mediate_hilma {hdmed}

High-Dimensional Linear Mediation Analysis

Description

mediate_hilma applies high-dimensional linear mediation analysis (HILMA) as proposed by Zhou et al. (2020).

Usage

mediate_hilma(
  A,
  M,
  Y,
  aic_tuning = FALSE,
  nlambda = 5,
  lambda_minmax_ratio = 0.1,
  center = TRUE
)

Arguments

A

length n numeric vector containing a single exposure variable or size n x q numeric matrix containing multiple exposures. If a matrix, should be low-dimensional (q << n).

M

n x p numeric matrix of high-dimensional mediators.

Y

length n numeric vector containing continuous outcome variable.

aic_tuning

logical flag for whether to select the tuning parameter using AIC. Default is FALSE as this was not the preferred approach by the HILMA authors (see references for more detail).

nlambda

number of candidate lambdas for AIC tuning. Default is 5. If aic_tuning=F this parameter is ignored.

lambda_minmax_ratio

ratio of the minimum lambda attempted in AIC tuning to the maximum. If aic_tuning=F, ignored.

center

logical flag for whether the variables should be centered. Default is TRUE.

Details

mediate_hilma is a wrapper function for the freebird::hilma() function, which fits the "high-dimensional linear mediation analysis" model proposed by Zhou et al. (2020) for mediation settings when there are high-dimensional mediators and one or several exposures. The function returns estimates of the direct effect, total effect, and global mediation effect, the last of which is tested for statistical significance with a reported p-value. For additional detail, see the attached reference as well as the freebird::hilma() documentation.

Value

A list containing, for each exposure, a data frame of the estimated direct, total, and global mediation effects. A p-value is provided for the global mediation effect.

References

Zhou, R. R., Wang, L. & Zhao, S. D. Estimation and inference for the indirect effect in high-dimensional linear mediation models. Biometrika 107, 573-589 (2020)

Examples

A <- med_dat$A
M <- med_dat$M
Y <- med_dat$Y

# Implement HILMA with one exposure
out <- mediate_hilma(A, M, Y)
out$a1