Replication package for "Same score, different institutional orders: academic selection, school segregation and the comparative reading of PISA's equity gradient"

This package contains derived, system-level estimates computed from the OECD PISA 2025 public-use student and school files (OECD 2026a). The student-level microdata are not included and must be downloaded from the OECD. The estimates cover 84 education systems: 649,174 students in 21,273 schools with at least ten students with a valid ESCS. Of the 90 systems in the PISA 2025 files, 6 are excluded: Costa Rica (no ESCS) and Iceland, Luxembourg, Malta, El Salvador and Macao (China), whose school identifiers are not in the public-use files. Every estimate uses final student weights, 80 Fay balanced repeated replication weights and ten plausible values combined by Rubin’s rules. Gradients are in science points per ESCS unit, where one ESCS unit is one OECD student standard deviation of the PISA index of economic, social and cultural status. P3_Data_and_Figures.xlsx (13 sheets) Systems: one row per system (84). It holds: Identifiers and flags: system code and name, World Bank region, OECD membership, paper-based assessment, and subnational or partial unit. Sample and scores: students and schools in the main sample, mean and standard deviation in science. Gradient estimates with standard errors: the total gradient b_T, the within-school gradient b_W and the between-school slope b_B. Segregation: η², the between-school share of ESCS variance. Formula columns: the within-school part (1 − η²)b_W, the between-school part η²b_B, an identity check showing the two parts sum to b_T, the contextual association b_B − b_W, the within-school share (shown where b_T is at least 10 points) and the map class. School-level shares: students in selective, vocational, ability-grouped and private schools, and school autonomy. Two-level model results: the standard deviation of school gradients τ₁, the intercept–slope correlation ρ₀₁ and the reliability of school slopes. Quartile: each system’s quartile of segregation. Summary block: formulas reproducing the medians and counts reported in the article. Figure1_data, Figure2_data, Figure3_data: the values plotted in each figure. They also include the rule-based contrast pairs, the map class counts and the random-effects meta-regression slopes by group. Table1_quartiles: Table 1’s quartile medians recomputed with formulas, next to the published values. Table1, Table2, Table3: the tables as printed in the article. Figure1, Figure2, Figure3, Graphical_abstract: the figures. In the workbook, estimates from the analysis code are shown in blue and Excel formulas in black, so the decomposition identity can be checked directly. figures/: Figures 1–3 as PNG and TIFF (600 dpi) and as vector PDF, plus the graphical abstract (PNG, 525 px wide). The world map in Figure 2 uses Natural Earth 1:50m boundaries (world-atlas 2.0.2) in the Equal Earth projection. code/: 31 scripts (30 Python, 1 R). They extract the variables from the raw OECD files, build the analysis data and estimate every result. They also run the integrity checks and build the tables, figures, supplement and this workbook. The run order and the software requirements are in README.md. Source: OECD (2026a), PISA 2025 Database, Paris: OECD. Calculations by the author.