Package 'freebird'

Title: Estimation and Inference for High Dimensional Mediation and Surrogate Analysis
Description: Estimates and provides inference for quantities that assess high dimensional mediation and potential surrogate markers including the direct effect of treatment, indirect effect of treatment, and the proportion of treatment effect explained by a surrogate/mediator; details are described in Zhou et al (2022) <doi:10.1002/sim.9352> and Zhou et al (2020) <doi:10.1093/biomet/asaa016>. This package relies on the optimization software 'MOSEK', <https://www.mosek.com>.
Authors: Ruixuan Zhou [aut, cph], Dave Zhao [aut, cph], Layla Parast [cre]
Maintainer: Layla Parast <[email protected]>
License: GPL (>= 2)
Version: 1.0
Built: 2024-12-21 06:41:55 UTC
Source: CRAN

Help Index

Estimation and Inference for High Dimensional Mediation Analysis

Description

This function implements the estimation and inference for the indirect effect in high dimensional linear mediation analysis models. It provides estimates and p-values under both incomplete mediation, where a direct effect may exist, as well as complete mediation, where the direct effect is known to be absent.

Usage

hilma(
  Y,
  G,
  S,
  mediation_setting = "incomplete",
  tuning_method = "uniform",
  lam_list = NA,
  min.ratio = 0.1,
  n.lambda = 5,
  center = TRUE
)

Arguments

Y

The n-dimensional outcome vector.
G

The n by p mediator matrix. p can be larger than n.
S

The n by q exposure matrix. q can be 1, and q < n is required.
mediation_setting

Either ‘incomplete’ or ‘complete’
tuning_method

‘uniform’ or ‘aic’, the default is ‘uniform’
lam_list

tuning parameter for uniform tuning or list of tuning parameter for aic tuning
min.ratio

the ratio of the minimum lambda to the maximum
n.lambda

number of tuning parameters to choose from
center

center the data or not, the default is TRUE

Value

A list with components:

beta_hat

estimated indirect effect
alpha1_hat

estimated direct effect
pvalue_beta_hat

the p value for testing the significance of the indirect effect
lambda_used

lambda used during optimization

Author(s)

Ruixuan Zhou

Examples

n = 30
p = 50
q = 2
G = MASS::mvrnorm(n, rep(0,p), diag(p))
S = as.matrix(MASS::mvrnorm(n, rep(0,q), diag(q)))
Y = as.matrix(rnorm(n))
out = hilma(Y,G,S, mediation_setting = 'complete', tuning_method = 'uniform', lam_list = 0.2)
out

Proportion of treatment effect explained by high-dimensional surrogates

Description

Estimates the proportion of the treatment effect explained by the indirect effect via high-dimensional surrogates.

Usage

ptehd(Yt, Yc, St, Sc, lambda_range = c(0, 1))

Arguments

Yt

The n-dmensional outcome vector in the treatment group.
Yc

The n-dmensional outcome vector in the control group.
St

The n x p matrix of surrogates in the treatment group.
Sc

The n x p matrix of surrogates in the treatment group.
lambda_range

Min and max of range of range of tuning parameter to use during the constrained l1 optimization step.

Value

A list with components:

est_id

Estimate of indirect effect, defined as ∫E(Y∣S=s,Z=1)dF(s∣Z=1)−∫E(Y∣S=s,Z=0)dF(s∣Z=0)\int E(Y | S = s, Z = 1) dF(s | Z = 1) - \int E(Y | S = s, Z = 0) dF(s | Z = 0)
sd_id

Standard deviation of indirect effect estimate
est_total

Estimate of total effect
sd_total

Standard deviation of total effect estimate
V

Covariance matrix of (est_id, est_total)
est_R

Estimate of proportion of treatment effect explained by surrogates
sd_R

Standard deviation of proportion estimate
lambda_used

lambda used during optimization

Author(s)

Ruixuan Zhou

Examples

n = 10
St = replicate(n, rnorm(20, mean = 1))
Sc = replicate(n, rnorm(20))
Yt = 1 + rowSums(St) / 2 + rnorm(n)
Yc = rowSums(Sc) / 3 + rnorm(n)
# Requires installation of mosek to run
## Not run: 
out = ptehd(Yt, Yc, St, Sc)

## End(Not run)