Fertility rate trends in India: tracking regional decline drivers through DLHS data

The headline is unambiguous: fertility is below the government’s replacement-level benchmark of 2.1.

The national mean is not the analytical endpoint. It is the point at which aggregation starts to conceal the problem. NFHS-5 still recorded a total fertility rate of 3.0 in Bihar, 2.9 in Meghalaya, and 2.4 in Uttar Pradesh. A country can be below replacement level while substantial population cohorts remain in high-fertility settings. India is precisely such a case.

The District Level Household Survey is useful here because it moves the unit of analysis below the state. DLHS does not supply a clean national district panel across every round, and it cannot establish causal effects by itself. It does, however, expose the service environment in which fertility decisions occur: contraceptive prevalence, method mix, unmet need, contact with maternal health services, and the distribution of reproductive-age women across districts.

India’s fertility decline is a national average assembled from unequal district transitions.

The granular lens of DLHS-3 and district-level realities

DLHS was designed around the practical geography of reproductive and child health administration. Its purpose was not merely to estimate a national rate. It was to identify districts where programme coverage, family-planning demand, maternal care, and child-health indicators diverged from state averages.

DLHS rounds were conducted in 1998–99, 2002–04, 2007–08, and 2012–13. The third round remains the strongest reference point for broad district-level analysis. DLHS-3, conducted in 2007–08, covered 720,320 households in 601 districts across 34 states and union territories. Nagaland was not included.

That sample scale matters. A national fertility estimate can be statistically stable while remaining operationally vague. A district estimate is less elegant but more actionable. It indicates where a contraceptive supply chain, counselling system, postpartum programme, or outreach model may be underperforming relative to adjacent areas.

DLHS-3 captured more than birth histories. Its family-planning indicators included current method use among married women aged 15–49 and total unmet need for family planning. These measures cannot be treated as direct substitutes for total fertility rate. They measure different constructs.

The DLHS fertility indicators analysis should therefore begin with a separation of levels. TFR is a population outcome. Contraceptive use is an intermediate behavioural measure. Unmet need is a coverage and access problem. None is interchangeable with another.

At the all-India level, DLHS-3 reported that 54.0% of currently married women aged 15–49 used any contraceptive method. Modern-method use stood at 47.1%. Total unmet need for family planning was 21.3%.

This produces an uncomfortable arithmetic. More than half of currently married women were using some method. Yet more than one in five still had an identified unmet need. The gap is not a marginal residual. It represents a large segment of reproductive demand that had not been converted into effective, voluntary contraceptive use.

The result should not be read as evidence that every non-user lacks access. Survey classifications include fertility preferences, timing of births, postpartum status, fecundity, and other conditions. But at district scale, a high unmet-need estimate remains a programme signal. It identifies an area where stated reproductive intentions and current method use are not aligned.

What fertility rate calculation can and cannot show

The fertility rate calculation methodology is frequently reduced to a slogan: births per woman. The actual measure is more specific.

Total fertility rate is constructed by summing age-specific fertility rates, generally across conventional reproductive-age groups, then converting that cumulative schedule into the number of births a woman would have if she experienced those rates throughout her reproductive lifetime. It is a period measure. It does not describe the completed fertility of a real birth cohort.

This distinction becomes material during rapid demographic change. If marriage is delayed, births are postponed, or spacing increases, a period TFR may fall even before the lifetime number of births for a cohort is fully observed. That is not an error. It is a property of period fertility measurement.

The second problem is source comparability. India’s major fertility data systems serve different statistical functions:

• DLHS supplies district-oriented reproductive and child-health indicators and selected fertility context measures.
• NFHS provides nationally and state-representative estimates, with extensive demographic, health, and household variables.
• SRS is a continuous registration-based sample system used for annual demographic estimates, including fertility and mortality.
• Census data provide the population denominators, age structures, and geographic frameworks required to interpret longer-run population change, but are not a direct replacement for survey fertility schedules.

The figures should be placed in sequence, not merged into a single measurement series without qualification.

The reported 2024 SRS figure also includes a rural TFR of 2.1 and an urban TFR of 1.5. That rural–urban difference is large. It confirms that national fertility convergence has not removed geographic and social stratification.

A common error in the India fertility rate trends report is to place a DLHS value, an NFHS value, and an SRS value in one graph, draw a continuous line, and call the slope a precise measure of programme performance. It is not. Differences in sampling frames, reference periods, geographic coverage, questionnaire design, and estimation procedures matter.

The direction of change is robust. The exact slope depends on the data system used.

Proximate determinants: interpreting the Bongaarts model without overstating it

The demographic transition factors in India are often described in broad terms: education, urbanisation, income, women’s employment, media exposure, migration, and health-service expansion. These factors may be relevant. They are not self-executing explanations.

A more disciplined approach separates distal social conditions from proximate determinants. Fertility is directly shaped by exposure to sexual union, contraception, induced abortion, postpartum infecundability, and biological fecundity. Broader social variables operate through these pathways, often in combination.

A 2022 analysis using NFHS-4 data and the Bongaarts proximate-determinants framework estimated the national contribution of several mechanisms to fertility inhibition relative to a biological maximum:

1. Marriage or exposure to union accounted for 36% of the estimated fertility reduction. Later marriage and reduced exposure to marital childbearing lower the number of years in which births are likely to occur.

2. Contraception accounted for 24%. This is a modelled contribution from contraceptive use and effectiveness, not a direct count of births prevented by government programmes.

3. Abortion accounted for 23%. The estimate reflects the model structure and underlying data assumptions. It should not be interpreted as a causal attribution to any single policy or service.

4. Postpartum infecundability accounted for 16%. Breastfeeding patterns and the interval following a birth affect the timing of subsequent conception.

The percentages sum to a framework for interpreting aggregate fertility inhibition. They are not experimental effect estimates. They do not establish that one determinant independently caused a specified share of India’s observed TFR decline. They also cannot be transferred without adjustment to every district, caste group, wealth stratum, or state.

A decomposition model allocates estimated inhibition within a system. It does not assign causal credit.

This matters for DLHS data. DLHS can show where contraceptive use is lower, where unmet need is higher, and where maternal-health contact is weaker. It cannot, without a defined longitudinal design and appropriate controls, prove that a specific district programme reduced fertility by a known quantity.

The appropriate use of district data is narrower and more useful: identify patterns, specify hypotheses, target follow-up, and test whether programme inputs correspond to persistent gaps in reproductive outcomes.

For example, a district with high unmet need and low modern-method use may require analysis of stock-outs, method availability, counselling quality, provider bias, travel time, or postpartum follow-up. The DLHS table is the start of that inquiry. It is not the conclusion.

Contraceptive prevalence and the 21.3% unmet-need gap

The all-India DLHS-3 contraceptive figures require joint interpretation.

Any-method prevalence was 54.0%. Modern-method prevalence was 47.1%. The difference indicates that a measurable share of contraceptive use was accounted for by traditional methods. That is not an automatic sign of programme failure. Some women choose traditional methods. But it does affect the interpretation of service coverage, method effectiveness, discontinuation risk, and counselling needs.

The larger issue is the 21.3% total unmet need for family planning. This indicator includes women who want to postpone the next birth or stop childbearing but are not using a method. It therefore sits at the intersection of stated preference, service availability, perceived side effects, autonomy, partner dynamics, postpartum care, and information quality.

At the national level, this figure does not identify a single deficiency. At district level, it can be disaggregated against complementary measures:

• modern-method use relative to any-method use;
• spacing need relative to limiting need;
• rural and urban residence;
• age group, especially adolescents and women in early childbearing years;
• parity and recent birth history;
• contact with antenatal, delivery, and postnatal services;
• household wealth strata and educational categories, where sample precision permits;
• local availability of sterilisation, reversible methods, and postpartum counselling.

This is where population analytics becomes operational rather than descriptive. A district with high unmet need for spacing is not equivalent to one with high unmet need for limiting. The first may indicate demand among younger women seeking longer birth intervals. The second may signal a failure to convert completed-family preferences into method use. The programme response should not be identical.

Method mix also matters. A high prevalence rate built around one permanent method can coexist with weak access to reversible contraception. That configuration may reduce aggregate fertility while leaving women with limited control over timing and spacing. Conversely, a district with moderate prevalence but a diversified method mix may have a different trajectory and different service needs.

The statistical objective is not to celebrate a lower rate in isolation. It is to determine whether observed fertility change is aligned with informed, voluntary reproductive choice and whether remaining demand is concentrated in identifiable populations.

Regional asymmetry: Bihar, Uttar Pradesh, Meghalaya, and the national mean

NFHS-5 estimated India’s TFR at 2.0, below the replacement benchmark of 2.1. That national position is frequently presented as the end of the demographic transition. It is more accurately described as a transition with uneven timing.

Bihar’s TFR was 3.0 in NFHS-5. Meghalaya was at 2.9. Uttar Pradesh was at 2.4. These are not minor departures from the national estimate. They represent population systems in which age structure, cohort size, reproductive exposure, and service conditions continue to generate higher levels of childbearing.

No single explanation should be attached to these estimates without a specified dataset and model. It is methodologically unsound to infer that a higher state TFR is caused by religion, poverty, education, culture, or health-service access merely because those variables are frequently discussed. Each claim requires measured covariates, a stated comparison group, controls for confounding, and uncertainty intervals.

The relevant demographic observation is simpler. Regional convergence has not occurred.

A total fertility rate regional breakdown should be read alongside population momentum. States that reached low fertility earlier may continue to grow because large cohorts remain in reproductive ages. States with higher TFRs may contribute disproportionately to future births because both the fertility schedule and the age distribution remain younger.

This produces a policy asymmetry. Low-fertility states may face slower school-age population growth, changing labour-force composition, and eventually population ageing. Higher-fertility states face continued demand for maternal services, newborn care, immunisation, schooling, contraception, and employment creation. A national fertility target does not resolve either set of pressures.

The district perspective sharpens this further. State averages conceal internal dispersion. Urban districts can move toward low fertility while remote, rural, or underserved districts retain a different reproductive profile. A state-level mean is therefore insufficient for resource allocation when service delivery is organised through district systems.

Why DLHS cannot be used as a simple district trend panel

DLHS has four rounds, but chronological repetition does not automatically create a longitudinal district dataset.

DLHS-3 covered 601 districts across 34 states and union territories. DLHS-4 did not repeat the same all-India coverage. One methodological description records DLHS-4 coverage of 18 states and 3 union territories, comprising 271 districts. This is a substantial reduction in geographic scope.

The implication is direct. A nationwide district-by-district comparison of DLHS-3 and DLHS-4 is not valid unless the analyst documents matched geography, exclusions, boundary changes, harmonised definitions, and comparable denominators.

Several additional problems arise in trend construction:

1. District boundaries change. A district split creates smaller units with different denominators and service geographies. Comparing the pre-split district with either successor district can manufacture a false trend.

2. Indicator definitions may change. A label can remain constant while question wording, eligibility rules, or classification algorithms differ.

3. Sampling precision varies. District estimates have wider uncertainty than national estimates. Small apparent changes may sit within confidence intervals.

4. Coverage is incomplete across rounds. Absence from a round is not a zero value. It is missing geography.

5. Survey and registration systems differ. DLHS, NFHS, and SRS should be triangulated, not treated as repeated measurements from one instrument.

6. Temporal reference periods differ. A TFR estimate captures fertility over a defined period, while contraceptive indicators refer to current use at interview. They should not be interpreted as simultaneous measurements of a single behavioural state.

A defensible district analysis uses a documented crosswalk. It limits comparisons to matched districts. It reports coverage exclusions. It distinguishes observed changes from changes that may reflect revised boundaries or survey design. It avoids ranking districts when confidence intervals overlap materially.

This is less dramatic than a national heat map. It is also more credible.

A practical analytic route for fertility decline drivers in India

The phrase “fertility rate decline drivers India analysis” implies explanation. Explanation requires a design, not a list of plausible variables.

A robust workflow would proceed in stages.

First, establish the outcome source. If the objective is a current national TFR estimate, use the appropriate NFHS or SRS series and retain its original reference period. If the objective is district service context, use DLHS indicators for the relevant round and geography.

Second, define the unit of analysis. A state, district, rural district population, urban district population, and currently married women aged 15–49 are not interchangeable cohorts. The denominator controls the meaning of the rate.

Third, construct a harmonisation file before modelling. This should record district codes, boundary changes, coverage status by round, indicator definitions, weighting procedures, and missing values. Analysts routinely treat this as administrative work. It is the main safeguard against false temporal inference.

Fourth, separate descriptive association from causal estimation. A cross-sectional correlation between modern contraceptive prevalence and TFR may be informative. It does not prove that increasing prevalence by one percentage point will reduce TFR by a fixed amount. Reverse causation, unmeasured preferences, age composition, and programme placement can all distort the association.

Fifth, report uncertainty. District rankings without standard errors, confidence intervals, or reliability flags are presentation devices, not population analysis. This is especially relevant when comparing small subgroups or looking for changes over short intervals.

Finally, link findings to programme questions that can be acted upon. A district with high unmet need should trigger a service-delivery audit. Are reversible methods available? Is counselling offered after delivery? Are clients returning for resupply? Are frontline workers reaching younger and lower-parity women? Are reported preferences consistent with available methods? These questions are measurable. Broad claims about “awareness” are not enough.

The next fertility transition will be measured in dispersion

India’s fertility decline is real. NFHS-5 placed TFR at 2.0 in 2019–21. The later SRS figure reported for 2024 was 1.9. Both sit below the 2.1 replacement benchmark, though they should not be treated as numerically interchangeable.

The remaining analytical task is distributional. Bihar, Meghalaya, and Uttar Pradesh show that national replacement-level fertility does not imply uniform transition. The rural TFR reported for 2024, 2.1, also indicates that the national figure contains a pronounced rural–urban divide.

DLHS-3 remains valuable because it documents district-level reproductive and family-planning conditions at scale: 720,320 households, 601 districts, and a measurable gap between contraceptive use and unmet need. Its 21.3% unmet-need estimate is not a historical footnote. It is evidence that fertility decline and reproductive-service coverage do not move in perfect synchrony.

The policy implication is narrow. National TFR should be monitored through NFHS and SRS. District intervention should be guided by carefully harmonised DLHS and related RCH indicators. The next useful map is not a map of India’s average fertility. It is a map of where fertility preferences, contraceptive access, and service capacity remain misaligned.