Caveats & limitations

All ISORA indicators are submitted by each jurisdiction’s tax administration. There is no central audit or harmonisation step — definitions, scope and reporting practices vary across jurisdictions. Cross-country comparisons should be read as approximate; large outliers may reflect reporting conventions or unit-of-measure errors rather than substantive differences in administration.

The PIT-administration maturity score is an AHP-weighted composite of four pillars — Assessment & Filing, Enforcement, Digital Transformation, Registration — each pillar averaging a handful of standardised ISORA / ITTI features. The pillar weights (Filing > Enforcement > Digital > Registration) come from a Saaty pairwise matrix and are a judgement call; an equal-weighted variant produces a Spearman-rank-correlated ranking (ρ ≈ 0.98), so the headline ordering is robust to the weighting scheme. Residuals from the regression on log₁₀ GDP per capita carry composite noise plus regression noise; values within ±0.5σ sit inside that noise band and shouldn’t be over-interpreted.

An unsupervised K-Means run on the same standardised feature matrix was retained as a diagnostic. Clusters do not separate cleanly into the four GNI tiers (ARI ≈ 0.14 vs four-tier GNI, silhouette ≈ 0.11), and silhouette is flat across k = 2–8 (best k by silhouette = 2). The headline OLS regression of the maturity composite on log GDP per capita is therefore preferred: PIT-administration capability behaves as a continuous gradient anchored on income, not a tiered system.

Across the 102 in-sample jurisdictions GDP per capita spans roughly $300–$100,000+ (a ~300× range), is heavily right-skewed, and shows a multiplicative rather than additive relationship with administrative maturity — equal ratio steps in income (every doubling, every tenfold) tend to buy roughly equal gains in maturity, while equal dollar steps do not. Taking the base-10 logarithm linearises the relationship (so OLS is the right tool), homogenises residual variance (so the ±1σ band is meaningful across the full x-range), and lets the slope read naturally as “σ change in maturity per tenfold change in GDP per capita”. Trade-off: visual compression understates how large the absolute income gap is between low- and high-income countries. The transformation is standard practice in cross-country development economics.

The maturity-vs-income scatter uses log GDP per capita as the continuous x-axis and the World Bank’s GNI per capita tier (Atlas method) as the categorical colour. The two measures are rank-correlated at >0.99 across the 102 in-sample jurisdictions, so the regression slope and residual ordering are robust to which is used. A small number of jurisdictions with significant cross-border income flows — multinational / finance hubs (e.g. Ireland, Singapore) where GDP/cap exceeds GNI/cap, and remittance economies (e.g. Tajikistan, Nepal) where GNI/cap exceeds GDP/cap — sit at x-positions that diverge by 10–25% from where GNI alone would place them. This is acknowledged but does not materially affect the residual ordering.

In creating this dashboard, I collaborated with Claude (Anthropic’s AI assistant) to assist with code drafting, design polish, written-content drafting and editing, and debugging. I affirm that all AI-generated and co-created content underwent thorough review and evaluation. The final output accurately reflects my understanding, expertise, and intended meaning. While AI assistance was instrumental in the process, I maintain full responsibility for the content, its accuracy, and its presentation. This disclosure is made in the spirit of transparency and to acknowledge the role of AI in the creation process.