Principal Component Analysis applied to 64 formerly colonized countries to identify the structural dimensions of comparative development, interpreted through Acemoglu, Johnson & Robinson's (2001) institutional framework.
Correlation circle — PC1 & PC2 |
Individuals by continent — PC1 & PC2 |
Biplot — variables & individuals |
Correlation matrix |
What are the main dimensions that explain differences in economic development across formerly colonized countries, and to what extent do institutional, geographic, and historical factors structure these differences?
Source: Cross-sectional dataset assembled from Acemoglu, Johnson & Robinson (2001) and related sources (df_DM_2025_ADD.csv).
Coverage: 64 countries, one observation per country (circa 1995–2016).
Active variables (10):
| Variable | Description |
|---|---|
risk |
Expropriation risk index (0–10, higher = better institutions) |
edes1975 |
Share of population with European descent in 1975 (%) |
malfal94 |
Malaria prevalence in 1994 (%) |
mort0 |
Estimated settler mortality (per 10,000) |
gdp |
GDP per capita at PPP, 1995 (USD) |
population |
Population size, 2024 |
land |
Territory area (km²) |
density |
Population density (inhabitants/km²) |
latitude |
Absolute mean latitude of country |
average_temperature |
Mean temperature 1991–2020 (°C) |
Illustrative variables (4): continent, hemisphere (qualitative); slave, neoeuro (quantitative). These are projected onto PCA axes without influencing their construction.
Transformations: All active variables standardized (zero mean, unit variance) prior to PCA, given the heterogeneous units of measurement.
Method: Principal Component Analysis (PCA) via FactoMineR::PCA(), with standardization (scale.unit = TRUE). All countries receive equal weight, consistent with the objective of a balanced comparative analysis rather than a size-weighted study.
Number of components retained: 4 axes, selected on three converging criteria:
- Kaiser criterion (eigenvalue > 1): 4 axes
- Scree plot: elbow clearly visible after axis 4
- Cumulative inertia: 79.68% explained by the first four axes
Qualitative illustrative variables (continent, hemisphere) were introduced via the quali.sup argument; quantitative illustratives (slave, neoeuro) via quanti.sup.
Four principal components account for 79.68% of total variance:
| Component | Eigenvalue | % Inertia | Cumulative % |
|---|---|---|---|
| PC1 | 4.284 | 42.84 | 42.84 |
| PC2 | 1.519 | 15.19 | 58.03 |
| PC3 | 1.157 | 11.57 | 69.60 |
| PC4 | 1.008 | 10.08 | 79.68 |
PC1 — Economic and institutional development (42.84%).
Dominated by GDP (gdp, 16.6%), European settler share (edes1975, 16.0%), temperature (average_temperature, 15.6%), expropriation risk (risk, 13.3%), and latitude (12.8%). This axis opposes settler colonies with strong institutions (Canada, USA, Australia, New Zealand — high GDP, low malaria, temperate climate) to extractive colonies with fragile institutions (Mali, Niger, Burkina Faso — low GDP, high malaria, tropical climate). The result corroborates Acemoglu et al. (2001): institutional quality is the primary predictor of comparative development.
PC2 — Demographic density (15.19%).
Driven almost entirely by density (49.7%). Opposes small, densely populated territories (Singapore, Hong Kong) to large, sparsely populated ones (Canada, Australia). This dimension is structurally independent of PC1: high development can coexist with either high or low density.
PC3 — Absolute population size (11.57%).
Dominated by population (63.6%). Distinguishes demographically large countries (India, Indonesia, Nigeria, Pakistan) from smaller ones, independently of density — revealing governance challenges specific to demographic scale.
PC4 — Colonial sanitary legacy (10.08%).
Driven by settler mortality mort0 (53.7%) and malaria prevalence malfal94 (15.1%). Countries with historically high settler mortality (tropical West and Central Africa, tropical Latin America) established extractive institutions; those with low settler mortality (Canada, USA, Australia) established inclusive ones. This dimension shows that colonial-era health conditions continue to structure development more than 150 years later.
Illustrative variables: The neoeuro variable projects firmly into the high-development quadrant of PC1, confirming the "Neo-Europe" model. Continent and hemisphere overlay onto PC1 as expected: African countries cluster in the low-development pole, North American and Oceanian countries in the high-development pole.
Representation quality: 33 countries (51.6%) are well-represented on the PC1–PC2 plane (cos² > 0.5); 17 (26.6%) are poorly represented (cos² < 0.2), notably India, Guatemala, Ecuador, and Indonesia — countries whose development trajectories are atypical and require axes 3–4 for fuller interpretation.
R version: 4.3 or later Required packages:
install.packages(c("corrplot", "dplyr", "FactoMineR", "factoextra",
"ggplot2", "here", "readr"))Run order:
- Open an R session with the working directory set to the project root (or use RStudio with the project file).
- Source
dm.R. The script will createresults/and populate it with all tables.
source("dm.R")The written report (rapport/main.typ) is compiled with Typst. With Typst installed:
cd rapport && typst compile main.typ- PCA identifies correlational structure, not causal mechanisms. The institutional interpretation draws on external theory (Acemoglu et al., 2001) and cannot be inferred from the PCA alone.
- The 64-country sample covers formerly colonized societies; results are not generalizable to non-colonial development trajectories.
- Data represent a single cross-section and do not capture convergence dynamics or structural breaks.
- Several potentially relevant variables are absent: human capital, trade openness, natural resource endowments, and contemporary political stability.
- 26.6% of countries are poorly represented on the main factorial plane, indicating that four dimensions do not fully capture development heterogeneity.
- The analysis reduces each country's history to its colonial experience, ignoring pre-colonial structures.
.
├── README.md # This file
├── .gitignore
├── dm.R # Full PCA analysis script
├── df_DM_2025_ADD.csv # Dataset (64 countries × 17 variables)
├── results/ # Generated by dm.R — eigenvalues, coordinates,
│ # contributions and cos² per axis (CSV, gitignored)
└── rapport/
├── main.typ # Report source (Typst)
├── refs.bib # Bibliography
├── rapport.pdf # Compiled report
├── figures/ # 16 PNG figures — generated by dm.R
└── template/
├── lib.typ # Typst template library ('Ilm)
└── typst.toml # Typst package configuration
Analyse de données — Antoine C. · Noah D.-G. · Jules D. · Hans P. | Grade: 19/20 | S5, Université Catholique de Lille