- complexity-analysis.R aggregates the block level data and generates several visualizations used in the analysis. The script uses aggregation_func.R to facilitate aggregation from the the block level to higher geographic scales.
- complexity-analysis.R will automatically download several GB of block level data and add it to folders within the top level of the repo. The database is available at the website millionneighborhoods.africa with a data dictionary available here. The script also uses the following files in this repo: data /buildings_lusaka.geojson, data/streets_lusaka.geojson. Data sources and methods for generating the block level population and informality data are available in the mansueto-institute/kblock repo.
- dhs-analysis.R downloads DHS data via an API connection, spatial joins DHS administrative regions to the block level database, aggregates blocks statistics to the DHS regions, and then performs a statistical analysis that runs correlations, PCA, and regressions on the relationships between block statistics and social development indicators corresponding to human well-being and household characteristics.
- dhs-analysis.R must be run after the step in the complexity-analysis.R that downloads the block level database. When running on your own computer please be sure to register and add your own DHS login credentials to the script (this requires undergoing a dataset access approval process). The pre-processing steps that involve API calls to the DHS website and spatial joins may take an hour of processing time. This script also uses data downloaded from data/un-habitat. To account for street densities this script uses streets_dhs_regions.csv.
- graph-viz.R visualizes block complexity in the format of a network graph. This script uses the graph_funcs.R to generate a complexity graph. These functions were specifically designed for the R programming language to replicate the basic functionality of the kblock repo, which performs the actual method used in the scientific article and associated datasets.
- Below are functions applied in the graph-viz.R script which demonstrate how to compute block complexity graphs and visualize parcel level diagrams. Again these methods were designed for heuristic purposes and not used in production at scale.
# Function to generate graph using building parcels (graph_polys) and on-network streets that connect interior buildings to exterior roads (graph_exclude) graph_lines <- generate_graph(graph_polys = building_parcels_polygons, graph_exclude = street_linestrings) # Function to convert graph to polygon representation graph_polys <- convert_to_poly(lines = graph_lines) # Function to compute block complexity at the parcel level such that each row to the output represents each subsequent interior layer of buildings. graph_duals <- extract_duals(graph_sf = graph_polys) - The script uses data/layer_lusaka.geojson, which contains data for a community area in Lusaka, Zambia.
- To ground-truth block complexity statistics use sdi-analysis.R which compares block geometries to SDI settlements. Note, the data requires special access permissions. Contact authors for more information.
- To validate the completeness of OSM street data use street-analysis.R which utilizes three data files: streets_metrics.parquet, streets_summary.csv, and streets_validation.csv.
- To compare block complexity performance to alternative predictors of wealth, see these statistical comparisons with Blumenstock et al. (2022) and Yeh et al. (2020)
graph LR
G[DHS API] --> C
A[africa_data.parquet] --> B[Step 1: complexity-analysis.R]
M[africa_geodata.parquet] --> B
M[africa_geodata.parquet] --> C
Q[Step 0: kblock.git] --> M
E[aggregation_func.R] --> B
A --> C[Step 2: dhs-analysis.R]
B --> Z[Summary stats and<br>visualizations]
C --> X[Statistical analysis and<br>visualizations]
Q[Step 0: kblock.git] --> A
F[graph_funcs.R] --> D[Step 3: graph-viz.R]
J[layer_lusaka.geojson] --> D
K[buildings_lusaka.geojson] --> B
L[streets_lusaka.geojson] --> B
D --> Y[Complexity graph<br>visualization]
style A fill:#eba8d3
style G fill:#eba8d3
style J fill:#eba8d3
style K fill:#eba8d3
style L fill:#eba8d3
style Z fill: #f7f5eb
style X fill: #f7f5eb
style Y fill: #f7f5eb
style F fill: #ADD8E6
style E fill: #ADD8E6
style Q fill: #b1d8b7
- R version 4.1.2 (2021-11-01)
- Platform: aarch64-apple-darwin20 (64-bit)
- Running under: macOS 13.4.1
| Package | Version | Package | Version | Package | Version | ||
|---|---|---|---|---|---|---|---|
| tidyverse | 1.3.1 | broom | 1.0.3 | readxl | 1.3.1 | ||
| sf | 1.0-8 | betareg | 3.1-4 | osmdata | 0.1.9 | ||
| units | 0.8-0 | ggpmisc | 0.5.2 | Hmisc | 4.8-0 | ||
| viridis | 0.6.2 | kableExtra | 1.3.4 | scatterpie | 0.1.7 | ||
| patchwork | 1.1.1 | xtable | 1.8-4 | tidymodels | 0.2.0 | ||
| scales | 1.2.1 | ggplot2 | 3.4.1 | ggrepel | 0.9.3 | ||
| rdhs | 0.7.5 | tidygeocoder | 1.0.5 | readr | 2.1.1 | ||
| rmapshaper | 0.4.6 | writexl | 1.4.0 | DescTools | 0.99.48 | ||
| arrow | 7.0.0 | ggsn | 0.5.0 | ggpubr | 0.6.0 | ||
| sfarrow | 0.4.1 | geoarrow | 0.0.0.9000 |
Nicholas Marchio, data scientist at the Mansueto Institute. For any technical inquiries please feel free to create a Git issue and tag nmarchio.
For related technical work see the following repos:
- kblock: Python codebase for generating the underlying block complexity and population data available at millionneighborhoods.africa.
- geopull: Python package for extracting OSM data and generating street block delineations.
- cloudtile: CLI for converting (Geo)Parquet files to PMTiles on AWS.
- prclz: Python codebase from previous iteration of block complexity research (see kblock for more recent version).
- Google drive with restricted data for full reproducibility for analysis.
- UChicago Box with restricted data for full reproducibility for analysis.
- Google drive with data for minimal reproducible example of block complexity metrics.