SUNDE T WORLD DEVELOPMENT
Description
The empirical analysis combines annual country-level data from the United Nations Conference on Trade and Development (UNCTAD) Data Hub, the World Bank’s World Development Indicators (WDI), International Telecommunication Union (ITU) indicators accessed through WDI/ITU, and the Worldwide Governance Indicators (WGI). The study population comprises economies classified as developing by UNCTAD, while regional and subregional groupings follow the United Nations M49 classification. The candidate sample contains 74 developing economies observed over 2010–2024. The starting year is determined by the availability and comparability of the digitally deliverable services export series used as the principal explanatory variable, giving a maximum of 1,110 country-year observations. The main analysis uses an unbalanced panel rather than imposing a balanced-panel requirement that would discard otherwise usable within-country information. No interpolation or imputation is undertaken. The main estimation sample is determined by the joint availability of RGPC, LDSE, DINF and the six baseline controls and contains 918 country-year observations for 71 economies; Jamaica, Jordan and Nigeria contribute no complete baseline observations. DSEX is available for 1,065 candidate observations, while DSEX and the variables required to construct DINF are jointly available for 1,017 observations. The balanced panel used only for the threshold test contains 38 economies observed every year, yielding 570 observations. Model-specific complete cases are used elsewhere so that missingness, rather than an arbitrary minimum-history rule, determines the estimation sample. Before estimation, country-year identifiers were checked for duplication, numeric variables were parsed consistently, and formula-derived variables, including LPRO, DSGD and INST, were reconstructed from their underlying series. The baseline data are not winsorised; tail treatment is assessed separately as a robustness exercise. These procedures preserve the sample and transformations used in the reported results.
Files
Steps to reproduce
The study uses a reproducible country-year panel compiled from publicly available international databases for 2010–2024. The starting population comprised 74 economies classified as developing by UNCTAD. Data were merged by country and year from the UNCTAD Data Hub, World Bank World Development Indicators (WDI), International Telecommunication Union (ITU) indicators accessed through WDI/ITU, and the Worldwide Governance Indicators (WGI). No surveys, laboratory instruments or reagents were used. The dependent variable is annual real GDP per capita growth from WDI. Digitally deliverable services exports (DSEX) were obtained from UNCTAD in current US$ millions. Because DSEX is strongly right-skewed, the main export variable was transformed as LDSE = ln(1 + DSEX). Alternative export measures were also constructed for robustness checks. Digital connectivity infrastructure (DINF) was constructed using principal component analysis (PCA) from three standardised indicators: individuals using the Internet (% of population), fixed-broadband subscriptions per 100 people, and mobile-cellular subscriptions per 100 people. PCA was estimated using observations from 2010–2024 only. The first principal component explained 72.95% of common variance and was restandardised so that higher values represent stronger connectivity. A four-component alternative index adding secure Internet servers was also constructed. Baseline controls from WDI include gross capital formation, FDI inflows, trade openness, inflation, government consumption and population growth. Additional specifications incorporated education and institutional indicators from WDI and WGI. Formula-derived variables were reconstructed programmatically in R. Country-year identifiers were checked for duplicates and numeric variables were parsed consistently before estimation. No missing observations were imputed and the baseline data were not winsorised. Complete-case availability produced the main unbalanced panel of 918 observations across 71 countries. The balanced sample used for threshold analysis contained 570 observations across 38 countries. The primary analysis used two-way country and year fixed-effects regressions with country-clustered HC1 standard errors. LDSE and DINF were mean-centred before constructing their interaction term. Robustness procedures included alternative outcomes and explanatory measures, alternative covariance estimators, a wild-cluster bootstrap, threshold estimation, lagged and differenced specifications, forward three- and five-year growth outcomes, sample restrictions, influence checks and two-step System GMM diagnostics. Reported hypothesis conclusions were based on the primary specifications rather than selectively significant robustness results. These procedures were implemented consistently across specifications so that changes in results could be traced to measurement, timing, sample composition, inference method or estimator rather than undocumented data handling
Institutions
- Namibia University of Science and TechnologyKhomas Region, Windhoek