Educational Stratification in PISA – A Browser-Based Statistical Tool for Exploratory Analysis of PISA Microdata, 2000–2022

1. Introduction

OECD PISA is widely used in secondary analyses because it combines cross-nationally comparable achievement measures with extensive student background information, repeated in a regular cycle since 2000 (OECD, 2009, 2024). At the same time, routine empirical tasks with PISA microdata remain implementation-heavy for many users: the data arise from complex sampling; achievement is released as plausible values rather than single test scores; and results can be sensitive to choices such as which weights are used, which cycles and countries are included, and which regression specification is treated as the descriptive model of interest (Mislevy et al., 1992; Wu, 2005; OECD, 2009).

the tool was developed as a technical and educational tool to make these common analytic operations easier to run, easier to iterate, and easier to export in a structured form, without requiring local installation or bespoke scripts for each exploratory question. The central research question the tool supports is: how do parental characteristics—specifically education, occupational status, and household resources—predict children's educational achievement at age 15? This is the core question of intergenerational transmission of educational achievement, and the tool operationalizes it through the PISA ESCS index and related family background measures.

1.1 Contribution and Audience

The application is intended for researchers, instructors, and students who need rigorous comparative analysis without bespoke statistical code. Its contribution is both methodological and infrastructural:

• Accessibility: Non-specialist users can reproduce core metrics and regression models directly in the browser, examining how parental background relates to student outcomes across countries and time.
• Comparative sensitivity: Users can systematically compare estimators and country selections to assess how methodological choices shape conclusions about intergenerational transmission patterns.
• Open science: Every analysis is exportable with metadata that preserves the analytic context, supporting transparent reporting and replication.

2. Data Coverage, Provenance, and Processing

the tool currently covers eight PISA cycles (2000, 2003, 2006, 2009, 2012, 2015, 2018, 2022), encompassing over 100 participating countries and economies and approximately 3.5 million student observations. Data access and harmonization are supported through the learningtower R package, which provides cleaned and harmonized PISA extracts with standardized variable naming across cycles; the CRAN release for learningtower (version 1.1.0) documents coverage for 2000–2022 and provides the recommended package citation (Wang et al., 2024).

• Detailed data sources and variable documentation
• Full methodology and technical documentation

2.1 Static Data Architecture

A central implementation choice is a static, browser-friendly data layout. Instead of loading a full cross-cycle microdata file into memory, the tool stores data as country–year JSON "chunks" and loads only the requested subsets when a user selects countries and years. This reduces memory load (typically 30–40 MB for a working selection versus 1.25 GB for the full dataset) and supports interactive use in standard browsers. The chunking strategy and sample-construction rules (including current listwise deletion for required fields) are documented in the project's data and methodology pages.

2.2 Key Variables: ESCS and the Measurement of Family Background

The primary outcome variables are PISA achievement scores in mathematics, reading, and science—measuring what 15-year-olds know and can do in these domains. The primary stratification variable is ESCS (Economic, Social, and Cultural Status), which operationalizes family background through three components:

• Parental education: Highest educational attainment of either parent, coded using ISCED classifications.
• Parental occupational status: Highest occupational status of either parent, measured using the ISEI (International Socio-Economic Index of Occupational Status) or HISEI (Highest parental ISEI) family of indices.
• Home possessions: An index of household wealth and educational resources, including books at home, educational possessions, and material resources.

ESCS is standardized to an OECD-referenced scale (mean 0, SD 1 for OECD countries), making it interpretable as a composite measure of family socioeconomic advantage. Recent OECD technical documentation summarizes the construction and intended use for cross-country and trend analyses (Wuyts, 2024). When users examine ESCS gradients or quartile gaps in the tool, they are directly measuring the strength of intergenerational transmission—how strongly these parental characteristics predict their children's academic performance at age 15.

3. Outcomes, Weights, and the Interpretation of Achievement

3.1 Survey Weighting

Because PISA is a complex sample, the tool's descriptive and regression outputs are computed using sampling weights. The tool defaults to the final student weight (W_FSTUWT) and optionally supports senate weights (W_FSENWT) for alternative estimands; replicate weights are described in the documentation but are not currently used in the interactive variance estimation (OECD, 2009).

3.2 Plausible Values

Achievement in PISA is released as plausible values, which are designed to support valid population inference under matrix sampling and latent proficiency estimation. Standard practice treats plausible values analogously to multiple imputations: analysts repeat computations across plausible values and combine estimates and variances accordingly (Mislevy et al., 1992; Rubin, 1987; Wu, 2005). For interactivity, the tool currently uses the first plausible value per domain in the web interface and flags this explicitly; users aiming for publication-grade inference are expected to replicate selected specifications using all plausible values and appropriate variance procedures.

4. Statistical Functionality

the tool implements a set of recurring "building blocks" that users typically assemble when producing descriptive briefs, teaching demonstrations, or preliminary robustness checks focused on intergenerational transmission of educational achievement.

4.1 Descriptive Statistics

The tool computes survey-weighted descriptive statistics, including weighted means, variances, and weighted quantiles. These quantities are used both for direct reporting and as inputs to distributional summaries.

4.2 Distributional Summaries

the tool summarizes achievement distributions using indices such as the Gini coefficient (with Lorenz curve display), the coefficient of variation, and percentile ratios. These measures characterize the spread and shape of achievement distributions within and across countries.

4.3 ESCS Gradients: Measuring Intergenerational Transmission

The ESCS gradient is the central measure for quantifying intergenerational transmission. It is computed as the slope from a survey-weighted regression of achievement on ESCS (optionally including simple controls such as gender and parental education). This coefficient represents how many score points of achievement are associated with a one-unit increase in family socioeconomic status—directly measuring how strongly parental characteristics predict children's outcomes.

A larger ESCS gradient indicates stronger intergenerational transmission—children's achievement is more strongly predicted by their parents' education, occupation, and wealth. Cross-country comparisons of ESCS gradients reveal which education systems show more or less mobility in educational outcomes.

4.4 Quartile Gaps: Family Background Differences in Achievement

the tool computes quartile-based ESCS gaps by constructing ESCS quartiles using weighted ranks and reporting Q4–Q1 differences (the gap between children from the most and least advantaged family backgrounds) along with standardized effect sizes (Cohen's d). This provides an intuitive measure of how much family background matters: the score-point difference between students from high-ESCS versus low-ESCS families.

4.5 Regression Models

the tool provides a small family of regression specifications intended for quick comparison rather than definitive modeling:

• Pooled OLS: Baseline weighted regression pooling all observations, showing the overall relationship between family background and achievement.
• Fixed Effects: Country fixed effects (with optional year fixed effects) controlling for time-invariant country heterogeneity, isolating within-country variation in how family background relates to outcomes.
• Random Effects: Country-level random intercepts with quasi-demeaning transformation, modeling country differences as draws from a common distribution.

The interface reports coefficient tables and conventional fit metrics (including AIC/BIC) and offers standard diagnostic plots (residuals, Q-Q plots). Model diagnostics include Hausman tests when both FE and RE models are available.

4.6 Variance Decomposition

the tool includes a variance decomposition module that partitions variance into within- and between-country components and reports an ICC-style summary of clustering at the country level. This reveals how much variation in the relationship between family background and achievement occurs within versus between countries.

5. Implementation Architecture, Exports, and Reproducibility

the tool is implemented as a modular ES6 JavaScript application with an R-based preprocessing pipeline that generates country–year data chunks and a metadata catalog used to drive the interface and loading logic. All computations in the interactive application are performed client-side after data are loaded, which keeps deployment simple (static hosting) and makes the computational steps inspectable in the project materials.

Component	Purpose	Reference
Data generation scripts	Extract and chunk country-year microdata	Pipeline scripts
Methodology appendix	Formal statistical definitions and formulas	Methodology document
Data sources	Variable provenance and coverage	Data sources document

5.1 Exportability

A practical design priority is exportability. the tool is intended to support workflows where users explore specifications interactively and then export outputs for inclusion in briefs, appendices, slides, or follow-up coding. The tool exports:

• Tabular outputs (summary statistics, regression tables, comparative analyses) as CSV
• Graphics as high-resolution PNG or SVG
• Self-contained HTML reports with embedded visualizations
• Analysis metadata recording all selections (countries, years, domain, weight choice, estimator family, and optional controls)

This is not a substitute for a full scripted pipeline, but it aligns with standard expectations for computational reproducibility in exploratory work, where the key requirement is that the analytic choices are recorded and can be re-executed in a more formal environment (Peng, 2011).

6. Typical Use Cases

the tool is most useful when the goal is to move quickly from a question about intergenerational transmission to a defensible descriptive answer, while keeping specification choices visible.

6.1 Cross-Country Profiling

Select a year and a set of countries to compare weighted distributional summaries and ESCS gradients, then export a compact table and figures for briefing or teaching. This reveals which countries show stronger or weaker intergenerational transmission of educational achievement.

6.2 Sensitivity Screening

Compare pooled and country-panel specifications, toggle weight conventions (where appropriate to the estimand), and assess whether headline patterns about family background effects are stable to common alternatives before deciding which models deserve full PV-complete, replicate-weight inference in a code-based workflow.

6.3 Trend Exploration

Inspect how ESCS gaps and gradients evolve across cycles for selected countries, treating the outputs as descriptive signals about whether intergenerational transmission is strengthening or weakening over time.

6.4 Classroom Instruction

the tool can be used to demonstrate why weights matter, what plausible values are, and how different regression specifications change descriptive associations between family background and achievement, without requiring students to install specialized software or manage large microdata files (OECD, 2009; Wu, 2005).

7. Limitations and Appropriate Interpretation

the tool's outputs should be interpreted as descriptive associations produced by a transparent, interactive implementation, not as definitive inferential results.

• Plausible values: The web interface currently uses a single plausible value per domain; estimates can differ from PV-combined results, especially for subgroup analyses and regression relationships (Mislevy et al., 1992; Rubin, 1987; Wu, 2005).
• Variance estimation: The interactive layer does not implement replicate-weight variance procedures, which are a central element of design-consistent standard errors in PISA analyses (OECD, 2009).
• Missing data: the tool currently relies on listwise deletion for required fields, which changes the effective sample and can bias results when missingness is systematic.
• Cross-sectional design: PISA is cross-sectional within cycles, and the tool is designed for exploration and descriptive reporting rather than causal identification strategies. ESCS gradients measure association, not causation.

8. Conclusion

the tool is a browser-based statistical tool that packages common first-pass operations used in large-scale assessment work—weighted descriptives, distributional summaries, ESCS gradients and quartile gaps measuring intergenerational transmission, specification comparisons, and variance decomposition—into a static, export-oriented interface for PISA cycles 2000–2022. Its value is primarily practical: it shortens the path from question to documented descriptive output about how parental education, occupation, and wealth relate to children's educational achievement at age 15 worldwide. It supports teaching and methods demonstrations, and helps users define and screen specifications before re-estimating selected models with full plausible-value combination and replicate-weight inference in a dedicated statistical environment.

References

Mislevy, R. J., Beaton, A. E., Kaplan, B., & Sheehan, K. M. (1992). Estimating population characteristics from sparse matrix samples of item responses. Journal of Educational Measurement, 29(2), 133–161. https://doi.org/10.1111/j.1745-3984.1992.tb00371.x

OECD. (2009). PISA data analysis manual: SPSS (2nd ed.). OECD Publishing. https://doi.org/10.1787/9789264056275-en

OECD. (2024). PISA 2022 Technical Report. OECD Publishing. https://doi.org/10.1787/01820d6d-en

Peng, R. D. (2011). Reproducible research in computational science. Science, 334(6060), 1226–1227. https://doi.org/10.1126/science.1213847

Rubin, D. B. (1987). Multiple imputation for nonresponse in surveys. John Wiley & Sons. https://doi.org/10.1002/9780470316696

Wang, K., Yacobellis, P., Siregar, E., Romanes, S., Fitter, K., Dalla Riva, G. V., Cook, D., Tierney, N., Dingorkar, P., Sai Subramanian, S., & Chen, G. (2024). learningtower: OECD PISA datasets from 2000–2022 in an easy-to-use format (R package, Version 1.1.0). https://doi.org/10.32614/CRAN.package.learningtower

Wu, M. (2005). The role of plausible values in large-scale surveys. Studies in Educational Evaluation, 31(2–3), 114–128. https://doi.org/10.1016/j.stueduc.2005.05.005

Wuyts, C. (2024). The measurement of socio-economic status in PISA. OECD.

Back to Application