R package downloads: reality, proxies, and better metrics

rprogramming
Author

Maciej Nasinski

Published

October 10, 2025

TL;DR

  • There is no official, global count of CRAN package downloads.
  • Popular services (e.g. cranlogs badges) measure only the RStudio/Posit CRAN mirror logs (http://cran.rstudio.com / http://cran-logs.rstudio.com).
  • If you install from other mirrors, corporate repos, or Posit Package Manager, those downloads aren’t counted.
  • Treat cranlogs numbers as a convenient proxy, not ground truth.
  • There is most probably around 200-300 downloads a month on Posit/RStudio CRAN mirror which comes from servers rather than users. Read more about it the in ‘ThreeWiseMonkeys’ R package README file.
  • For organizations, use your own repository logs (Posit Package Manager metrics) for accurate internal usage.

The problem: “CRAN downloads” don’t exist (globally)

CRAN is a network of mirrors. There’s no single authoritative counter across all mirrors, binary caches, containers, or corporate pipelines. What’s widely used instead:

These are useful, but partial. If you’re not using http://cran.rstudio.com mirror (e.g., other GUIs, corporate pipelines, or Posit Public/Private Package Manager), your downloads won’t be captured.

What cranlogs measures (and what it misses)

Measures:

Downloads observed on the RStudio/Posit CRAN mirror, aggregated per day/package.

FYI. The is a problem of ghost downloads on Posit CRAN mirror which was recognized by ‘ThreeWiseMonkeys’ R package author. ‘ThreeWiseMonkeys’ was created to recogize the servers baseline downloads on Posit CRAN mirror. By using ‘ThreeWiseMonkeys’ as reference we can estimate that there is around 200-300 downloads a month which not comes from RStudio users but rather servers.

Misses:

  • Other CRAN mirrors (cloud.r-project.org and regional mirrors)
  • Corporate repos (Posit Package Manager, Artifactory, Nexus)
  • CI/CD caches, Docker layers, offline installs, shared binaries
  • Reinstalls from local caches (pak, renv)

Interpretation:

  • Use cranlogs as a trend indicator and rough popularity proxy.
  • Don’t equate cranlogs counts with “true installs” or “active users.”

Getting proxy counts with cranlogs (R)

library(cranlogs)
library(dplyr)
library(ggplot2)

pkg <- "yourpkg"   # change to your package name
from <- "2024-01-01"
to <- "2025-10-01"

dl <- cran_downloads(packages = pkg, from = from, to = to)

# Quick summary
dl %>% 
  summarize(total = sum(count), days = n(), avg_per_day = mean(count))

# Visualize trend
ggplot(dl, aes(date, count)) +
  geom_col(fill = "#2C7FB8") +
  labs(
    title = paste0("cranlogs downloads (", pkg, ")"),
    subtitle = "Proxy counts from the RStudio/Posit CRAN mirror only",
    x = NULL, y = "Downloads/day",
    caption = "Source: cranlogs (cran.rstudio.com mirror)"
  ) +
  theme_minimal(base_size = 12)

You can also call the API directly:

# API docs: https://cranlogs.r-pkg.org/
# Example: total downloads over a date range
library(httr2)
pkg <- "yourpkg"
resp <- request(
  paste0("https://cranlogs.r-pkg.org/downloads/total/2024-01-01:2025-10-01/", pkg)
) |>
  req_user_agent("your-app/1.0") |>
  req_perform() |>
  resp_body_json()

resp

What to use for accurate organizational metrics

If your team installs via Posit Package Manager (PPM), use PPM’s usage/metrics APIs or exported logs. That’s the only way to measure internal downloads reliably (across dev laptops, servers, and CI) for your org.

Ask your admin to enable metrics and provide API access. Then query by repo, package, version, and time window.

Complementary signals beyond downloads

  • Reverse dependencies on CRAN/Bioconductor
  • GitHub stars, forks, issues, discussions
  • Citations (if you have a paper or JOSS entry)
  • Container pulls (if you publish images)
  • Internal telemetry (opt‑in), surveys, and support tickets

These provide richer pictures of adoption and impact.

Recommendations

  • Use cranlogs for public, comparable proxy trends—with clear caveats.
  • For real adoption in your org, use your repository manager’s metrics.
  • Combine multiple signals (deps, GitHub, citations) for a fuller view.
  • Avoid treating any single metric as “true downloads.”