Local R Install

When do I need local R?

For the full capability matrix between browser-only, WebR, and local R:

Try WebR Studio first Back to hub

Install R + core stack

Step 1 — base R

Download from CRAN. On Windows, also install Rtools so that packages with compiled code build correctly. Recommended R version: ≥ 4.3.

Step 2 — editor (optional but recommended)

RStudio Desktop (free) or Quarto for reproducible reports.

Step 3 — the allmeta package set

Copy this into your R console. It covers everything this hub does natively, plus the MCMC packages that only work locally:

buttonCopy
# allmeta — full CRAN stack
pkgs <- c(
  # Pairwise + random-effects
  "metafor", "meta", "metaSEM", "clubSandwich", "robumeta",
  # Network meta-analysis
  "netmeta", "gemtc", "multinma", "bnma",
  # Diagnostic test accuracy
  "mada", "CopulaDTA",
  # Publication bias / sensitivity
  "metasens", "puniform", "weightr", "dmetar",
  # Bayesian
  "bayesmeta", "brms", "baggr", "RoBMA",
  # IPD
  "ipdmeta", "ecoreg",
  # Risk of bias / GRADE
  "robvis", "metaumbrella",
  # K-M reconstruction
  "IPDfromKM",
  # Screening / data formatting
  "revtools", "synthesisr", "metagear"
)
install.packages(pkgs, dependencies = TRUE)
# MCMC back-ends
# JAGS: install from https://sourceforge.net/projects/mcmc-jags/
# Stan / rstan:
install.packages("rstan", repos = c("https://stan-dev.r-universe.dev", "https://cloud.r-project.org"))
# Or cmdstanr + CmdStan:
install.packages("cmdstanr", repos = c("https://stan-dev.r-universe.dev", "https://cloud.r-project.org"))
cmdstanr::install_cmdstan()

Reproducing each allmeta app

Every browser app in allmeta can be replicated in local R. Here are the one-liners:

Pairwise MA equivalent to MA Workbench / Forest Plot / TruthCert

library(metafor)
dat <- data.frame(yi = c(0.25, 0.18, 0.40, 0.30, 0.12),
                  sei = c(0.08, 0.10, 0.15, 0.09, 0.07))
res <- rma(yi = yi, sei = sei, data = dat, method = "REML", test = "knha")
print(res); forest(res)

NMA equivalent to NMA Pro / Bayesian NMA / Inconsistency

library(netmeta)
d <- data.frame(treat1 = c("A","A","B","A","B","C"),
                treat2 = c("B","C","C","D","D","D"),
                TE     = c(-0.30, -0.20, 0.10, -0.50, -0.25, -0.35),
                seTE   = c( 0.10,  0.12, 0.11,  0.15,  0.18,  0.14),
                studlab= paste0("S", 1:6))
nm <- netmeta(TE, seTE, treat1, treat2, studlab, data = d, reference.group = "A", random = TRUE)
print(nm); netsplit(nm); netheat(nm); netrank(nm)

Bayesian NMA equivalent to Network MA with SUCRA — real MCMC

library(gemtc)
# Requires JAGS (install separately)
net <- mtc.network(data.re = d)  # d = per-comparison effect + se
model <- mtc.model(net, type = "consistency", link = "identity", likelihood = "normal")
fit <- mtc.run(model, n.adapt = 2000, n.iter = 20000, thin = 5, n.chain = 3)
print(summary(fit)); rank.probability(fit); sucra(rank.probability(fit))

DTA equivalent to HSROC / DTA SROC

library(mada)
d <- data.frame(TP = c(85, 72, 110, 45, 92),
                FP = c( 9, 11,  18,  6, 14),
                FN = c(15, 18,  20,  8, 14),
                TN = c(91, 99, 152, 41, 80))
reit <- reitsma(d); print(summary(reit)); plot(reit)

Publication bias — Copas

library(meta); library(metasens)
m <- metagen(TE, seTE, data = d)
co <- copas(m); print(co); plot(co)

Multi-level (3-level) MA

library(metafor)
# data: study, outcome, yi, vi
res <- rma.mv(yi, V = vi, random = ~ 1 | study/outcome, data = d, method = "REML")
print(res); summary(res)

Risk of bias — RoB 2 / ROBINS-I / QUADAS-2

library(robvis)
d <- read.csv("your_rob2_assessments.csv")  # one row per study, columns per domain
rob_traffic_light(data = d, tool = "ROB2")
rob_summary(data = d, tool = "ROB2")

K-M IPD reconstruction

library(IPDfromKM)
# curve: data.frame(time, surv)
# risk:  data.frame(time, nrisk)
pre <- preprocess(dat = curve, trisk = risk$time, nrisk = risk$nrisk)
ipd <- getIPD(pre, armID = 1, tot.events = 145)
head(ipd)

MCMC back-ends

JAGS (for gemtc, rjags, R2jags, multinma)

Download from mcmc-jags. On Windows, JAGS auto-registers with R. On macOS with Homebrew: brew install jags. On Linux: apt-get install jags.

Stan (for rstan, brms, cmdstanr, bayesmeta+Stan)

cmdstanr is the modern, recommended path. After install.packages("cmdstanr") run cmdstanr::install_cmdstan() which compiles Stan in the background (5-15 minutes first time). On Windows, Rtools provides the C++ toolchain.

Verify

library(rstan)
# Tiny HMC test
fit <- stan(model_code = "parameters { real y; } model { y ~ normal(0, 1); }", iter = 200)
print(fit)

Reproducibility harness

For every allmeta result you publish, save the exact R session that produced it:

library(renv)
renv::init()           # creates a lock-file of package versions
renv::snapshot()       # call this after you finish the analysis
# distribute your script + renv.lock so others can reproduce exactly

Or use a Docker image: rocker/verse has R, RStudio, and the tidyverse pre-installed; add the allmeta stack on top.