrix (or list of projection matrices)

Description

Helper function to make projection matrix (or list of projection matrices), which is P in the notation of Boonstra and Barbaro.

Usage

create_projection(
  x_curr_orig,
  x_curr_aug,
  eigenvec_hist_var,
  imputes_list = list(c(1, 500)),
  seed_start = sample(.Machine$integer.max, 1),
  predictorMatrix = NULL
)

Arguments

x_curr_orig	(matrices) matrices with equal numbers of rows and p & q columns, respectively. These are used to estimate the joint association between the original and augmented covariates, which is needed for the imputation model
x_curr_aug	(matrices) matrices with equal numbers of rows and p & q columns, respectively. These are used to estimate the joint association between the original and augmented covariates, which is needed for the imputation model
eigenvec_hist_var	(matrix) pxp matrix with each row corresponding to an eigenvector. This is v_0 in Boonstra and Barbaro.
imputes_list	(list) list of length-2 vectors, with each vector containing the lower and upper indices of the imputations to use for a projection matrix in the SAB method. This is best explained with an example: if imputes_list = list(c(1,15),c(16,100),c(1,100)), then three projection matrices will be returned. One will be based upon the first 15 imputations from a run of MICE, the second based upon the last 85 imputations from that same run (i.e. the 16th to100th imputations), and the third will be based upon all 100 imputations from this same run. This is coded as such to allow for flexible exploration of the impact of number of imputations or variability due to imputations.
seed_start	(pos. integer) random seed to start each imputation
predictorMatrix	(matrix) (p+q)x(p+q) matrix equivalent to argument of the same name in in the 'mice()' function (type '?mice'). It is specially calculated based upon a monotone missingness pattern (x^o is fully observed, x^a is not) Thus it samples from

Value

A list as long as the the length of imputes_list, with each element containing a different projection matrix using the indices of the imputations specified in the corresponding element of imputes_list.

umich-biostatistics/AdaptiveBayesianUpdates documentation built on May 21, 2024, 5:29 a.m.