DLHS versus NFHS: which Indian health survey to choose

Someone wants to understand what is actually happening at the district level, whether the question is fertility, child nutrition, family-planning uptake, or institutional delivery, and reaches for either the DLHS or the NFHS without really understanding what each was built to do. The result is usually a quietly inflated claim, a buried caveat, or a trend line that joins two surveys that were never designed to be the same instrument.

The question of dlhs vs nfhs data for district level research sounds technical. Practically, it is a clinical-style matching decision: what is the patient asking, and which dataset is fit for that question? The rest of this piece is the kind of orientation I wish I had been given when I first tried to use these datasets to make sense of what I was seeing on the ground.

The Evolution of District-Level Estimates: From DLHS to NFHS-6

India's appetite for district-level evidence grew out of a specific policy moment. In the late 1990s, when DLHS-1 (1998-99) was being designed, the district was emerging as the operational unit for reproductive and child health programming. National surveys existed, but planners could not disaggregate below the state, and the gap between a state average and what was actually happening in a single district was too wide to ignore. The District Level Household Surveys were commissioned precisely to close that gap.

DLHS-1, DLHS-2 (2002-04), and DLHS-3 (2007-08) were all products of that era, implemented with IIPS Mumbai as the nodal agency under the Ministry of Health and Family Welfare. Their explicit remit was to estimate indicators — antenatal care, institutional delivery, full immunisation, family-planning method mix — at the district level, with enough precision to feed directly into Reproductive and Child Health programme reviews. DLHS-3 was the largest and most ambitious of the three. It covered approximately 700,000 sampled households across 612 districts. Its sampling design was deliberately tiered: better-performing districts contributed around 1,000 households, medium-performing districts 1,200, and low-performing districts 1,500, with the classification itself built from indicators like antenatal care coverage, institutional delivery, and immunisation status. That choice is worth pausing on. The designers were not just trying to measure; they were trying to oversample where the burden was likely heaviest.

DLHS-3 was the only round that linked household responses to a structured audit of the public facilities meant to serve those households. That structural decision is the reason it still matters.

DLHS-4 (2012-13) followed, but its story is one of partial coverage. According to government data catalogues, DLHS-4 published data for only 21 States and Union Territories, which means it is not a complete all-India replacement for DLHS-3, even if it is sometimes described as one in casual conversation. For researchers working on those 21 states, however, DLHS-4 brought something genuinely new: a clinical, anthropometric, and biochemical (CAB) component that measured height, weight, blood pressure, haemoglobin, blood glucose, and the iodine content of household salt at the district level. That made it a real workhorse for nutrition and lifestyle-disorder work in its covered geography, even as it left the rest of the country with a gap.

The National Family Health Surveys, meanwhile, had been running in parallel since 1992-93, but they were not originally designed for district-level disaggregation. NFHS-3 (2005-06) began to push in that direction. NFHS-4 (2015-16) and NFHS-5 (2019-21) became truly district-representative. NFHS-6 (2023-24), released by the Ministry on May 29, 2026 and again implemented with IIPS Mumbai as the nodal agency, is now the primary source for contemporary district-level demographic research — 715 districts, 679,238 households, 716,397 women interviewed, and 100,977 men.

Why NFHS-6 is the Primary Choice for Contemporary Demographic Research

If you are writing a paper today about current fertility, maternal health-seeking behaviour, child nutrition, or family-planning uptake at the district level, the honest answer is that NFHS-6 is where you start. The reasons are practical, not ideological, and they matter for what your conclusions will be allowed to claim.

First, coverage. NFHS-6 reaches 715 districts — more than DLHS-3's 612 — and it does so as a single, contemporary, all-India dataset. The provisional national fact sheets released in 2026 show a total fertility rate of 2.0, an institutional delivery estimate of 90.6%, and full immunisation coverage of 87.1% among children aged 12 to 23 months. Each of those numbers is a landmark in its own right. A national TFR at replacement means district planners are no longer wrestling with a single national fertility transition story; they are wrestling with a divergent one, where some districts are still well above replacement and others are already slipping below it. That divergence is only visible clearly in a dataset designed for district disaggregation in the recent period, and NFHS-6 is the only such dataset currently available across most of the country.

Second, content. NFHS-6 retained broadly similar questionnaire content to NFHS-5, which is what allows clean five-year trend comparisons at the district level. But it also added topics that did not exist in earlier rounds: Direct Bank Transfer coverage for health-related payments, Self-Help Group participation, digital literacy, and financial transaction patterns. For researchers interested in how digital and financial inclusion intersect with reproductive and child health decisions — a question that comes up constantly in my own clinical networks — those modules are essential. NFHS-6 also expanded its clinical, anthropometric, and biochemical testing to include HIV testing, which strengthens its use for population-level disease burden estimates.

Third, sample size and design. With roughly 679,000 households, NFHS-6 produces district estimates with reasonable precision for most core indicators. The IIPS Mumbai reports that the NFHS-6 sample was designed to provide national, State/UT, and district-level estimates for various indicators, but the documentation is explicit that indicators on husbands' background and women's work, and on attitudes and behaviour, are available only at the national and State/UT level. That is a critical caveat. Researchers who assume every NFHS-6 indicator is district-representable will overclaim, and reviewers who know the documentation will catch them.

The Unique Value of DLHS-3: Linking Household Data with Public Facility Infrastructure

There is one thing DLHS-3 did that no NFHS round has done since, and it is the reason DLHS-3 remains on my shelf rather than in storage. Alongside its household survey, DLHS-3 ran a population-linked facility survey. In every district it covered, the survey team visited every Community Health Centre and district hospital, plus the sub-centres and primary health centres expected to serve the selected primary sampling units. The facility questionnaires covered infrastructure, human resources, drugs and instruments, and service performance.

What this means in practice is that a researcher can still pair a DLHS-3 household observation — say, that 38% of women in a particular district reported delivering in a public facility — with a DLHS-3 facility observation that tells them whether that public facility had a functional labour room, an anaesthetist on roster, and oxytocin on the shelf. That linkage is rare in population-level data. NFHS collects household data with extraordinary depth, but it does not, in any recent round, link those household responses to a contemporaneous, structured audit of the public facility environment at the same geographic unit. SRS and other facility-side datasets exist, but they do not give you the household-to-facility pairing at the district level that DLHS-3 still offers.

For facility-readiness research, for service-environment analyses, and for any study trying to understand why institutional delivery rates are what they are — not just whether they are what they are — DLHS-3 remains the standard. The 2007-08 vintage is a limitation, of course, and I will come back to that. But the structure of the data is unique, and discarding it because it is old is a real loss.

Navigating the Limitations of DLHS-4 and Provisional NFHS-6 Results

Every dataset has edges, and the edges of DLHS-4 and NFHS-6 are the ones most likely to trip up a careful researcher.

DLHS-4's published coverage was limited to 21 States and Union Territories. If your research question concerns a state outside that set, DLHS-4 is simply not an option for a district-level comparison in that round, no matter how attractive its CAB component looks on paper. The government data catalogue for DLHS-4 also specifies a disclosure rule: percentages are not shown when they are based on fewer than 20 cases. That is reasonable from a privacy standpoint, but it means small-district or rare-indicator estimates may be suppressed in published tables, and you may need to apply for the unit-level data to reconstruct them yourself.

The released NFHS-6 fact-sheet results are explicitly provisional. Any analysis built on those values should carry that caveat in the abstract, not bury it in a footnote.

NFHS-6 has a different limitation, and it is one every analyst needs to write into their methods section explicitly. The fact-sheet results released on May 29, 2026 are explicitly provisional. The availability and timing of final district fact sheets and public microdata are not established in the released material. That uncertainty is not a minor stylistic detail. It changes how much weight a reader should give the numbers, and any district-level analysis built on provisional values carries that uncertainty into every downstream planning conversation.

There is also the question of what NFHS-6 actually estimates at the district level versus what it estimates only at the State/UT level. As I noted earlier, husbands' background, women's work patterns, and attitudes and behaviour domains are not district-representable in NFHS-6. If your study hinges on one of those domains at the district scale, NFHS-6 will not give you what you need, and no amount of disaggregation will rescue it.

Methodological Hurdles in Constructing Longitudinal District-Level Trends

This is the section where I slow down, because this is where the comparison most often goes wrong. A 2016 review of India's large health surveys, published in the Bulletin of the World Health Organization, found only modest comparability across surveys for child mortality, maternal mortality, and immunisation trend analysis. That sentence sounds technical, but its consequences are immediate. You cannot simply join DLHS-3 (2007-08), DLHS-4 (2012-13), and NFHS-6 (2023-24) district indicators end-to-end and call the result a trend.

Why not? Four reasons keep coming up in practice, and I run through them with every research assistant before they touch a district-level time series:

1. Indicator definitions change. The denominator for “full immunisation” in DLHS rounds is not identical to the denominator in NFHS rounds, and the reference periods for antenatal care recall questions have been revised over time. Treat the question wording as a moving target.

2. Sampling frames shift. DLHS-3 used the 2001 Census as its sampling frame. By the time NFHS-5 and NFHS-6 were designed, the frame had moved, district boundaries had been reorganised in several states, and new districts had been carved out of old ones. A district called “X” in 2007-08 may not be the same geographic unit as “X” in 2023-24, and there is no single clean, authoritative crosswalk that resolves every DLHS district identifier against every NFHS-6 district identifier.

3. Suppression rules differ. DLHS-4's “fewer than 20 cases” suppression rule does not apply in the same way to earlier DLHS rounds or to NFHS, so apparent gaps in a published series may be a reporting artefact rather than a real change.

4. The endpoint is provisional. NFHS-6’s released fact-sheet values are provisional. A long trend that uses them as its endpoint needs to say so clearly and avoid treating that endpoint as settled final evidence.

A practical approach I recommend: before you build a DLHS-versus-NFHS trend line, sit with the questionnaire documents for each round and check, indicator by indicator, that the wording, reference period, denominator, and eligibility criteria are comparable. If they are not, document the difference, restrict your trend to the subset of indicators that genuinely are comparable, and be transparent about what you have left out. Reviewers will respect the constraint; they will not respect a trend line built on sand.

Choosing the Right Dataset for the Question You Are Actually Asking

The choice between DLHS and NFHS is not a referendum on which survey is “better.” It is a clinical-style matching question, and I treat it the way I treat any referral. What is the patient asking? What is the right dataset for that question, given what we know about coverage, timing, content, and methodological caveats?

For contemporary district-level demographic research on most indicators across most of India, NFHS-6 is the right starting point, with its provisional fact-sheet status openly flagged in the abstract and methods section. For historical RCH-era analysis and especially for facility-readiness research, DLHS-3 remains uniquely valuable and irreplaceable. For DLHS-4, the answer is geography-specific: if your state is among the 21 covered, its CAB component is genuinely useful for nutrition and lifestyle-disorder work; if not, you are looking at a different dataset altogether.

The mistake is to treat the choice as binary and final. In practice, the strongest district-level studies I see use both traditions carefully — DLHS for the period and the questions only it can answer, NFHS-6 for the contemporary picture, and explicit methodological honesty about what can and cannot be joined across them. If you are sitting at a desk right now, trying to decide, slow down, match the dataset to the question, and write your caveats into the abstract. That is what I tell every junior researcher who walks into my clinic with a stack of printouts and a deadline, and it is what I would tell anyone reading this. The data are extraordinary. The care we take with them is what makes them useful.