Bivariate MR simulation

mr_basics.R

library(dplyr)
library(tidyr)
library(purrr)
library(broom)
library(tibble)

n <- 1000
dat <- matrix(runif(6*n),nrow=n)
colnames(dat) <- paste0("parent", rep(c("x","y"),c(3,3)), c(1:3,1:3))
dat <- as_tibble(dat)

myscale <- function(x) x / sd(x)

coin <- TRUE

if (coin) {
  dat <- dat %>%
    mutate(x = myscale(1 * parentx1 + 2 * parentx2 + 3 * parentx3 + rnorm(n,0,2)),
           y = myscale(1 * parenty1 + 2 * parenty2 + 3 * parenty3 + x + rnorm(n)))
} else {
  dat <- dat %>%
    mutate(y = myscale(1 * parenty1 + 2 * parenty2 + 3 * parenty3 + rnorm(n,0,2)),
           x = myscale(1 * parentx1 + 2 * parentx2 + 3 * parentx3 + y + rnorm(n)))
}

tidy_named_coef <- function(x,suffix) {
  setNames(tidy(x)[,1:3],
           paste0(c("term","est","se"),suffix))
}

dat %>%
  pivot_longer(cols=starts_with("parent"), names_to="variable", values_to="independent") %>%
  nest(.by = variable) %>%
  mutate(x_model = map(data, ~ lm(x ~ independent, data=.)),
         y_model = map(data, ~ lm(y ~ independent, data=.))) %>%
  mutate(x_coef = map(x_model, tidy_named_coef, "_x"),
         y_coef = map(y_model, tidy_named_coef, "_y")) %>%
  unnest(cols=c(x_coef, y_coef)) %>%
  filter(term_x == "independent") %>%
  select(variable, starts_with(c("est","se")))