The core analysis uses one official public OECD dataset. Here are the origins of each data type, how the fields are linked, the limits of the evidence, and the files needed to reproduce the work.
Check uniqueness of CNTSTUID
Skills + family + stratum + weights
Every required field is in the same Student PUF. The core paper does not require inferred matches between examinations, schools and air-raid alerts, or linkage of personal registries.
STRATUM is interpreted using value labels in the original SAV dictionary. Each of the 33 strata maps unambiguously to a rural or urban school category. CNTSCHID verifies the school count; CNTSTUID verifies student uniqueness.
The archive was downloaded and extracted, and the SAV file read. Checks cover unique rows, stratum labels, completeness of skills and weights, and missing family characteristics.
Separate code independently reproduced 20 main estimates and their standard errors. The largest discrepancy was 1.4 × 10⁻¹⁶. This verifies computation, not causality.
| Field | Purpose | Verification |
|---|---|---|
| PV1MATH…PV10MATH | Mathematics skills and their uncertainty | All 10 values for all 7,105 students |
| W_FSTUWT + 80 W_FSTURWT | Weighted estimates and standard errors | 81 columns without missing values |
| STRATUM | Priority by school stratum | 33 labels; 1,619 rural rows |
| ESCS | Family-resource index | 495 missing values; weighted share 7.11% |
| HISCED | Highest parental education | 801 missing values; weighted share 11.46% |
CY09_MS_STU_PUF.sav · 2,168,666,883 bytes
6f6fa1e1ecce23d39183d44ec82ba05b245fe88a2416c8d8dc1b016d556fc379The ZIP requires a Deflate64-compatible extractor. File size alone does not confirm a complete download: verify the hash after extraction.
Compare expected coverage of need under different rules in the represented population. Assess sensitivity to capacity and incomplete information. Calculate a conditional spending limit for additional screening.
Estimating the causal effect of war, actual learning gains from tutoring, or future earnings. Extending results to excluded territories, children outside school, or all refugees. Treating ESCS as an administrative income register.
Primary source: the Student Public Use File, containing skills, family characteristics, strata and survey weights.
Description of the Ukrainian population covered. This defines limits on generalization; it does not justify extending estimates to the whole country.
Guidance on plausible values and replicate weights.
Published notes on the analysis of results. We do not claim to have read the forthcoming full PISA 2025 technical report.
PISA 2022 technical report: the Level 2 mathematics threshold of 420.07 on the common scale.
We bring targeting, information costs and incomplete family information together in Ukraine’s learning-recovery problem. A new PISA cycle or the idea of screening alone does not establish novelty.
Ukrainian tutoring experiments already used diagnostic assessment to group students.
Our question concerns the allocation of scarce places and the maximum justified price of additional information. Effects from that experiment are not attributed to our pilot.
Social differences in PISA performance have already been studied.
The contribution is a quantitative comparison of support-allocation decisions, including information costs and incomplete information.
Different information sources produce different errors in placement into remedial education.
A different population and a Ukrainian budget-allocation problem; effects from higher education are not transferred without validation.
The economic value of information has already been considered for school-based preventive programs.
We derive a spending limit when need is observed but a specific screening instrument’s effectiveness is unknown. Our review of this source covered its bibliographic information and abstract.
This is a targeted review of related work, not proof that no competing study exists. The review should be expanded before the main paper. Publication is not guaranteed.
Question, rationale, hypotheses, model, pilot and plan for the paper.
ZIP · code and aggregatesReproduce the analysis ↓Three Python scripts, package versions, tables, methods and the original file’s hash. Download microdata directly from OECD.
CSV · main pilotAll main estimates ↓Four rules, outcome measures, standard errors and intervals.
CSV · exploratory checksCapacity and missing information ↓Capacities of 10/20/30%, incomplete information and a mixed rule.