Linking School Infrastructure to Learning Outcomes: a UDISE+ and NAS District-Level Study

Linking School Infrastructure to Learning Outcomes is an M.Sc (IT) project that builds a single, clean, district-level dataset by joining two major government education data systems — UDISE+ (the annual school-information collection covering enrolment, teachers and infrastructure) and the National Achievement Survey (NAS) (NCERT's nationwide learning assessment) — and then measures, honestly, how far infrastructure indicators are associated with learning outcomes.

Every year schools report to UDISE+ whether they have functional girls' toilets, drinking water, electricity, internet and libraries, along with pupil–teacher ratios. Every three years the NAS publishes district report cards with subject-wise mean performance for Classes 3, 5 and 8. Both are public. Neither is easy to use together: UDISE+ publishes indicator tables per district while the NAS publishes scored report cards, and the two systems name and code districts differently. The integration itself — harmonising districts, standardising indicators, auditing missing values — is the core engineering contribution.

On top of the integrated table, the study runs a defensible statistical analysis: Spearman and partial correlations between a constructed infrastructure-composite index and NAS mean scores, followed by ordinary least-squares regression with covariates such as pupil–teacher ratio, and full regression diagnostics. A Streamlit dashboard lets a reader explore any district's profile and see every claimed association as a plot, not just a table.

The topic matters because infrastructure spending is enormous and its learning payoff is contested; the method matters because joining, cleaning and diagnosing a real multi-source dataset is exactly what an M.Sc (IT) degree claims to certify.

Policy conversations about school infrastructure in India run on intuition: toilets, electricity and internet are obviously good, so spending on them must help learning. But the claim is rarely tested district by district against the government's own measurement systems, and the analyses that do exist are usually single-source summaries that never join UDISE+ infrastructure indicators with NAS learning scores. The obstacles are technical rather than conceptual — district naming mismatches between systems, indicator definitions that shift between UDISE+ rounds, and missing values that are silently ignored.

There is a need for a reproducible integration and analysis: one harmonised district-level table built from the two official sources with every mapping documented, an index whose construction is transparent and sensitivity-tested, correlation and regression analyses whose assumptions are checked rather than assumed, and a presentation layer where a reader can inspect any district and every fitted relationship directly. Without this, infrastructure budgeting continues on faith, and the two most relevant public datasets remain strangers to each other.

Abstract

Does a functional girls' toilet, an electricity connection or an internet link in every school translate into better learning? India spends substantially on school infrastructure, yet public debate rarely tests the question against the two datasets designed to answer it: UDISE+ infrastructure indicators and NAS learning-outcome scores, both published district by district. This project integrates the two systems for Tamil Nadu's districts, constructs a documented infrastructure-composite index, and measures its association with NAS mean scores using correlation and regression with full diagnostics, delivered as reusable code and an exploratory dashboard.

Introduction

UDISE+ records, for every school, the facilities that policy targets promise; the NAS measures, for sampled children in every district, what learning actually happened. Each dataset is analysed alone all the time — dashboards quote toilet percentages, reports quote NAS means — but the two are almost never joined, because joining them is unglamorous work: different district names and codes, different school levels, changing indicator definitions across UDISE+ rounds. That integration gap is precisely where a data-science project earns its marks.

Existing system and gap

Published studies of Indian education data tend to either (a) analyse UDISE+ indicators as national summaries with no outcome link, (b) analyse NAS scores alone against census-style covariates, or (c) claim strong infrastructure–outcome links from pooled cross-country data whose relevance to Indian districts is weak. District-level, code-harmonised, diagnostics-complete analyses of the two Indian systems together are rare, and almost none ship the code. The gap this project fills: a reproducible, district-level integration of UDISE+ and NAS for one state, with the harmonisation documented, the index construction transparent, and the statistics honest about what correlation can and cannot show.

Proposed system

• Sources: UDISE+ district report tables (dashboard.udiseplus.gov.in) and NAS 2021 district report cards (nas.gov.in), for Tamil Nadu (38 districts), with the all-India table built as a by-product.
• Integration: a district crosswalk mapping UDISE+ and NAS spellings/codes to one master list; indicator extraction for six facilities plus PTR; score extraction for Class 3, 5 and 8 subject means.
• Index construction: z-score standardisation of indicators into a weighted infrastructure-composite index, with weights and sensitivity checks published.
• Analysis: Spearman and partial correlations; OLS regression of NAS mean scores on the index with PTR and enrolment covariates; residual, leverage and influence diagnostics; a significance framework fixed before analysis.
• Interface: a Streamlit dashboard — district profile pages, scatter with fitted lines, league tables and full regression output.

Feasibility study

• Technical: both sources are free public downloads; volumes are tiny (tens of districts, dozens of indicators); every library is open source. The only genuinely hard step — district harmonisation — is bounded and testable.
• Economic: zero cost; no paid data or hosting.
• Operational: all data is aggregate and public, so no human-subjects clearance is required; the integrated SQLite dataset and dashboard run offline on a laptop, which also removes any dependency on the portals during the viva.

In scope

• One state (Tamil Nadu, 38 districts) as the headline analysis; an all-India district table built as a by-product for robustness checks.
• Public aggregate data only: published UDISE+ report tables and NAS 2021 district report cards.
• Infrastructure indicators, PTR, and NAS mean scores at Class 3/5/8 level; composite index; correlation and regression with diagnostics; dashboard.

Out of scope

• Individual-school or student-level analysis — the published district aggregates cannot support it, and claiming more would be dishonest.
• Causal claims: the study measures association and says so; no instrumental variables or randomised evidence are attempted.
• Private-school unaided categories and detailed expenditure tracking.
• Time-series analysis across multiple NAS rounds (the next NAS round would make this possible and is listed as future scope).

Research design: quantitative, secondary-data, cross-sectional study at district level, with an engineering deliverable (integration pipeline + dashboard). All data are published aggregates; no human participants are involved, so no ethics clearance is required. Sources are used with attribution.

Analysis plan (fixed before results): α = 0.05 throughout; Spearman as the primary correlation (indicators are proportions bounded at 0–1); OLS with Class 8 mathematics mean as the headline outcome, infrastructure index as the primary predictor, PTR and log-enrolment as covariates; diagnostics include residual-vs-fitted plots, Q–Q plots, VIF for multicollinearity and Cook's distance for influential districts. Every figure in the report regenerates from scripts, never by hand.

The system is a study pipeline with an integration core: two sources in, one harmonised dataset out, statistics and dashboard downstream.

flowchart TD
  U["UDISE+ district report tables"] --> A["Indicator extraction: facilities, PTR"]
  N["NAS 2021 district report cards"] --> B["Score extraction: Class 3 / 5 / 8 means"]
  A --> C["District crosswalk: harmonised names and codes"]
  B --> C
  C --> D[("SQLite: district master, indicators, scores")]
  D --> E["Missing-value audit and exclusion log"]
  E --> F["Composite infrastructure index (z-scores, sensitivity-checked)"]
  F --> G["Correlation: Spearman + partial"]
  F --> H["OLS regression + diagnostics"]
  G --> I["Results tables and figures"]
  H --> I
  I --> J["Streamlit dashboard: profiles, scatters, output"]

erDiagram
  DISTRICT ||--o{ UDISE_INDICATOR : "reports"
  DISTRICT ||--o{ NAS_SCORE : "achieves"
  DISTRICT ||--o| INFRA_INDEX : "receives one"

  DISTRICT {
    string district_id PK
    string udise_name
    string nas_name
    string state_code
  }
  UDISE_INDICATOR {
    string id PK
    string district_id FK
    string indicator_name
    real value
    string source_round
  }
  NAS_SCORE {
    string id PK
    string district_id FK
    string grade
    string subject
    real mean_score
    int students_sampled
  }
  INFRA_INDEX {
    string district_id PK
    real composite_score
    string weighting_scheme
  }

Two design decisions carry the analysis. First, the crosswalk is data, not code: district mappings live in a reviewable table (with the source spelling from each system), so a disagreement can be resolved by editing an entry rather than a script — and the mapping itself becomes a citable artefact of the report. Second, exclusions are logged, not silent: any district missing an indicator or a score stays in the master table with NULLs, and each analysis states its own effective sample, so a reader can see exactly which regression ran on how many districts.

• Add the next NAS round for a two-period panel: does an improvement in infrastructure between rounds predict an improvement in scores?
• Extend the integration to all-India headline analysis and test whether the Tamil Nadu associations hold nationally.
• Join UDISE+ with NITI Aayog's SDG India index indicators for a richer covariate set.
• Replace the composite index with a principal-component construction and compare loadings.
• Package the crosswalk as a small open library so other students can reuse the hardest part of this work.