Do infrastructure and human capital work together for environmental sustainability in Sub-Saharan Africa? This observatory publishes the panel data, the econometric estimates and a scenario simulator behind the study, so every number can be inspected and reused.
A one-standard-deviation rise in the infrastructure index is associated with +1.2 percentage points of land under arable use. The link holds in every fixed-effects specification, when any single country is dropped and at lags of one to three years.
Resource rents rise by about 8–10% per SD of infrastructure, with p between 0.05 and 0.13. Emissions rise with infrastructure only in the broad-control sample.
The INFRA × HCR interaction is negative for resource rents (p = 0.02 with broad controls, 0.09 with full controls), which is the pressure-reducing direction. It does not survive every leave-one-country-out check, and there is no such synergy for emissions.
| Hypothesis | Status | Evidence |
|---|---|---|
| H1 · Infrastructure improves sustainability (reduces pressure) | Not supported | Infrastructure is associated with more pressure: more arable land (robust), higher resource rents (borderline) and higher emissions (broad samples only). |
| H2 · Human capital strengthens that effect | Weak, partial | A pressure-reducing interaction for resource rents that is sensitive to sample and to which countries are included. None for emissions or land in the fixed-effects models. |
| H3 · Institutional quality moderates the effects | Pending | WGI-based INST index not yet merged. |
Grey countries have no observation for the selected year. The colour scale is fixed across years and capped at the 98th percentile. Click a country to add it to the comparison.
The panel is unbalanced. Emissions (THE) and energy mix are observed mostly after 2000, which shrinks any model that uses them. Coverage matters for how much weight each result can bear.
Points are coefficients on standardised predictors with 95% intervals. Filled markers: p < 0.05. Hollow: not significant.
Quantile estimates (QR) pool within- and between-country variation and are not directly comparable in size with fixed-effects (FE) estimates.
Two checks on the preferred model (two-way fixed effects, full controls, Driscoll–Kraay SEs). Leave-one-country-out re-estimates the model 40-plus times, dropping one country each time, to see whether any single country drives the result. Lagged infrastructure asks whether the association appears in the same year or builds over one to three years. Predictors are standardised once, on the full sample, so coefficients are comparable across runs.
Each point is the estimate with the named country omitted, with its 95% interval. The solid line is the full-sample estimate and the shaded band its 95% interval. Filled markers: p < 0.05.
Own sample: each model uses every observation available for that lag. Common sample: all five models use the same observations (INFRA at t to t−3, the outcome and all controls observed), so differences reflect timing rather than sample composition.
*** p < 0.01, ** p < 0.05, * p < 0.10. The interaction uses the lagged INFRA times current HCR.
Translate the estimates into a scenario. Choose a country and change the levers; the simulator applies the estimated coefficients to that country's latest observed values and reports the implied change with a 95% interval. These are statistical associations, not causal forecasts.
Orange bars raise environmental pressure; blue bars lower it. Each bar applies one lever on its own, so bars need not sum exactly to the total.
Every variable is screened for coverage, persistence and plausibility. Real macro indicators change slowly, so their year-to-year (lag-1) autocorrelation within a country is typically above 0.8. A value near zero indicates noise. The original workbook indices (rows marked "superseded") failed these checks and were rebuilt from their public sources.
Both series are standardised within the country so they share one axis.
An extended STIRPAT model with country and year fixed effects:
Yit = αi + λt + β1INFRAit + β2HCRit + β3(INFRA×HCR)it + γ′Xit + εit
INFRA: electricity access (WDI), mobile subscriptions, internet users and fixed telephone lines (ITU via WDI). HCR: mean and expected years of schooling (UNDP HDR), tertiary and secondary gross enrolment (UNESCO UIS via WDI). Each component is z-scored within year across the 49 countries, and the index is the mean of the available z-scores, with at least two required. Gaps inside a country's series of up to five years are interpolated linearly. Series are never extrapolated.
For lever changes Δx (in standard-deviation units), the predicted change in the transformed outcome is Δŷ = β′Δx, including the change in the INFRA × HCR product. The 95% interval uses the model's variance–covariance matrix. Results are back-transformed to natural units at the country's latest observed value.
The pipeline (analysis/fetch_sources.py → build_indices.py → build_panel.py → estimate.py) downloads the source series and regenerates every file above. Source code: GitHub repository.
Tackie, G., Adams, J., Maisuh, A., & Moore, S. E. (2026). Paving an extractive path? Infrastructure, human capital and environmental pressure in Sub-Saharan Africa. University of Cape Coast.
Supported by a Group-led Research Support Grant from the Directorate of Research, Innovation and Consultancy (DRIC), University of Cape Coast.