# Supplementary Dataset 1

## Solar and Wind Expansion and Uneven Fossil-Generation Displacement Across 29 Emerging and Developing Economies

**Author:** Rasool Bux Khoso  
**Affiliation:** Department of Earth Sciences, University of Turin, Turin, Italy  
**Contact:** rasool.khoso@edu.unito.it  
**Dataset file:** `Supplementary_Data_1.xlsx`  
**Version date:** 24 July 2026  
**File size:** 725,091 bytes  
**SHA-256:** `7de8f9727bca3e57a12f3f9c3efffcc8c3773293249b6e942e9b37d705a4eedc`  
**Repository DOI:** To be assigned on Zenodo  
**Associated manuscript DOI:** To be added after publication

## 1. Purpose and scope

This workbook provides the machine-readable data and outputs supporting the manuscript **“Solar and Wind Expansion and Uneven Fossil-Generation Displacement Across 29 Emerging and Developing Economies.”** It contains:

- the harmonized country-year electricity and socioeconomic panel;
- the annual-change analytical panel;
- derived accounting indicators, including the gross fossil-displacement ratio and clean-growth coverage;
- transition classifications and the binary fossil-decline outcome;
- out-of-fold predictions and fold-specific logistic-regression coefficients;
- fixed-effects, robustness, source-consistency, and data-quality outputs;
- source data underlying all manuscript figures and tables; and
- machine-readable versions of Supplementary Tables S1–S12.

The balanced level panel covers **29 countries from 2010 to 2024** and contains **435 country-year observations**. The annual-change panel covers **2011 to 2024** and contains **406 country-year observations**.

The country sample comprises Algeria, Argentina, Bangladesh, Brazil, Chile, Colombia, Egypt, Ethiopia, Ghana, India, Indonesia, Jordan, Kazakhstan, Kenya, Malaysia, Mexico, Morocco, Nepal, Nigeria, Pakistan, Peru, the Philippines, South Africa, Sri Lanka, Thailand, Tunisia, Türkiye, Uzbekistan, and Viet Nam.

## 2. Workbook structure

The workbook contains 30 worksheets. The sheets are grouped below by function.

### 2.1 Core analytical data

| Worksheet | Description |
|---|---|
| `Main_panel` | Balanced country-year level panel for 2010–2024. Contains electricity generation, demand, imports, emissions, carbon intensity, socioeconomic variables, generation shares, annual changes, lagged variables, normalized changes, accounting indicators, transition outcomes, and the binary fossil-decline outcome. |
| `Annual_changes` | Annual-change panel for 2011–2024. This is the principal analytical dataset used for first-difference regressions, transition accounting, prospective classification, ex-post diagnostic classification, and robustness analyses. |
| `OOF_predictions` | Concatenated leave-one-country-out out-of-fold probabilities and binary predictions for the prospective structural, ex-post diagnostic, and full-balance sensitivity logistic models. |
| `Fold_coefficients` | Fold-specific standardized logistic-regression coefficients estimated after excluding each held-out country. |
| `Data_quality` | Row-count, country-count, year-count, duplicate, and core-variable completeness checks used to verify the balanced panel. |

### 2.2 Supplementary tables

| Worksheet | Supplementary table | Description |
|---|---|---|
| `S1` | Table S1 | Country-level cumulative solar-wind, fossil-generation, and demand changes; cumulative GDR and CGC; and counts of fossil-decline and positive solar-wind-growth years. |
| `S2` | Table S2 | Equal-weighted summary across the 29 countries, including median cumulative changes, median GDR and CGC, and the number and share of countries with cumulative fossil decline. |
| `S3` | Table S3 | Regional cumulative electricity changes and pooled GDR and CGC values for six regional groups. |
| `S4` | Table S4 | Restricted Rademacher wild-cluster-bootstrap inference for the two principal fixed-effects coefficients, based on 999 replications. |
| `S5` | Table S5 | Variance inflation factors and the standardized condition index for the principal fixed-effects predictor set. |
| `S6` | Table S6 | Leave-one-country-out fixed-effects coefficient stability results. |
| `S7` | Table S7 | Fold-level classification thresholds for all held-out countries and model designs. A fixed threshold of 0.50 was used. |
| `S8` | Table S8 | Majority-class classification benchmark for the binary fossil-decline outcome. |
| `S9` | Table S9 | Conditional positive-solar-wind-growth prospective sensitivity analysis, with country-bootstrap confidence intervals based on 500 replications. |
| `S10` | Table S10 | Source-consistency comparison between Ember and Our World in Data electricity series. These are consistency checks rather than independent validation because OWID electricity series partly incorporate Ember data. |
| `S11` | Table S11 | Complete country-year electricity-balance residuals for the 406 annual-change observations. |
| `S12` | Table S12 | Expanded variable dictionary with definitions, units, sources, and construction procedures. |
| `S7_Summary` | Compact S7 | Compact summary showing the fixed 0.50 threshold for each of the three primary classification designs. |
| `S11_Summary` | Compact S11 | Summary statistics for the electricity-balance residuals; the complete country-year records remain in `S11`. |

### 2.3 Figure source data

| Worksheet | Figure | Description |
|---|---|---|
| `Figure_1_source` | Figure 1 | Annual aggregate electricity demand, fossil generation, solar-wind generation, and corresponding annual changes. |
| `Figure_2_source` | Figure 2 | Overall and regional counts and shares of the four mutually exclusive transition outcomes. |
| `Figure_3_source` | Figure 3 | Country-level cumulative solar-wind, fossil-generation, and demand changes, with cumulative GDR and CGC. |
| `Figure_4_curves` | Figure 4a–b | Receiver-operating-characteristic and grouped calibration coordinates for the classification models. |
| `Figure_4_coefficients` | Figure 4c | Fold-level coefficient summaries used to display the principal ex-post diagnostic predictors. |

### 2.4 Main-manuscript table source data

| Worksheet | Description |
|---|---|
| `Table_1_source` | Analytical samples and data coverage. |
| `Table_2_source` | Aggregate electricity-system changes between 2010 and 2024. |
| `Table_3_source` | Distribution of transition outcomes among positive solar-wind-growth observations. |
| `Table_4_source` | Fixed-effects coefficients, confidence intervals, clustered P values, Driscoll–Kraay P values, and within R² values. |
| `Table_5_source` | Leave-one-country-out classification performance for the prospective, ex-post diagnostic, and full-balance sensitivity designs. |

### 2.5 Workbook documentation

| Worksheet | Description |
|---|---|
| `00_README` | Short in-workbook summary of the dataset, sources, missing-value conventions, model threshold, bootstrap settings, and worksheet contents. |

## 3. Data sources

The analytical panel was constructed from the following public sources.

### 3.1 Ember Yearly Electricity Data

**Use in this study:** Electricity generation by source, total domestic generation, electricity demand, net imports, power-sector emissions, and carbon intensity.

**Official source:**  
https://ember-energy.org/data/yearly-electricity-data/

**Expected source filename:**  
`yearly_full_release_long_format.csv`

### 3.2 World Bank World Development Indicators

**Use in this study:** GDP growth, real GDP per capita, population growth, urban population share, industry value added, manufacturing value added, electricity access, and related socioeconomic variables.

**Official source:**  
https://databank.worldbank.org/source/world-development-indicators

The analysis code accepts one of the following input formats:

- `wdi_selected_panel.csv`;
- `WDIEXCEL.xlsx`; or
- `WB_WDI_WIDEF.csv`.

### 3.3 Our World in Data Energy Dataset

**Use in this study:** Supplementary source-consistency checks for electricity and emissions variables.

**Official source:**  
https://github.com/owid/energy-data

**Expected files:**

- `owid-energy-data.csv`;
- `owid-energy-codebook.csv` (optional documentation file).

The OWID comparison is not treated as independent validation because OWID electricity series partly incorporate Ember data.

## 4. Key variable definitions

The complete variable dictionary is provided in worksheet `S12`. The principal variables are summarized below.

### 4.1 Electricity variables

| Variable | Definition | Unit |
|---|---|---|
| `demand_twh` | Annual electricity demand | TWh |
| `total_generation_twh` | Annual domestic electricity generation | TWh |
| `net_imports_twh` | Net electricity imports; positive values indicate net imports | TWh |
| `fossil_twh` | Electricity generation from coal, gas, oil, and other fossil fuels | TWh |
| `wind_solar_twh` | Combined solar and wind electricity generation | TWh |
| `hydro_twh` | Hydropower generation | TWh |
| `nuclear_twh` | Nuclear generation | TWh |
| `other_clean_twh` | Clean generation excluding solar-wind, hydropower, and nuclear | TWh |
| `total_emissions_mtco2` | Power-sector greenhouse-gas emissions | Mt CO₂ |
| `co2_intensity_g_per_kwh` | Power-sector carbon intensity | g CO₂ kWh⁻¹ |

### 4.2 Annual changes

Variables beginning with `d_` are annual first differences. For example:

- `d_fossil_twh` = fossil generation in year *t* minus fossil generation in year *t − 1*;
- `d_wind_solar_twh` = solar-wind generation in year *t* minus solar-wind generation in year *t − 1*;
- `d_demand_twh` = electricity demand in year *t* minus electricity demand in year *t − 1*.

Positive values indicate annual increases and negative values indicate annual declines.

### 4.3 Gross fossil-displacement ratio

The gross fossil-displacement ratio is defined only for observations with positive solar-wind growth:

```text
GDR = -Δ fossil generation / Δ solar-wind generation
```

Interpretation:

- `GDR > 1`: fossil decline exceeds the solar-wind increment;
- `0 < GDR < 1`: fossil generation declines, but by less than solar-wind growth;
- `GDR = 0`: no fossil-generation change;
- `GDR < 0`: fossil generation and solar-wind generation expand concurrently.

GDR is a descriptive accounting ratio, not a causal efficiency estimate. It should be interpreted together with its numerator, denominator, demand growth, and changes in other electricity-balance components.

### 4.4 Clean-growth coverage

Clean-growth coverage is defined for observations with positive electricity-demand growth:

```text
CGC = Δ solar-wind generation / Δ electricity demand
```

Interpretation:

- `CGC >= 1`: solar-wind growth equals or exceeds incremental electricity demand;
- `CGC < 1`: solar-wind growth does not fully cover incremental demand.

### 4.5 Transition outcomes

For observations with positive solar-wind growth, `transition_outcome` assigns one of four mutually exclusive categories:

1. **Fossil decline exceeding solar-wind growth**;
2. **Fossil decline below solar-wind growth**;
3. **Demand-dominant co-expansion**; or
4. **Fossil co-expansion despite demand coverage**.

### 4.6 Binary classification outcome

`fossil_decline` equals:

- `1` when annual fossil generation declines; and
- `0` when fossil generation is unchanged or increases.

This binary outcome is used for the prospective structural, ex-post diagnostic, and full-balance sensitivity logistic classifiers.

### 4.7 Lagged and normalized variables

- Variables beginning with `lag_` or `lag1_` are one-year lags.
- Variables beginning with `log_` use natural logarithms.
- Variables ending in `_pct_lag_demand` are calculated as:

```text
100 × annual change / electricity demand in the preceding year
```

These normalized variables are expressed in percentage-point-equivalent units relative to lagged demand.

## 5. Units and coding conventions

| Data type | Unit or coding |
|---|---|
| Electricity generation, demand, and imports | TWh |
| Power-sector emissions | Mt CO₂ |
| Carbon intensity | g CO₂ kWh⁻¹ |
| Generation shares and socioeconomic rates | % |
| Normalized annual changes | 100 × change / lagged demand |
| GDR and CGC | Unitless ratios |
| Binary outcomes and predictions | 0 or 1 |
| Predicted probabilities | 0 to 1 |
| Country identifier | ISO alpha-3 code |
| Year | Four-digit calendar year |
| Log-transformed variables | Natural logarithm |

## 6. Missing values and special treatments

Blank cells represent values that are unavailable, undefined, or inapplicable.

Important conventions are:

- Core electricity variables are complete for the balanced 29-country panel.
- Annual-change and lagged variables are blank for 2010 because a preceding year is unavailable within the study window.
- `gdr` is blank when solar-wind growth is zero or negative.
- `cgc` is blank when electricity-demand growth is zero or negative.
- `transition_outcome` is blank when solar-wind growth is zero or negative.
- Countries with no nuclear generation throughout the period had an all-missing nuclear series in the source data; these complete no-nuclear series were set to zero.
- No imputation was applied to the core electricity variables.
- Isolated missing socioeconomic predictor values used in classification were imputed with the median calculated only from the training countries within each leave-one-country-out fold.
- Predictor standardization for logistic classification was also estimated separately within each training fold to prevent leakage.

## 7. Classification designs

Three logistic-regression designs are reported.

### 7.1 Prospective structural model

Uses only one-year-lagged electricity-system and socioeconomic characteristics. It is intended to test out-of-country predictive discrimination using information available before the annual fossil outcome is observed.

### 7.2 Ex-post diagnostic model

Adds contemporaneous changes in electricity demand, solar-wind generation, and hydropower, normalized by lagged demand. It is an ex-post diagnostic model and should not be interpreted as a prospective forecast or causal model.

### 7.3 Full-balance sensitivity model

Further adds normalized nuclear-generation and net-import changes. Because these variables approach reconstruction of the annual electricity balance, this design is treated as a sensitivity analysis rather than an independent forecasting model.

All three designs use:

- leave-one-country-out validation;
- logistic regression with L2 regularization;
- training-fold median imputation;
- training-fold standardization;
- a fixed classification threshold of 0.50; and
- a fixed random seed of 42.

## 8. Reproduction workflow

### 8.1 Required files

Place the following files in the Google Colab `/content` directory or upload them through the Colab file-selection dialogue:

1. `yearly_full_release_long_format.csv`;
2. one World Bank input file:
   - `wdi_selected_panel.csv`, or
   - `WDIEXCEL.xlsx`, or
   - `WB_WDI_WIDEF.csv`;
3. `owid-energy-data.csv` for the source-consistency checks; and
4. `owid-energy-codebook.csv` if desired for documentation.

### 8.2 Software files

The reproducibility package should contain:

- `GDR_Script.ipynb`;
- `gdr_script.py`;
- `README.md`;
- `requirements.txt`; and
- `environment.yml`.

### 8.3 Google Colab workflow

1. Open `GDR_Script.ipynb` in Google Colab.
2. Upload the required source files to `/content`.
3. Run the notebook from the first cell to the final cell without changing the default analytical settings.
4. The notebook creates the project directory:

```text
/content/GDR_Files/
```

5. Outputs are written to:

```text
/content/GDR_Files/output/data/
/content/GDR_Files/output/tables/
/content/GDR_Files/output/figures/
/content/GDR_Files/output/models/
```

6. The notebook creates a ZIP archive named:

```text
GDR_Reproducibility_Package.zip
```

### 8.4 Local environment

Using Conda:

```bash
conda env create -f environment.yml
conda activate gdr-analysis
```

Using pip:

```bash
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
```

The current script is optimized for Google Colab and uses `/content` paths. For local execution, update the input and project directory paths in the configuration section before running the script.

### 8.5 Default analytical settings

| Setting | Default |
|---|---:|
| Study period | 2010–2024 |
| Countries | 29 |
| Random seed | 42 |
| Wild-cluster-bootstrap replications | 999 |
| Country-bootstrap replications for primary classification models | 2,000 |
| Country-bootstrap replications for conditional positive-growth sensitivity | 500 |
| Driscoll–Kraay lags | 2 |
| Classification threshold | 0.50 |
| Threshold tuning | Disabled |
| Tree-model sensitivity | Disabled |
| TIFF output | Disabled |

### 8.6 Expected verification checks

A successful run should reproduce the following structural checks:

| Check | Expected result |
|---|---:|
| Countries in balanced panel | 29 |
| Years in level panel | 15 |
| Level-panel rows | 435 |
| Annual-change rows | 406 |
| Duplicate country-years | 0 |
| Missing core level values | 0 |
| Positive solar-wind-growth observations | 317 |
| Fossil-decline observations | 129 |
| Maximum absolute electricity-balance residual | Approximately 2.84 × 10⁻¹³ TWh |

Small differences in the final displayed decimal place may occur because of operating-system, library, or floating-point differences. Material differences should be investigated before the outputs are used.

## 9. Relationship to submitted files

This workbook is intended to accompany:

- the main manuscript;
- the Supplementary Information PDF containing formatted Tables S1–S12; and
- the software repository or Supplementary Software 1 containing the notebook, Python script, environment files, and execution instructions.

The workbook contains machine-readable source values. The formatted Supplementary Information PDF should be used for reading the table notes and presentation-ready versions of the supplementary analyses.

## 10. Citation

Until a repository DOI is assigned, cite the associated manuscript and identify the workbook as Supplementary Dataset 1.

## 11. License and reuse

The original compilation, harmonisation, documentation, derived variables, analytical outputs, and figure and table source data in this repository are licensed under the Creative Commons Attribution 4.0 International licence
(CC BY 4.0).

Users may share and adapt these materials for any purpose, provided appropriate credit is given to the author, a link to the licence is supplied, and any changes are indicated.

The underlying source data remain subject to the licences and terms of their original providers, including Ember, the World Bank World Development Indicators, Our World in Data, and any original third-party data providers. Users should cite both this repository and the relevant original data sources.

## 12. Contact

Questions about the dataset or reproduction workflow may be directed to:

**Rasool Bux Khoso**  
Department of Earth Sciences  
University of Turin  
Turin, Italy  
Email: rasool.khoso@edu.unito.it