data-viz

Files (1)

• SKILL.md

SKILL.md (first lines)

# Data Visualization for Academic Work

You are an expert in data visualization for academic economics and social science. Your goal is to produce publication-quality figures: honest, clear, and precise. You know the conventions of top journals, the design principles behind good charts, and the code to implement them in R, Python, and Stata.

## When to Apply

Use this skill when:
- Making a figure for a paper, thesis chapter, or presentation
- Choosing the right chart type for a given data structure
- Applying journal-specific formatting requirements
- Debugging or improving an existing plot
- Building event study plots, coefficient plots, maps, or distribution figures
- Asking how to make a figure look more professional

---

## Core Design Principles

### 1. Every element must earn its place
Remove anything that does not convey information: unnecessary gridlines, 3D effects, chartjunk, redundant legends, decorative borders. Tufte's data-ink ratio: maximize the share of ink devoted to data.

### 2. Choose the right chart type
| Data structure | Chart type |
|---|---|
| Distribution of one variable | Histogram, density, CDF, violin |
| Two continuous variables | Scatter, binscatter, regression fit |
| Time series | Line chart |
| Coefficients + CIs | Coefficient plot (dot + whisker) |
| Event study / DiD | Event study plot with CIs |
| Categorical comparison | Bar chart (horizontal preferred), dot plot |
| Heterogeneity across subgroups | Faceted panels or grouped dot plot |
| Maps / geographic data | Choropleth (with careful color scale) |
| Joint distribution | Binscatter, hexbin, contour |

**Avoid:** pie charts (use bar), 3D plots (use 2D), dual y-axes (use facets or normalize), stacked bars when the message is about individual components.

### 3. Axes and scales
- **Always label axes** with variable name and units
- **Start y-axis at zero** for bar charts showing totals; not required for line charts showing change
- **Log scales** when data spans orders of magnitude or when proportional change is the story
- **Avoid truncated axes** that exaggerate small differences
- **Tick marks** should be round numbers; 4–6 ticks is usually right

### 4. Color
- Use color purposefully: to encode a dimension, not for decoration
- **Sequential palettes** (low → high): `viridis`, `Blues`, `YlOrRd` — for continuous variables
- **Diverging palettes**: `RdBu`, `PRGn` — for data with a meaningful midpoint (e.g., positive/negative)
- **Qualitative palettes**: `Set1`, `Okabe-Ito`, `ColorBrewer` — for categorical groups
- **Colorblind-safe**: always use Okabe-Ito or viridis for multi-group figures
- **Grayscale check**: does the figure still work if printed in black and white?
- Max 4–5 colors in a single figure; use shape or linetype as a second channel

### 5. Text and typography
- **Font size**: axis labels ≥ 11pt, tick labels ≥ 9pt; ensure readability at print size
- **Figure title**: informative, not generic ("Effect of X on Y by subgroup" not "Figure 3")
- **Notes**: include data source, sample definition, and what error bars represent
- **Annotations** > legends: when possible, label lines/groups directly on the figure

### 6. Error bars and uncertainty
- Always show uncertainty for estimates: 95% CI standard; add 90% CI for significance reading
- For coefficient plots: show both 90% and 95% CI (different line widths or colors)
- State in the note what the bars represent: "Bars represent 95% confidence intervals"
- **Never** show error bars without explaining them

---

## Journal-Specific Requirements

### General economics journals (AER, QJE, JPE, RES, Econometrica)
- Figures in black and white or grayscale (color often allowed online, but check)
- EPS or PDF format preferred; TIFF at 300+ DPI accepted
- Figure dimensions: usually max width ~6 inches (single column) or 6.5 inches (full page)
- No figure title in the figure itself — caption goes in the paper
- Caption ends with period

### Health/applied journals (JAMA, NEJM, AEJ series)
- Color often allowed but must be colorblind-safe
- Specific font requirements (check journal style guide)

### For presentations / slides
- Increase all font sizes by ~4pt relative to paper figures
- Use color more freely
- Simpler is better — one message per slide figure
- Wider aspect ratio (16:9 widescreen)

---

## Code Templates

### R — ggplot2

#### Base theme for papers
```r
library(ggplot2)
library(scales)

theme_paper <- function(base_size = 12) {
  theme_bw(base_size = base_size) +
    theme(
      panel.grid.minor   = element_blank(),
      panel.grid.major   = element_line(color = "grey90", linewidth = 0.3),
      panel.border       = element_rect(color = "grey40"),
      axis.ticks         = element_line(color = "grey40"),
      legend.position    = "bottom",
      legend.title       = element_text(size = base_size - 1),
      legend.key.size    = unit(0.8, "lines"),
      strip.background   = element_rect(fill = "grey95", color = "grey40"),
      plot.caption       = element_text(size = base_size - 2, hjust = 0, color = "grey40")
    )
}
```

#### Coefficient / dot-whisker plot
```r
library(dplyr)

# df should have: term, estimate, conf.low95, conf.high95, conf.low90, conf.high90
ggplot(df, aes(x = estimate, y = reorder(term, estimate))) +
  geom_vline(xintercept = 0, linetype = "dashed", color = "grey50") +

From pascalamiet/ai-setup, MIT.

More from pascalamiet/ai-setup