Introduction

The UN E-Government Development Index (EGDI) measures digital governance maturity biennially for 193 countries. We ask: how much of a country's EGDI score is explained by its socioeconomic development level, and which countries significantly over- or under-perform this expectation?

We deliberately exclude features that overlap with EGDI components (internet penetration is part of the Telecommunication Infrastructure Index; years of schooling is part of the Human Capital Index). Using only four non-overlapping socioeconomic indicators — GDP per capita, corruption perceptions, urbanization, and government expenditure — a Random Forest model explains 93.5% of 2022 EGDI variance (R² = 0.935) on a held-out test set.

We compare against two baselines: linear regression (R² = 0.778) and a persistence model using prior EGDI scores (R² = 0.987). The persistence model is a better forecaster, but it cannot explain why countries achieve their scores or identify policy-driven outperformance. Our model's residuals reveal countries whose digital maturity exceeds socioeconomic expectations — quantifying the impact of deliberate policy intervention.

Data

Target: EGDI scores from UN DESA E-Government Survey (2018, 2020, 2022).

Features (4, non-overlapping with EGDI):

Features deliberately excluded to avoid circularity:

Sample: 54 countries spanning all income groups and regions (76% of world population, 89% of world GDP). Selected for data completeness across all three survey years.

Split: Train on 2018+2020 (104 observations). Test on 2022 (52 observations, strictly held out).

Model and Baselines

The persistence model achieves R² = 0.987 because EGDI scores are highly autocorrelated between surveys. We do not claim to outperform it for forecasting. Our contribution is explanatory: the Random Forest captures non-linear relationships between development indicators and digital maturity that linear regression misses (R² 0.935 vs 0.778), and its residuals identify policy-driven outperformance.

Random Forest: 200 trees, max depth 8, 3 minimum samples per leaf, pure NumPy implementation. No scikit-learn dependency to minimize execution requirements, though scikit-learn would produce equivalent results.

Results

Feature Importance

GDP per capita and governance quality (CPI) jointly account for 92.8% of the model's explanatory power. Government spending level and urbanization contribute minimally, suggesting that digital governance maturity is driven by economic capacity and institutional quality rather than fiscal effort or population distribution.

Residual Analysis: Policy Outperformers

Countries with large positive residuals achieve EGDI scores significantly above their socioeconomic prediction — the gap is attributable to factors outside the four indicators, plausibly including deliberate digital transformation policy.

Top positive residuals (outperformers):

Top negative residuals (underperformers):

Saudi Arabia's residual of +0.075 is the largest in the dataset. Its EGDI score (0.880) is 7.5 points higher than what GDP per capita ($30,436), CPI (51), urbanization (84%), and government expenditure (31%) would predict. This residual is consistent with — though not proof of — the impact of Vision 2030 digital transformation investments. A causal interpretation would require controlling for additional confounders.

Full Country Predictions

The model predicts 2022 EGDI within ±0.04 for 35 of 52 countries (67%).

Related Work

Predictive modeling of e-government development has been explored through several approaches. Krishnan et al. (2013, Government Information Quarterly 30(4)) analyzed institutional and economic factors influencing e-government maturity using structural equation modeling across 72 countries. Zhao et al. (2014, GIQ 31(4)) examined how national governance quality affects e-government performance. Ingrams et al. (2020, GIQ 37(3)) studied the relationship between government transparency and EGDI scores. Our work differs by using non-linear machine learning rather than linear econometric methods, and by focusing on residual analysis to identify policy outperformers rather than causal inference.

Limitations

1. 54 of 193 countries (28%). Expanding to full UN membership would improve generalizability. Countries were selected for data completeness, introducing potential selection bias toward data-rich nations.
2. Persistence baseline is stronger for forecasting. Our model's value is explanatory (identifying outperformers), not predictive. For pure EGDI forecasting, prior scores are more accurate.
3. Residual interpretation is associative, not causal. Saudi Arabia's positive residual is consistent with policy impact but could reflect unmeasured variables (oil wealth effects, expatriate workforce composition, measurement methodology differences).
4. No cross-validation on the temporal split. A single train/test split provides one estimate of generalization error. Additional survey years (2024) would enable more robust temporal evaluation.
5. Embedded dataset is a snapshot. Values should be verified against primary UN DESA and World Bank publications for operational use.

Conclusion

Using four non-overlapping socioeconomic indicators, a Random Forest model explains 93.5% of 2022 EGDI variance — substantially more than linear regression (77.8%) and without the circularity of using EGDI component features as predictors. While the model does not outperform a simple persistence baseline for forecasting, its residuals identify countries whose digital maturity exceeds development-level expectations. Saudi Arabia shows the largest positive residual (+0.075), consistent with the impact of deliberate digital transformation policy.

References

1. UN DESA, "E-Government Survey 2018," United Nations, 2018.
2. UN DESA, "E-Government Survey 2020," United Nations, 2020.
3. UN DESA, "E-Government Survey 2022," United Nations, 2022.
4. World Bank, "World Development Indicators," 2024.
5. IMF, "World Economic Outlook Database," October 2024.
6. Transparency International, "Corruption Perceptions Index," 2018-2022.
7. Breiman L., "Random Forests," Machine Learning 45(1), pp. 5-32, 2001.
8. Krishnan S., Teo T., Lim V., "Examining the Relationships Among E-Government Maturity, Corruption, Economic Prosperity, and Environmental Degradation," Information & Management 50(8), 2013.
9. Zhao F., Collier A., Deng H., "A Multidimensional and Integrative Approach to Study Global Digital Divide and E-Government Development," Information Technology & People 27(1), 2014.
10. Ingrams A., Piotrowski S., Berliner D., "Learning from Our Mistakes: Public Management Reform and the Hope of Open Government," Perspectives on Public Management and Governance 3(4), 2020.