Viz

Data Visualization

Create publication-ready figures using ggplot2 in R. Validate data and gather all visual requirements before generating the first plot.

When to Use

• Creating any chart, figure, or plot
• Refining an existing visualization (spacing, labels, colors, layout)
• User says "plot", "graph", "visualize", "figure", "chart"

When NOT to Use

• Quick exploratory plot() or hist() for debugging (just do it)
• User explicitly asks for base R graphics

Core Rule

Ask about ALL visual elements upfront before generating the first plot. Do not produce a draft and iterate one tweak at a time.

Workflow

Step 1: Validate Data

Before any plotting code, confirm:

1. Read the data and print column names, types, and 3-5 sample rows
2. Confirm column mappings with the user: which columns map to x, y, color, facet, etc.
3. Check for issues: NAs in plot variables, unexpected factor levels, wrong types

Do NOT guess column mappings silently. If ambiguous, ask.

Step 2: Gather Visual Requirements

Ask about ALL of these before the first plot:

| Element | Ask about | Default if not specified | |---------|-----------|------------------------| | Chart type | bar, coef/forest, line, network, other | Infer from data structure | | Layout | single panel, faceted, multi-panel | Single panel | | Axes | labels, limits, breaks, log scale | Auto with clear labels | | Colors | palette, specific mappings | Colorblind-safe, clean | | Error bars | SE, 95% CI, none | 95% CI when applicable | | Legend | position, title, label text | Right side, auto title | | Separators | lines between groups | None unless grouped | | Text | title, subtitle, caption, annotations | Minimal | | Dimensions | width x height in inches | 7 x 5 |

Step 3: Build the Plot

Language: R with ggplot2. Always.

Theme baseline:

theme_minimal(base_size = 12) +
  theme(
    panel.grid.minor = element_blank(),
    strip.text = element_text(face = "bold"),
    legend.position = "bottom"
  )

Color palette: Colorblind-safe. No strong default - pick appropriate per chart:

• Categorical (2-8 groups): scale_color_brewer(palette = "Set2") or similar
• Sequential: scale_fill_viridis_c()
• Diverging: scale_fill_distiller(palette = "RdBu")

Multi-panel approach:

• Same chart type across panels: facet_wrap() / facet_grid()
• Different chart types combined: patchwork package

Step 4: Chart-Specific Patterns

Bar Charts

ggplot(df, aes(x = group, y = estimate, fill = condition)) +
  geom_col(position = position_dodge(width = 0.8), width = 0.7) +
  geom_errorbar(aes(ymin = ci_low, ymax = ci_high),
                position = position_dodge(width = 0.8), width = 0.2) +
  # 95% CI error bars by default

Coefficient / Forest Plots

ggplot(df, aes(x = estimate, y = reorder(term, estimate))) +
  geom_vline(xintercept = 0, linetype = "dashed", color = "gray50") +
  geom_point(size = 2) +
  geom_errorbarh(aes(xmin = ci_low, xmax = ci_high), height = 0.2)

Input sources (handle all three):

• R model objects: Use broom::tidy(conf.int = TRUE) or marginaleffects::avg_slopes()
• Stata margins output: Parse CSV/text from margins or esttab export
• Pre-computed CSV: Expect columns: term, estimate, ci_low, ci_high

Line / Time Series

ggplot(df, aes(x = time, y = value, color = group)) +
  geom_line(linewidth = 0.8) +
  geom_ribbon(aes(ymin = ci_low, ymax = ci_high, fill = group), alpha = 0.15)

Network Graphs

Use igraph base plotting (not ggraph):

library(igraph)
plot(g,
     vertex.size = degree(g) * 2,
     vertex.label.cex = 0.7,
     vertex.color = V(g)$color,
     edge.arrow.size = 0.3,
     layout = layout_with_fr(g))

Step 5: Stata-to-R Pipeline

When combining Stata models with R visualization:

1. In Stata: Export estimates with esttab using "estimates.csv", csv ci

or: margins, post then matrix list e(b), matrix list r(table)

1. In R: Parse the CSV, clean column names, build ggplot

# Parse Stata esttab CSV output
est <- read.csv("estimates.csv", skip = 1)  # skip header row
# Clean: remove significance stars, convert to numeric
est$estimate <- as.numeric(gsub("[*]", "", est$estimate))

Step 6: Save Output

Always save both formats:

ggsave("figure.png", width = 7, height = 5, dpi = 300)
ggsave("figure.pdf", width = 7, height = 5)

Adjust dimensions based on content:

• Single panel: 7 x 5
• Two panels side by side: 10 x 5
• Tall coefficient plot (many terms): 7 x 8
• Network graph: 7 x 7

Quick Reference

| Chart type | Key geom | Error bars | Default | |------------|----------|------------|---------| | Bar | geom_col + position_dodge | geom_errorbar (95% CI) | Dodged, 0.7 width | | Coefficient | geom_point + geom_errorbarh | Built-in (CI) | Horizontal, ref line at 0 | | Line | geom_line + geom_ribbon | geom_ribbon (CI band) | 0.15 alpha ribbon | | Network | igraph::plot() | N/A | Fruchterman-Reingold layout |

Common Mistakes

| Mistake | Fix | |---------|-----| | Guessing column names | Always read data first, confirm with user | | Iterating one tweak at a time | Ask about ALL visual elements before first plot | | Using theme_gray() default | Always start with theme_minimal(base_size = 12) | | Forgetting position_dodge on error bars | Error bars must match bar dodge width exactly | | Network plots with ggraph when user expects igraph | Default to igraph base plot() | | Not saving both PNG and PDF | Always ggsave() both formats | | Wrong dimensions for multi-panel | Scale width with number of panels | | Parsing Stata output without cleaning stars | Strip *, **, *** before as.numeric() |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: letitbk
• Source: letitbk/claude-academic-setup
• License: MIT

Install and usage instructions live in the source repository linked above.