---

name: r-data-science description: "R programming for data analysis, visualization, and statistical workflows. Use when working with R scripts (.R), Quarto documents (.qmd), RMarkdown (.Rmd), or R projects. Covers tidyverse workflows, ggplot2 visualizations, statistical analysis, epidemiological methods, and reproducible research practices."

R Data Science

Overview

Generate high-quality R code following tidyverse conventions and modern best practices. This skill covers data manipulation, visualization, statistical analysis, and reproducible research workflows commonly used in public health, epidemiology, and data science.

Core Principles

1. Tidyverse-first: Use tidyverse packages (dplyr, tidyr, ggplot2, purrr, readr) as the default approach
2. Pipe-forward: Use the native pipe |> for chains (R 4.1+); fall back to %>% for older versions
3. Reproducibility: Structure all work for reproducibility using Quarto, renv, and clear documentation
4. Defensive coding: Validate inputs, handle missing data explicitly, and fail informatively

Quick Reference: Common Patterns

Data Import

library(tidyverse)

# CSV (most common)
df <- read_csv("data/raw/dataset.csv")

# Excel
df <- readxl::read_excel("data/raw/dataset.xlsx", sheet = "Sheet1")

# Clean column names immediately
df <- df |> janitor::clean_names()

Data Wrangling Pipeline

analysis_data <- raw_data |>
  # Clean and filter
  filter(!is.na(key_variable)) |>

  # Transform variables
  mutate(
    date = as.Date(date_string, format = "%Y-%m-%d"),
    age_group = cut(age, breaks = c(0, 18, 45, 65, Inf),
                    labels = c("0-17", "18-44", "45-64", "65+"))
  ) |>

  # Summarize
  group_by(region, age_group) |>
  summarize(
    n = n(),
    mean_value = mean(outcome, na.rm = TRUE),
    .groups = "drop"
  )

Basic ggplot2 Visualization

ggplot(df, aes(x = date, y = count, color = category)) +
  geom_line(linewidth = 1) +
  scale_color_brewer(palette = "Set2") +
  labs(
    title = "Trend Over Time",
    subtitle = "By category",
    x = "Date",
    y = "Count",
    color = "Category",
    caption = "Source: Dataset Name"
  ) +
  theme_minimal(base_size = 12) +
  theme(
    legend.position = "bottom",
    plot.title = element_text(face = "bold")
  )

Tidyverse Style Guide Essentials

Naming Conventions

• snake_case for objects and functions: case_counts, calculate_rate()
• Verbs for functions: filter_outliers(), compute_summary()
• Nouns for data: patient_data, surveillance_df
• Avoid: dots in names (reserved for S3), single letters except in lambdas

Code Formatting

• Indentation: 2 spaces (never tabs)
• Line length: 80 characters maximum
• Operators: Spaces around <-, =, +, |>, but not :, ::, $
• Commas: Space after, never before
• Pipes: New line after each |>

# Good
result <- data |>
  filter(year >= 2020) |>
  group_by(county) |>
  summarize(total = sum(cases))

# Bad
result<-data|>filter(year>=2020)|>group_by(county)|>summarize(total=sum(cases))

Assignment

• Use <- for assignment, never = or ->
• Use = only for function arguments

Comments

# Load and clean surveillance data ------------------------------------------

# Calculate age-adjusted rates
# Using direct standardization method per CDC guidelines
adjusted_rate <- calculate_adjusted_rate(df, standard_pop)

Package Ecosystem

Core Tidyverse (Always Load)

library(tidyverse)  # Loads: ggplot2, dplyr, tidyr, readr, purrr, tibble, stringr, forcats

Data Import/Export

Task	Package	Key Functions
CSV/TSV	readr	read_csv(), write_csv()
Excel	readxl, writexl	read_excel(), write_xlsx()
SAS/SPSS/Stata	haven	read_sas(), read_spss(), read_stata()
JSON	jsonlite	read_json(), fromJSON()
Databases	DBI, dbplyr	dbConnect(), tbl()

Data Manipulation

Task	Package	Key Functions
Column cleaning	janitor	clean_names(), tabyl()
Date handling	lubridate	ymd(), mdy(), floor_date()
String operations	stringr	str_detect(), str_extract()
Missing data	naniar	vis_miss(), replace_with_na()

Visualization

Task	Package	Key Functions
Core plotting	ggplot2	ggplot(), geom_*()
Extensions	ggrepel, patchwork	geom_text_repel(), + operator
Interactive	plotly	ggplotly()
Tables	gt, kableExtra	gt(), kable()

Statistical Analysis

Task	Package	Key Functions
Model summaries	broom	tidy(), glance(), augment()
Regression	stats, lme4	lm(), glm(), lmer()
Survival	survival	Surv(), survfit(), coxph()
Survey data	survey	svydesign(), svymean()

Epidemiology & Public Health

Task	Package	Key Functions
Epi calculations	epiR	epi.2by2(), epi.conf()
Outbreak tools	incidence2, epicontacts	incidence(), make_epicontacts()
Disease mapping	SpatialEpi	expected(), EBlocal()
Surveillance	surveillance	sts(), farrington()
Rate calculations	epitools	riskratio(), oddsratio(), ageadjust.direct()

Reproducibility Standards

Project Structure

project/
├── project.Rproj
├── renv.lock
├── CLAUDE.md              # Claude Code configuration
├── README.md
├── data/
│   ├── raw/               # Never modify
│   └── processed/         # Analysis-ready
├── R/                     # Custom functions
├── scripts/               # Pipeline scripts
├── analysis/              # Quarto documents
└── output/
    ├── figures/
    └── tables/

Quarto Document Header

---
title: "Analysis Title"
author: "Your Name"
date: today
format:
  html:
    toc: true
    code-fold: true
    embed-resources: true
execute:
  warning: false
  message: false
---

Package Management with renv

# Initialize (once per project)
renv::init()

# Snapshot dependencies after installing packages
renv::snapshot()

# Restore environment (for collaborators)
renv::restore()

Workflow Documentation

Always include at the top of scripts:

# ============================================================================
# Title: Analysis of [Subject]
# Author: [Name]
# Date: [Date]
# Purpose: [One-sentence description]
# Input: data/processed/clean_data.csv
# Output: output/figures/trend_plot.png
# ============================================================================

Common Analysis Patterns

Descriptive Statistics Table

df |>
  group_by(category) |>
  summarize(
    n = n(),
    mean = mean(value, na.rm = TRUE),
    sd = sd(value, na.rm = TRUE),
    median = median(value, na.rm = TRUE),
    q25 = quantile(value, 0.25, na.rm = TRUE),
    q75 = quantile(value, 0.75, na.rm = TRUE)
  ) |>
  gt::gt() |>
  gt::fmt_number(columns = where(is.numeric), decimals = 2)

Regression with Tidy Output

model <- glm(outcome ~ exposure + age + sex, data = df, family = binomial)

# Tidy coefficients
tidy_results <- broom::tidy(model, conf.int = TRUE, exponentiate = TRUE) |>
  select(term, estimate, conf.low, conf.high, p.value)

# Model diagnostics
glance_results <- broom::glance(model)

Epi Curve (Epidemic Curve)

library(incidence2)

# Create incidence object
inc <- incidence(
  df,
  date_index = "onset_date",
  interval = "week",
  groups = "outcome_category"
)

# Plot
plot(inc) +
  labs(
    title = "Epidemic Curve",
    x = "Week of Onset",
    y = "Number of Cases"
  ) +
  theme_minimal()

Rate Calculation

# Age-adjusted rates using direct standardization
library(epitools)

# Stratum-specific counts and populations
result <- ageadjust.direct(
  count = df$cases,
  pop = df$population,
  stdpop = standard_population$pop  # e.g., US 2000 standard
)

Error Handling

Defensive Data Checks

# Validate data before analysis
stopifnot(
  "Data frame is empty" = nrow(df) > 0,
  "Missing required columns" = all(c("id", "date", "value") %in% names(df)),
  "Duplicate IDs found" = !any(duplicated(df$id))
)

# Informative warnings for data quality issues
if (sum(is.na(df$key_var)) > 0) {
  warning(sprintf("%d missing values in key_var (%.1f%%)",
                  sum(is.na(df$key_var)),
                  100 * mean(is.na(df$key_var))))
}

Safe File Operations

# Check file exists before reading
if (!file.exists(filepath)) {
  stop(sprintf("File not found: %s", filepath))
}

# Create directories if needed
dir.create("output/figures", recursive = TRUE, showWarnings = FALSE)

Performance Tips

For Large Datasets

# Use data.table for >1M rows
library(data.table)
dt <- fread("large_file.csv")

# Or use arrow for very large/parquet files
library(arrow)
df <- read_parquet("data.parquet")

# Lazy evaluation with duckdb
library(duckdb)
con <- dbConnect(duckdb())
df_lazy <- tbl(con, "data.csv")

Vectorization Over Loops

# Good: vectorized
df$rate <- df$cases / df$population * 100000

# Avoid: row-by-row loop
for (i in 1:nrow(df)) {
  df$rate[i] <- df$cases[i] / df$population[i] * 100000
}

Additional Resources

For detailed patterns, consult:

• Tidyverse Style Guide: https://style.tidyverse.org/
• R for Data Science (2e): https://r4ds.hadley.nz/
• The Epidemiologist R Handbook: https://epirhandbook.com/
• Quarto Documentation: https://quarto.org/

Version History

• v1.0.0 (2025-12-04): Initial release for PubHealthAI community