ggplot2 Grammar-of-Graphics Visualization

DSCI 324 | Data Visualization | Python (plotnine)

A comprehensive implementation of the full ggplot2 grammar-of-graphics framework in Python using plotnine, producing 20+ charts across 12 core concepts. Each concept is demonstrated with a dedicated chart and side-by-side wrong/correct or before/after comparisons where applicable.

---

Project Overview

Attribute	Detail
Language	Python 3
Core Library	plotnine (R ggplot2 equivalent in Python)
Datasets	Gapminder (142 countries, 1952–2007) · GSS Survey (2,867 respondents, synthetic)
Charts Produced	20+ across 12 core ggplot2 concepts
Notebook	ggplot2_All12_Concepts.ipynb

---

Datasets

Gapminder — gapminder (PyPI package)

• 142 countries across 12 time points (1952–2007, every 5 years)
• Variables: country, continent, year, lifeExp, gdpPercap, pop
• Derived column: log_gdp = log(gdpPercap)

GSS Survey — Synthetic GSS-style data (N = 2,867)

• Variables: age, childs, sex, race, religion, bigregion, degree
• Distributions calibrated to match General Social Survey population proportions
• Seed: np.random.seed(42) for reproducibility

---

Concepts Covered

#	Concept	Chart Type	Key Demonstration
1	aes() mapping	Scatter	4 simultaneous aesthetics: x, y, color, size
2	geom_line()	Line chart	Wrong (no group=) vs correct (group=country)
3	geom_point()	Scatter	alpha, size, overplotting control
4	geom_bar()	Bar chart	Auto-count stat; plain vs fill-encoded
5	geom_smooth()	Scatter + trend	se=True (ribbon) vs se=False (clean line)
6	facet_wrap()	Small-multiples	Default layout vs nrow=1; one-variable faceting
7	facet_grid()	2D panel grid	Two-variable row × col faceting
8	scale_y_log10()	Log-scaled axis	Before/after: skewed GDP normalized with dollar labels
9	scale_x_continuous()	Custom axis	Dollar-formatted tick labels ($30k, $60k)
10	labs()	Any	Bare plot vs full title, subtitle, caption, axis, legend
11	guides()	Any	Redundant legend suppression; when to keep vs remove
12	position=	Bar chart	stack → dodge → fill progression

---

Setup

pip install plotnine gapminder

import pandas as pd
import numpy as np
from plotnine import *
from gapminder import gapminder

No external data files required — both datasets are generated in the setup cell. Run all cells top to bottom.

---

Chart Highlights

Chart 2 — geom_line() Wrong vs Correct The most common ggplot2 mistake: omitting group=country collapses all 142 countries into a single jagged line. Chart 2a shows the broken output; Chart 2b shows the fix with group=country.

Chart 6 — facet_wrap() Three Variants Three layouts demonstrated: default 2-row wrap by continent, nrow=1 single-row (matching the lecture 04 style), and a GSS age × children scatter faceted by sex.

Chart 8 — scale_y_log10() Before/After Kuwait's GDP spike dominates the raw y-axis and compresses all other countries. Log transformation makes all 142 country trends readable simultaneously — the canonical use case for log-scale in data analytics.

Chart 11 — guides() Three-Panel Sequence Shows exactly when to suppress vs keep a legend: redundant (suppress), redundant with color (suppress), two different variables mapped to x and fill (keep).

Chart 12 — Position Variants Full stack → dodge → fill progression on religion × census region data — demonstrating when each encoding is appropriate for categorical composition analysis.

---

File Structure

├── ggplot2_All12_Concepts.ipynb    # Main notebook — all 12 concepts, 20+ charts
└── README.md

---

Key Takeaways

• aes() controls all data-driven visual mappings; fixed properties (e.g. color='gray') go outside aes()
• group= is mandatory for geom_line() on multi-series data — the single most common ggplot2 error
• scale_y_log10() is the correct tool for right-skewed distributions (GDP, costs, population)
• facet_wrap() splits by one variable; facet_grid() splits by two in a true 2D layout
• guides(fill=False) removes a legend when x= and fill= encode the same variable
• position='fill' normalizes to 100% for rate comparison; position='dodge' enables direct value comparison

---

References

• plotnine documentation
• ggplot2 documentation
• Gapminder dataset: Hans Rosling / Gapminder Foundation
• GSS: NORC at the University of Chicago (synthetic approximation)