elviscuihan / pbiv

Parametric Bayesian Instrumental Variable Methods.

Parametric Bayesian Instrumental Variable (PBIV)

This R package extends Li-Lu's PBIV estimation, which was orignally for right-censored data only, to arbitrary censoring time-to-event outcome. The PBIV package deals with causal effect estimation using instrumental variables.

Installation:

devtools::install_github("https://github.com/ElvisCuiHan/PBIV/")

There is one main function IV_MH_IC in the package. It deals with bivariate normal random errors with multiple instruments and multiple observed potential confounders within the two-stage linear model framework.

Instrumental variables regression:

library("PBIV")
IV_MH_IC(L, R, d, X, G, U = NULL, m, wid, init = NULL, prior_1, prior_2)

where

• L: $n\times1$ vector refers to left-observed time to event or censoring
• R: $n\times1$ vector refers to right-observed time to event or censoring
• d: $n\times1$ vector refers to censoring status, 4=event, 3=right-censored, 2=interval-censored, 1=left-censored
• X: $n\times1$ vector refers to the covariate of interest, i.e., the exposure using the language of randomized clinical trials (RCT)
• G: $n\times kG$ matrix if multiple instruments are used
• U: $n \times kU$ matrix if multiple observed confounders are used; the default is NULL where n is sample size, $kG$ is number of instruments, $kU$ is number of observed confounders A total of $(6+kG+2*kU)$ parameters to estimate: $a_0$, $a_1$ (length=$kG$), $a_2$ (length=$kU$), $\sigma_1^2$, $b_0$, $b_1$, $b_2$ (length=$kU$), $\sigma_2$, and $\rho$
• m: A scalr refers to the number of iterations of the MCMC algorithm
• wid: vector of the random walk width for $(a_0,a_1,a_2,\sigma_1^2,b_0,b_1,b_2,\sigma_2^2,\rho)$ in the MCMC algorithm
• init: A vector of the initial values for $(a_0,a_1,a_2,\sigma_1^2,b_0,b_1,b_2,\sigma_2^2,\rho)$, default value is NULL
• prior_1: A vector of the first parameter of the priors for $(a_0,a_1,a_2,\sigma_1^2,b_0,b_1,b_2,\sigma^2_2)$: mean of the normal priors for $a_0,a_1,a_2,b_0,b_1,b_2$; shape parameter of the inverse-gamma priors for $\sigma_1^2, \sigma_2^2$
• prior_2: A vector of the second parameter of the priors for $(a_0,a_1,a_2,\sigma_1^2,b_0,b_1,b_2,\sigma_2^2)$: variance of the normal priors for $a_0,a_1,a_2,b_0,b_1,b_2$; scale parameter of the inverse-gamma priors for $\sigma_1^2, \sigma_2^2$

The IV_MH_IC function returns a list containing MCMC traces and accept rates.