Estimates of the number of births and deaths in India every year come from two key sources: the Civil Registration System (CRS) which relies on registered events, and the Sample Registration System (SRS), which is a large demographic survey.
The CRS operates under the provisions of the Registration of Births and Deaths Act, 1969, which makes birth and death registration mandatory across India. Despite its legal mandate, problems of under-registration and irregular reporting persisted for many years[1], limiting the use of CRS data in demographic research and planning until recently.
Against this backdrop, the Registrar General of India introduced the SRS in the 1960s. Designed as a sample-based registration system, the SRS was intended to provide reliable estimates of key demographic indicators at the national and state levels. Over time, it expanded in both scope and application and came to be regarded as the benchmark for demographic estimates in India. SRS data is widely used by researchers and policymakers to study population dynamics of India and states. Both the UN Population Division and the Registrar General of India rely heavily on SRS estimates for population projections.
Moreover, SRS estimates are used to assess the completeness of birth and death registration under the CRS.
(See our work for details on the methodology of the CRS and SRS.)
Over time, questions have emerged over the completeness and appropriateness of the use of both data sources. We examine different approaches to estimating births and deaths in India, their successes and shortcomings, and suggest in which circumstances which one might be better to use.
Measuring completeness
The CRS-SRS approach
CRS estimates of births and deaths come from the civil registration system, meaning from births and deaths registered with local authorities, in accordance with Indian law. SRS estimates, meanwhile, come from applying birth and death rates produced by the SRS survey to projected population figures for a given year. The level of birth and death registration in the CRS is estimated by comparing the number of registered birth and death counts from the CRS with the expected number of births and deaths from the SRS, for a given year, expressed as a percentage. We call this the CRS-SRS-based approach to estimating registration completeness. If a state were to successfully record all births and deaths, and if SRS estimates of births and deaths are accurate, the level of registration in the state should be close to 100 percent. Instead, we see large differences in both directions.
In the past decade, CRS-registered deaths in many states have frequently exceeded SRS-based estimates, often by substantial margins. Seven out of 18 major states[2] have consistently reported a higher number of registered deaths than SRS-based estimates of deaths. Between 2015 and 2023, Tamil Nadu reported over 100 percent registration in all nine years, while Kerala, Punjab and Haryana did so in eight years. Karnataka, Odisha and Gujarat reported such levels in seven years. The gap between registered and estimated deaths reached as high as 47 percent, with about one-third of the major states frequently reporting gaps exceeding 10 percent,[3] implying registration levels above 110 percent. These states include Tamil Nadu, Karnataka, Haryana, Kerala and Andhra Pradesh. One explanation for these discrepancies could be that the SRS is systematically undercounting deaths at least in the states where the level of death registration is recorded above 110 percent.
Moreover, since 2019, India's overall level of death registration has been close to 100 percent, even though two of its most populous states-Uttar Pradesh and Bihar, which together account for nearly one-fourth of the country's population-continue to report registration levels below 80 percent and 60 percent, respectively. This raises a clear inconsistency: it is implausible that deaths from these large states are being registered elsewhere in a way that offsets these deficits and effectively brings India's overall registration level close to 100 percent. Migration alone is unlikely to explain this offset.[4]
In the case of births, four major states have reported registration levels exceeding 100 percent frequently between 2015 and 2023-Telangana for six years, and Kerala, Assam and West Bengal for five years each. However, the gap between registered and estimated births generally remained below 10 percent. This deviation is more likely attributable to sampling variation in SRS or delayed registration in CRS-such as the recording of births from earlier years due to policy or documentation requirements or migration-rather than a systematic bias in SRS birth rate estimates.
The survey-based approach
Another way to assess the completeness of the CRS is to compare it with survey-based estimates of birth and death registration levels. Recent rounds of the National Family Health Survey (NFHS) in 2015-16 and 2019-21, as well as the National Sample Survey (NSS) in 2020-21 and 2022-23, included direct questions on whether a child's birth was registered with a civil authority. The NFHS-5 also introduced a question on death registration, asking the household head whether any death that occurred in the preceding three years had been registered.
Estimates from these surveys present a markedly different picture from those derived using the CRS-SRS-based approach for death registration. While CRS-SRS-based assessments suggest that death registration in India has risen from 85 percent in 2018 to 100 percent in 2020, NFHS-5 indicates substantially lower levels, around 70 percent for the same period.[5]
This divergence is not confined to the national level but is systematically evident across states. In all 18 major states, survey-based estimates of death registration are consistently lower than CRS-SRS-based estimates. In as many as 15 states, the gap exceeds 10 percentage points. The differences range from about 6 percentage points in states such as Chhattisgarh and Kerala to as high as 40 percentage points in states like Tamil Nadu and Andhra Pradesh. Even in states where the CRS-SRS approach finds registration levels above 100 percent, the NFHS reported substantially lower levels of death registration. Kerala is the only state that reported near-universal death registration in NFHS (98 percent).
Although survey responses from NFHS, reported by household heads, are likely to be affected by recall or reporting bias, the scale and consistency of these gaps point to deeper concerns. Viewed alongside earlier evidence of registration levels exceeding 110 percent, these findings suggest that the CRS-SRS-based approach may be overstating registration completeness and the SRS is likely to be systematically underestimating deaths in India and several states.
In the case of births, survey-based estimates of the level of registration are available from both NFHS and NSS, and the two align closely, with a high rank correlation (around 0.9),[6] indicating strong consistency between these two survey-based measures. At the all-India level, estimates of the level of registration from all three sources are broadly comparable, at around 90 percent. Differences between survey-based estimates and CRS-SRS estimates of the level of birth registration are relatively modest and non-systematic across states.[7] Five out of 18 states show a difference of more than 10 percentage points between CRS-SRS estimates and both survey-based estimates of the level of registration. Among these, Bihar and Telangana report higher CRS-SRS-based levels of registration, suggesting possible underestimation of births in the SRS, while Madhya Pradesh, Chhattisgarh and Tamil Nadu report lower levels, indicating potential overestimation of births in the SRS for these states. Three out of 18 major states-Kerala, West Bengal and Assam-reported birth registration levels above 95 percent consistently across all three sources: CRS-SRS-based estimates, NFHS-5 (2019-21), and NSS-78 (2020-21).
This suggests that levels of birth registration are broadly consistent across the CRS-SRS-based approach and survey-based approach, with only modest and non-systematic differences across a few states. In contrast, the large and persistent gaps observed for levels of death registration indicate that the CRS-SRS-based approach may overstate registration completeness, likely due to systematic underestimation of deaths in the SRS.
How complete is the SRS?
The Registrar General and Census Commissioner of India (RGI) and independent researchers have also conducted several evaluations of the SRS. Early evaluations conducted during the 1970s and 1980s, based on both indirect demographic techniques[8] and intensive field enquiries, generally indicated that omissions in SRS were within a modest range of about 3 to 10 percent.[9]
Further evidence from subsequent assessments reinforces this pattern of relatively modest but non-negligible omission. Using indirect demographic methods, the RGI estimated around 6 percent under-reporting of births in 1978, with substantial interstate variation ranging from less than 1 percent in Gujarat to over 17 percent in Karnataka.[10] Direct estimation studies based on intensive enquiry of sub-samples also indicate continued omission in SRS estimates.[11]
A recent evaluation finds that there was an undercounting of deaths in India by around 4.3% for males and 11.3% for females during 2001-10.[12]
Taken together, these evaluations suggest that while SRS omission rates at the aggregate level have broadly remained within the range of 10 percent, the extent of undercounting, particularly for deaths, is non-trivial and varies across states, time periods and demographic groups.[13]
Sources of omission
The reasons for omission of vital events in the SRS are not well-researched and are not systematically documented in the existing literature. Several plausible factors could contribute to such underreporting. The SRS follows a dual record system, combining continuous enumeration with independent half-yearly surveys across all households within selected sampling units.
The sampling frame is updated only after each census and could become outdated over time, limiting its representativeness at national and sub-national levels.[14] Rapid demographic changes-such as migration, urbanisation and uneven population growth-can lead to coverage errors and biased demographic estimates, as the sample may not reflect the current population distribution and composition. This could distort key indicators like birth and death rates, affect the accuracy of population projections.
In SRS, the enumerator is expected to record all births and deaths occurring within the sample unit, as well as those of usual residents occurring outside it. If a household migrates out of the sample unit, enumerators attempt to continue tracking its members to record births and deaths. However, maintaining such follow-up is not always possible. As a result, some events may go unrecorded, leading to omissions in the data.
Can we rely on CRS estimates?
Our analysis identifies Kerala as the most appropriate case for assessing the usability and accuracy of CRS-based demographic estimates.
For birth registration, Kerala, Assam and West Bengal reported near-universal coverage in CRS-SRS approach as well as survey-based approaches; however, detailed annual data on live births by mother's age are available only for Kerala. We therefore use Kerala's CRS data to compute age-specific fertility rates and compare them with estimates from SRS, NFHS, and Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) for the years 2017-19.[15] The GBD Study, led by the Institute for Health Metrics and Evaluation (IHME), produces model-based estimates of fertility, mortality, population and health indicators by combining multiple data sources-including censuses, surveys, registration systems, and other studies-and applying demographic and statistical modelling techniques to these inputs.[16]
The comparison of age-specific fertility rates shows that CRS estimates closely follow the patterns observed in NFHS and GBD across age groups in Kerala. SRS reported lower fertility rates in the early reproductive ages (15-29), suggesting a likely underestimation of births in this segment in the state.
A similar pattern is observed for mortality.
For death registration, Kerala is the only state with near-universal coverage. We use Kerala's age- and sex-specific death counts from the CRS to compute age-specific death rates separately for males and females for the years 2017-19. These estimates are then compared with corresponding estimates from SRS and GBD.[17] This exercise enables us to assess whether CRS data can be reliably used to generate key demographic indicators such as age-specific and total fertility rates, age-specific death rates, and life expectancy.
CRS-based age-specific death rates closely align with GBD estimates across most age groups for both males and females, with SRS broadly following the same pattern. However, at older ages-particularly in the open-ended age group (70+)-SRS reports noticeably lower death rates, while CRS and GBD remain closely aligned. This divergence suggests that SRS may be undercounting deaths at older ages. A study using anonymized individual-level CRS data for Kerala finds no evidence that CRS underestimates mortality; instead, CRS produces smoother and more stable age patterns than SRS. It also reports higher death rates than SRS in age groups where mortality is typically low and among women, suggesting possible undercounting in SRS.[18]
Overall, these results suggest that in states with near-universal registration,[19] CRS data can generate age-specific fertility and mortality estimates that are highly consistent with both survey-based sources such as NFHS and model-based estimates such as GBD. Given that the SRS was originally designed to compensate for incomplete civil registration, the evidence from Kerala indicates that where CRS systems are sufficiently complete and robust, they can serve as a reliable and regular alternative to SRS for producing key demographic indicators. For such states, CRS can potentially replace SRS as the primary source of timely and consistent demographic data.
However, to fully realize this potential, CRS data should be made available in anonymized, individual-level form to enable the computation of detailed indicators. CRS should also publish birth and death counts by usual place of residence, in addition to place of occurrence, to enable meaningful analysis of rural-urban differentials in demographic indicators. Reliance solely on place of occurrence can lead to misclassification-for instance, events from rural areas being recorded in urban locations-thereby limiting the representativeness of rural-urban estimates.
In states where registration is not complete, SRS could continue to remain the main source for demographic estimates until the completeness of CRS registration improves further. At the same time, there is a need for continued assessment of both the completeness of CRS registration and the quality and coverage of SRS estimates across states.
[1] Estimating mortality using data from civil registration: a cross-sectional study in India (2015), Mamta Gupta et al., Bulletin of the World Health Organization.
[2] Major states refer to those with a population exceeding 2.5 million, as per the 2011 Census; smaller states are excluded due to inadequate sample sizes in the SRS and other surveys used in this analysis to generate accurate and reliable estimates of age-specific fertility and mortality rates.
[3] A 10 percent threshold is used here as an indicative margin of error. Since SRS estimates are based on a sample survey and subject to sampling variation, small differences are expected. To assess the likely range of such variation, expected numbers of births and deaths were computed using both the point estimates and the upper and lower confidence interval bounds of SRS rates. In most states and years, the minimum-maximum expected estimates generally remained within ±10 percent of the point estimates. Differences beyond 10 percent are therefore unlikely to be explained by sampling variation alone.
[4] If out-migrants from Bihar and Uttar Pradesh experienced mortality rates similar to residents of their origin states, then compensating for the observed deficits would require out-migrant populations amounting to nearly 40 percent of Bihar's population and 20 percent of Uttar Pradesh's population. However, Census 2011 data indicate that out-migrants constitute less than 10 percent of the populations of these states, making it implausible that migration alone could offset such large registration deficits.
[5] NFHS-5 (2019-21) provided the percentage of deaths in the last three years reported as registered with a civil authority.
[6] Rank correlation measures how similarly units (such as states) are ranked across two measures. A value of 0.9 indicates very strong agreement, meaning states that rank high in terms of proportion of births registered with civil authority in one survey also tend to rank high in the other.
[7] NFHS-5 (2019-21) & NSS-79 (2020-21) shows the percentage of births in the last three years reported as registered by households. CRS and SRS figures are averaged over 2018-20 to ensure comparability.
[8] Indirect demographic techniques refer to methods that estimate demographic indicators from incomplete or imperfect data using mathematical models. For example, the proportion of children dead among those ever born to women aged 20-24 years can be used to estimate the probability of dying before age two.
[9] The Sample Registration System (SRS) in India, An Overview (2017), Prasanta Mahapatra.
[10] Estimates of Fertility and Child Mortality by Indirect Methods (1983), Registrar General of India.
[11] Report on the Intensive Enquiry Conducted in a Sub-sample of SRS Units (1988), Registrar General of India.
[12] There Is a Glaring Gender Bias in Death Registrations in India (2019), Yadav & Ram, EPW Engage.
[13] The Sample Registration System (SRS) in India, An Overview (2017), Prasanta Mahapatra.
[14] Household Sample Surveys in Developing and Transition Countries (2005), United Nations.
[15] NFHS-5 provides Age-Specific Fertility Rates (ASFR) based on births in the three years preceding the survey; for Kerala, this corresponds to 2017-2019. Accordingly, CRS, SRS and GBD estimates are averaged over 2017-2019 for comparability.
[16] Global Burden of Disease, Institute for Health Metrics and Evaluation.
[17] CRS publications provide death counts by selected age intervals (<1, 1-4, 5-14, 15-24, …, 55-64, 65-69, and 70+), while SRS and GBD estimates are primarily available in five-year age groups. For comparability, SRS and GBD estimates are re-aggregated to match the CRS age intervals.
[18] Assessing mortality registration in Kerala: the MARANAM study (2022), Gupta & Mani, Genus.
[19] The characterisation of "near-universal" registration is based on the convergence of estimates derived from the CRS-SRS comparison and independent survey-based measures such as NFHS and NSS. In scientific analysis, consistency across multiple data sources-each with distinct methodologies and potential sources of error-is generally taken to enhance the reliability of estimates, as it reduces the likelihood that observed patterns are driven by source-specific biases.
