Data Journalism · Public Policy · Inference & Evidence

Public data,
unlocked.

The Inference Project is an independent data journalism platform. We turn large public datasets — starting with India's National Family Health Survey — into interactive dashboards that anyone can read, question, and explore, so the inference is yours to draw, not ours to hand you. Three rounds are live — NFHS-4 (2015–16), NFHS-5 (2019–21) and NFHS-6 (2023–24) — with a Compare mode that maps what changed between any two of them. More datasets are on the way.

NFHS-4 · 2015–16 NFHS-5 · 2019–21 NFHS-6 · 2023–24 Compare · any two rounds

Explore the dashboard How the numbers are made

scroll — the inference is live below

The Inference Project

Every relationship here is inferred, not asserted.

Each dot is one of India's 36 states and union territories, plotted from survey-weighted NFHS-5 (2019–21) estimates. The dotted line is an ordinary-least-squares fit computed live — the inference the project is named for. Swap the pair and watch the relationship re-draw.

Each dot is a state or union territory. The dotted line is an ordinary-least-squares fit drawn from the data.

—

Pearson r

NFHS-5 (2019–21) · 36 states/UTs

What this is

An independent data journalism platform. Every dashboard is built from raw public microdata, the methods are open, and every figure traces back to the source files. Where an estimate is fragile, it is marked, not hidden.

survey rounds, eight years apart

indicators tracked

700k+

women interviewed per round

states & UTs

707

districts

levels of breakdown

On the platform

Three survey rounds live. More on the way.

Each dataset gets the same treatment: raw public microdata, weighted into clear, explorable estimates, delivered as an interactive dashboard. Three rounds of India's National Family Health Survey are live — eight years of change, with any two of them comparable side by side. DHS rounds from other countries and international datasets are on the roadmap.

Live now

Rounds 4 · 5 · 6

NFHS Dashboard

Three rounds of India's National Family Health Survey in one place, on one estimator — read any round on its own, or set any two side by side in Compare.

Open the dashboard →

NFHS-4 2015–16 · microdata

The baseline. Computed from the unit-level recodes, down to all 640 districts, five wealth bands and four education levels.

NFHS-5 2019–21 · microdata

The fullest round: 68 indicators across 707 districts, on the same weighted estimator and the same breakdowns as NFHS-4.

NFHS-6 2023–24 · fact sheets · provisional

The newest round. Its microdata is unreleased, so its 49 indicators are read from the official fact sheets — state and urban–rural level, no districts.

Coming soon

Census of India

Decennial population census

Population, literacy, work, migration and amenities — read down to the smallest administrative units.

Compare · NFHS-4 → NFHS-5

Six years of change, read directionally.

Compare mode sets 2015–16 against 2019–21 and asks a harder question than "what is the number?" — did it get better or worse? The dashboard knows which way is healthy for every indicator, so a fall in anaemia and a rise in immunization both read as progress. These are the real national movers.

Improved — moved the healthy way Slipped — moved the wrong way National figures · percentage-point change

Mortality & fertility — measured in their own units

Rates and ratios cannot be added to percentages, so Compare keeps them apart. Every child-mortality rate fell between the rounds — fewer deaths at every age.

Direction-aware: for "less is better" indicators — anaemia, stunting, mortality — a decline counts as a gain. 61 indicators are compared in total; the full scorecard, plus state dumbbells, a change map and district spreads, lives in Compare mode →

NFHS-4 · NFHS-5 · NFHS-6 — all live

Inside the NFHS datasets

What these datasets cover, and how deep they go. All three rounds run in the same dashboard on the same estimator, so you can read any one at full resolution or set any two side by side in Compare. The figures below are NFHS-5 — the fullest round — with each strip showing its move from NFHS-4; full indicator definitions live in the dashboard.

Six domains, dozens of indicators

Violet dots are the real NFHS-5 spread across the states. The tick shows the national average moving from NFHS-4 (grey, 2015–16) to NFHS-5 (violet, 2019–21) — teal when the move is healthy, coral when it is not.

One number, five resolutions

National stunting is 35.5% — an average that hides almost everything. The same figure, resolved from one country down to wealth quintiles:

Survey-weighted NFHS-5 estimates, children under five. Wealth bands run poorest → richest.

Inside Compare mode

Five ways to see what moved.

Pick any indicator and Compare reads the change at every resolution — nationally, by state, by district, and against other indicators. Teal is improvement, coral is a setback, throughout.

National scorecard

Every indicator ranked by direction-aware change, gains against setbacks on one axis.

Change map

All 707 districts shaded by whether they improved or slipped — teal to coral, on the map.

District beeswarm

The spread of districts in each round, stacked — search and hold any district to track it.

Change scatter

Any indicator on X and Y — round against round, or the change itself — to find who moved together.

Change in context

The notes that make the comparison honest — J&K and Ladakh recomputed on today's boundaries, districts reconciled, splits flagged.

The live dashboard

Pick an indicator. Watch it break apart.

The full interactive tool — choropleth maps, ranked state and district bars, equity dumbbells and correlation scatters — embedded below. Switch between NFHS-4, NFHS-5 and NFHS-6 in the header, or open Compare and pick any two rounds for a national scorecard, a teal-and-coral change map, a district beeswarm and change scatters that show exactly what moved between them.

nfhs-dashboard.html

A preview is embedded here for context. For the full scrolling experience — all sections, faster — open it on its own.

Open the full dashboard ↗

How the numbers are made

Traceable from raw recode to printed estimate

SourceNFHS-4 (2015–16) · NFHS-5 (2019–21) · NFHS-6 (2023–24) All conducted by IIPS, Mumbai for the Ministry of Health & Family Welfare; fieldwork support from ICF / DHS Program.
Raw inputsSix DHS recode files per round Women (IR), children (KR), men (MR), households (HR), persons / biomarkers (PR), births (BR) — NFHS-4 and NFHS-5 are built from these, with the identical pipeline.
WeightingSurvey design weights applied Each record weighted by its sampling weight so estimates represent the population, not the sample.
BreakdownsState · district · urban–rural · wealth · education Computed independently at each level; small-sample cells are flagged or suppressed.
ReliabilityDHS convention, shown not hidden Cells with n 25–49 are faded as fragile; n < 25 is withheld entirely.
NFHS-6Read from the fact sheets, not computed Its recodes are unreleased, so NFHS-6 is transcribed from the official fact sheets and remains provisional — state and urban–rural level only.
ValidationChecked against the official Fact Sheet National estimates for each round are reconciled to its published NFHS India Fact Sheet within tolerance.

Inside the dashboard

Every number, from every angle

One indicator can be read a dozen ways — nationally, by state and district, split by wealth and schooling, or set against another survey round. The dashboard makes each of those a click away.

Maps & rankings

Choropleth maps and ranked state and district bars for 60+ indicators — the whole country, down to one district.

Equity, education & urban–rural

Every indicator split by wealth quintile, mother's schooling, and urban vs rural — so a state average never hides who is left behind.

Beeswarms

Districts, wealth levels or residence groups as swarms of dots, national average marked — search and hold any one to follow it.

Distributions

The whole Z-score curve behind stunting, wasting and underweight, against the WHO standard — not just the headline cut-off.

By-state small multiples

Any chart re-drawn as a grid of small per-state panels, so all 36 states and UTs read side by side at a glance.

Compare across rounds

NFHS-4 against NFHS-5 as a scorecard, change map, state dumbbells and scatters — every indicator read directionally, better or worse.

Filter, sort & correlate

Pick the states you want, sort by value or by the gap, and plot any two indicators against each other to see what moves together.

Documented & flagged

A plain-language definition for every indicator, full methodology, and a reliability flag wherever a sample is too thin to trust.

About

A public platform for understanding through data

The Inference Project is a public data journalism platform built to take large public datasets that usually sit locked inside raw survey files and turn them into dashboards anyone can read, question and reuse.

It uses public data — India's large national surveys such as the NFHS, DHS surveys from other countries, and datasets from multilateral agencies — to build documented, source-traceable dashboards on health, demography and socio-economic issues. India is the primary focus, but the project is not restricted to it.

The project is independent and non-partisan, and is not affiliated with IIPS, the DHS Program, or any government body. The source data is theirs; the estimates, the code and any mistakes are the project's own.

Mission

Alongside widening access to fundamental information, the project's mission is to deepen data literacy — building shared knowledge around data sources, metrics and methodologies, and giving readers the tools to interrogate the numbers themselves rather than take a headline on faith.

Everything rests on being clear and transparent about sources and methods. Every dashboard documents exactly how its estimates are computed.

Found an error, or want to use the data? That is encouraged. Write to hello@theinferenceproject.com