Key Takeaways: Statistical and Scientific Visualization

  1. Publication figures have specific requirements. Journals specify figure widths (single-column ~3.5 in / 89 mm, double-column ~7 in / 183 mm), minimum font sizes (7-9 pt), font families (Arial/Helvetica), and file formats (PDF/EPS preferred, TIFF at 300+ DPI). Read the guidelines for your target journal before designing the figure.

  2. Font embedding is mandatory. Set mpl.rcParams["pdf.fonttype"] = 42 and mpl.rcParams["ps.fonttype"] = 42 to force Type 42 TrueType output. Journals reject figures with Type 3 fonts because many PDF processors cannot handle them.

  3. Reusable style modules save time. Create a Python module (nature_style.py, plos_style.py) that applies rcParams and provides convenience functions for sized figures and panel labels. Import at the top of every figure script. Ensures consistency and reduces per-figure effort.

  4. Panel labels use axes coordinates. Add labels with ax.text(-0.15, 1.05, "a", transform=ax.transAxes, fontsize=12, fontweight="bold"). The transform=ax.transAxes means coordinates are fractions of the axes. Negative x and y > 1 position text outside the top-left of the axes.

  5. Error bars and confidence bands are essential. Use ax.errorbar for discrete estimates and ax.fill_between for continuous bands. Always disclose what the error represents (SEM, SD, 95% CI, bootstrap interval) in the caption.

  6. Significance brackets via statannotations. The statannotations library automates pairwise significance annotations on seaborn categorical plots. Configure the test, significance format, and position; the library runs the test and draws the bracket.

  7. Colorblind and grayscale safety. Use the Wong palette (or similar) for categorical colors and verify grayscale readability. Redundant encoding (color + line style + marker) is the surest way to ensure accessibility.

  8. Forest plots, volcano plots, Manhattan plots, QQ plots. Each is a specialized format for a specific statistical context (meta-analysis, high-throughput screening, GWAS, distributional diagnostics). Know the conventions for any format your field uses regularly.

  9. Effect sizes beat p-value thresholds. Modern guidance emphasizes effect sizes with confidence intervals over p-value stars. Tools like DABEST make estimation plots accessible. The format you choose shapes the reasoning the reader applies, so choose deliberately.

  10. Reproducibility is part of figure quality. Publish the data and code alongside the paper. Show full distributions, not just summaries. Report sample sizes. Use version control. A figure is not just the image; it is also the data and code behind it.

Chapter 27 is applied rather than conceptually novel. The discipline it teaches — checking every figure against external requirements — serves you whenever you produce a chart for an audience with standards, whether a journal, a client, a brand guide, or a style committee. Chapter 28 closes Part VI with big data visualization strategies.