netmediate

The goal of latentnetmediate is to …

Installation

You can install the development version of latentnetmediate from GitHub with:

# install.packages("devtools")
devtools::install_github("alexpghayes/latentnetmediate")

latentnetmediate's People

Contributors

Stargazers

Watchers

latentnetmediate's Issues

Handle case without edge weight attribute

https://github.com/alexpghayes/netmediate/blob/main/R/mediation.R#L231

TODO: Get HC standard errors working

estimatr::lm_robust() is not a drop-replacement for lm(), in the sense that it sometimes gives NaN for covariance estimates when lm() does not. Need to investigate this and determine a fix, but for now just planning to drop lm() back in wherever there was an lm_robust() call. As a consequence, standard errors may not be heteroscedasticity robust until I figure out how to handle this.

Finish documenting NY Leukemia data

Finish cleaning glasgow data

https://www.stats.ox.ac.uk/~snijders/siena/Glasgow_data.html -- adolescent smoking data, used in \cite{di_maria_networks_2022}

Make sure to use edge weights

attr = "weight" in igraph::as_adjacency_matrix

Give warning if cannot get embeddings for some nodes in graph

Some sort of scoping issue

Programming around nodelm is somehow broken. Messy code dump below, not anything at all close to a reprex.

library(targets)
library(netmediate)

data_block2 <- tar_read(data_block2, 1)[[1]]
mediator_block2_models <- tar_read(mediator_block2_models, 1)[[1]]

length(data_block2)

purrr::map2(data_block2, mediator_block2_models, ~fit_mediator(graph = .x, k = .y$k))




fit_mediator <- function(graph, k) {
  nodelm_robust(US(A, k) ~ . - name - 1, graph = graph)
}


fit_mediator(data_block2[[1]], mediator_block2_models[[1]]$k)
k <- 5
nodelm_robust(US(A, k) ~ . - name - 1, graph = data_block2[[1]])
fit_mediator(data_block2[[1]], k)

fit_mediator <- function(graph, k) {
  formula <- as.formula(glue::glue("US(A, {k}) ~ . - name - 1"))
  nodelm_robust(formula, graph = graph)
}

fit_mediator <- function(graph, k) {
  netmediate::US  # hack for {targets} to track dependencies in formulas
  formula <- as.formula(glue::glue("US(A, {k}) ~ . - name - 1"))
  nodelm_robust(formula, graph = graph)
}