An Evaluation of Recent
Estimates of Fertility
Trends in India
Robert D. Retherford and Vinod Mishra
National Family Health Survey Subject Reports
Number 19 • November 2001
International Institute for Population Sciences
Mumbai, India
East-West Center, Population and Health Studies
Honolulu, Hawaii, U.S.A.
Correspondence addresses:
International Institute for Population Sciences
Govandi Station Road, Deonar, Mumbai - 400 088, India
Fax: 91-22-556-3257 • E-mail: iipsnfhs@vsnl.com
East-West Center, Population and Health Studies
1601 East-West Road, Honolulu, Hawaii 96848-1601, U.S.A.
Fax: 1-808-944-7490 • E-mail: poppubs@eastwestcenter.org
India’s first and second National Family Health Surveys (NFHS-1 and NFHS-2) were conducted
in 1992–93 and 1998–99 under the auspices of the Ministry of Health and Family Welfare. The
surveys provide national and state-level estimates of fertility, infant and child mortality, family
planning practice, maternal and child health, and the utilization of services available to mothers
and children. The International Institute for Population Sciences, Mumbai, coordinated the
surveys in cooperation with selected population research centres in India, the East-West
Center in Honolulu, Hawaii, and ORC Macro in Calverton, Maryland. The United States Agency
for International Development, via a subcontract from ORC Macro to the East-West Center,
provided support for the preparation and publication of this report.
ISSN 1026-4736
This publication may be reproduced for educational purposes.
An Evaluation of Recent
Estimates of Fertility
Trends in India
ABSTRACT
Comparison of fertility estimates from India’s Sample Registration System (SRS) and
two recent National Family Health Surveys (NFHS-1 in 1992–93 and NFHS-2 in 1998–
99) indicates major discrepancies between the three sources. This subject report
attempts to evaluate the reasons for these discrepancies and to assess the true trend
of fertility in all India and 16 major states.
The analysis indicates that, for all India, the true total fertility rate (TFR) for
1990–92 was probably around 3.92, and the true TFR for 1996–98 was probably be￾tween 3.39 and 3.55, somewhat higher than the raw SRS estimates and considerably
higher than the raw NFHS-1 and NFHS-2 estimates for the same time periods. The
new estimates correct for displacement of births in NFHS-1 and NFHS-2 and for
underregistration of births in the SRS. Although the new TFR estimates are higher, the
decline in the TFR between 1990–92 and 1996–98 (0.4 or 0.5 child) derived from them
is consistent with the declines calculated directly from the SRS (0.4 child) and from
NFHS-1 and NFHS-2 (0.5 child).
Misreporting of women’s ages in the two NFHS surveys does not have a large
effect on the estimates of the TFR, for either India or major states. On the other hand,
it does have a large effect on the estimates of ASFRs (age-specific fertility rates).
Because the extent of age misreporting did not change much between the two surveys
in India as a whole, our survey-based estimates of changes in ASFRs for all India are
probably fairly accurate, insofar as errors in the estimates of ASFRs tend to cancel
out when computing changes. In most individual states, however, the extent of age
misreporting changed enough between the two surveys that we are unable to arrive
at accurate estimates of changes in ASFRs. In the SRS, the pattern of change in
ASFRs is distorted at both the national and state levels by changes in the pattern
of age misreporting caused by the phasing in of a new SRS sample during the
period 1993–95.
4
National Family Health Survey Subject Reports, No. 19
Robert D. Retherford and Vinod Mishra
Robert D. Retherford is Coordinator, and Vinod Mishra is a Fellow in Population and
Health Studies at the East-West Center in Honolulu, Hawaii.
National Family Health Survey Subject Reports, Number 19 • November 2001
The analysis demonstrates that calculating TFR estimates for the 3-year period
immediately preceding a survey does not work well in India. In NFHS-1 and NFHS-2,
widespread ignorance of children’s ages results in substantial displacement of births
to earlier years in the birth histories and therefore to underestimation of the TFR for
this 3-year period. Moreover, because the extent of displacement changed in many
Indian states between NFHS-1 and NFHS-2, due to differences in the diligence of the
survey organizations in collecting accurate age data, the TFR estimates for the 3-year
period before each survey yield inaccurate estimates of the trend in the TFR in these
states. In this situation, the own-children method of fertility estimation, applied to the
household samples of NFHS-1 and NFHS-2, is a useful supplement to the birth-his￾tory method in that it yields TFR estimates for longer reference periods that minimize
bias from displacement and provide more accurate indications of both level and trend
in the TFR.
The analysis of bias in fertility estimates from the SRS could be improved further
if the Sample Registration System were to publish unsmoothed single-year age-sex
distributions of the SRS sample as well as sex ratios at birth on an annual basis.
5
National Family Health Survey Subject Reports, No. 19
P
ublished reports from India’s 1992–93 and 1998–99 National Family Health
Surveys (NFHS-1 and NFHS-2) have shown that estimates of the total ferti￾lity rate (TFR) for the 3-year period before each survey are consistently lower
than corresponding estimates from India’s Sample Registration System (SRS), for
both all India and individual states (IIPS 1995; IIPS and ORC Macro 2000). In addi￾tion, discrepancies among the three sources in the TFR estimates for the state of
Uttar Pradesh have been investigated in some depth in a recent study by Retherford
et al. (2001). The present report extends Retherford et al.’s analysis for Uttar Pradesh
to all India and an additional 15 major states. The goal of the analysis is to reconcile
differences between the three sources of TFR estimates.
DATA
NFHS-1 and NFHS-2 were designed along the lines of the Demographic and Health
Surveys (DHS) that have been conducted worldwide in many developing countries
during the past two decades. Both NFHS-1 and NFHS-2 are nationally representa￾tive surveys that include both a household sample, covering everyone in the sampled
households, and an individual sample, covering all ever-married women age 15–49
(13–49 in the case of NFHS-1) within those households. Corresponding to these two
samples are a household questionnaire and an individual questionnaire. In the case of
the household questionnaire, the household head or other knowledgeable adult in the
household responded for the entire household. In the case of the individual question￾naire, each ever-married woman responded for herself and her children. NFHS-1
comprises 88,562 households and 89,777 ever-married women. NFHS-2 comprises
92,486 households and 90,303 ever-married women. Details of the sample design are
contained in the basic survey reports for NFHS-1 and NFHS-2 (IIPS 1995; IIPS and
ORC Macro 2000).
The International Institute for Population Sciences in Mumbai, which was the
nodal agency for the two surveys, contracted with 7 survey organizations in NFHS-1
and 13 survey organizations in NFHS-2 to conduct the fieldwork in the various states.
In some cases the survey organizations were private survey-research organizations,
and in other cases they were population research centers or other organizations.
Each contractor handled one or more states.
India’s Sample Registration System (SRS) also provides national and state-level
estimates of fertility, and it does so on an annual basis. The system includes both
continuous registration of births and six-monthly surveys to catch missed births. The
SRS is based on a nationally representative sample of villages and urban blocks. A
new sample is selected every 10 years, following the decennial census. The most
recent sample was phased in during 1993–95, using the 1991 Census as the sampling
6
National Family Health Survey Subject Reports, No. 19
frame. In the SRS sample, the sampling unit in rural areas is an entire village or
segment of any village with a population of 1,500 or more. In urban areas the sam￾pling unit is a census enumeration block with a population ranging from about 750 to
1,000. The SRS sample currently includes 6,671 sampling units (4,436 in rural areas
and 2,235 in urban areas), comprising about 1.1 million households and about 6 million
population. The structure and procedures of the SRS have been described in more
detail by Retherford et al. (2001).
DATA QUALITY
Data quality in NFHS-1 and NFHS-2
This section focuses on all India. Discussion of data quality in each of the 16 major
states is deferred to the section of the report that analyzes state-level fertility esti￾mates.
In NFHS-1, both the household response rate and the individual response rate
for all India were 96 percent (IIPS 1995). In NFHS-2, the household response rate
was 98 percent, and the individual response rate was 96 percent. Nonresponse is rare
enough that it probably does not significantly bias the fertility estimates derived from
NFHS-1 and NFHS-2.
Age misreporting is a more serious problem. Accuracy of age reporting is criti￾cal for calculating accurate estimates of fertility derived by either the birth-history
method or the own-children method, which are the two methods used here to obtain
fertility estimates from NFHS-1 and NFHS-2. In all India, only about one-fourth of
ever-married women age 15–49 know their year of birth. For about three-fourths of
the women, interviewers had to make educated guesses of age. In many cases, inter￾viewers also had to make educated guesses of children’s ages, inasmuch as many
mothers could not report these ages accurately. There is, accordingly, great potential
for bias in the estimates of fertility derived from the two surveys.
Figures 1a and 1b show single-year age distributions of females in NFHS-1 and
NFHS-2 for all India. These age distributions are based on the household sample
(including persons of all ages), except that ages of ever-married women age 15–49
(13–49 in NFHS-1) are copied over from the individual sample. (This is done because
women’s self-reports of their age tend to be more accurate than the reports of the
household head or other knowledgeable person in the household who responded for
the household in the household questionnaire). Both surveys show evidence of heap￾ing on ages ending in 0 or 5, and the extent of such heaping is about the same in both
surveys. This is also indicated by Myers’ Index of age heaping among women age
7
National Family Health Survey Subject Reports, No. 19
Figure 1a Female age distribution, with ages of ever-married women age 13–49 taken from the
individual questionnaire and ages of other women taken from the household questionnaire: India,
NFHS-1, 1992–93
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
Age
Percent
8
National Family Health Survey Subject Reports, No. 19
Figure 1b Female age distribution, with ages of ever-married women age 15–49 taken
from the individual questionnaire and ages of other women taken from the household
questionnaire: India, NFHS-2, 1998–99
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
Age
Percent
9
National Family Health Survey Subject Reports, No. 19
10–69 (Shryock and Siegel 1973), which declined slightly from 10.4 to 10.0 between the
two surveys. Although the extent of age misreporting for women age 10–69 did not
change much between the two surveys at the all-India level, this is not true for many
Indian states considered individually, as will be seen later. In some of states the quality of
age reporting improved between the two surveys, while in others it worsened.
Figures 1a and 1b show considerable heaping on ages 8, 10, and 12, which
translates into peaks in the fertility estimates derived from NFHS-1 and NFHS-2 in
the 9th, 11th, and 13th years before the surveys. An additional problem contributing to
misreporting of children’s ages is that in both NFHS-1 and NFHS-2, young children
who were reported as older than a cut-off age became ineligible for a large block of
questions pertaining to children below this age, offering interviewers an opportunity to
reduce their workload by exaggerating children’s ages beyond the cut-off age. The
cut-off age for the large block of questions pertaining to young children was approxi￾mately age 4.5 years in NFHS-1 and age 3.5 years in NFHS-2. (These cut-off ages
are approximate because in the questionnaires the cut-off was specified as a date
rather than an age.)
As shown in Figures 1a and 1b, in NFHS-1 there is a substantial jump between
the proportion age 4 and the proportion age 5, and in NFHS-2 there is a substantial
jump between the proportion age 3 and the proportion age 4, suggesting that some
interviewers did indeed exaggerate children’s ages in order to lighten their workload.
This type of age exaggeration, which is equivalent to displacing births backward in
time, has the effect of lowering fertility estimates for the 3-year period immediately
preceding the survey, which is the period typically used for fertility estimates in demo￾graphic and health surveys, including NFHS-1 and NFHS-2. Because of the lower
age cut-off in NFHS-2, there may be more displacement of births out of this 3-year
period to earlier years in NFHS-2 than in NFHS-1. Some interviewers may also have
intentionally omitted some births that occurred in the 3-year period before the survey
in order to avoid having to ask the block of questions pertaining to young children.
Because of considerable preference for sons over daughters in India (mainly in
the northern states), estimated sex ratios at birth provide another useful indicator of
data quality. The sex ratio at birth is biologically determined for the most part and is
usually close to 105 male births for every 100 female births, or a ratio of 1.05. In
national populations with complete or near-complete vital registration, the sex ratio at
birth rarely exceeds 1.06 (Keyfitz and Flieger 1968). If female births are omitted
more than male births, this ratio will tend to be higher than 1.06. This occurs when
women who prefer sons over daughters forget to mention children who have died or
moved away, but more frequently daughters than sons. Such omissions are more
likely for children born several years before the survey than for children born closer
to the survey. If such omissions are common, one expects the sex ratio at birth, as
10
National Family Health Survey Subject Reports, No. 19
ascertained from women’s birth histories, to be higher in earlier years. In NFHS-1,
the sex ratio at birth was 1.05, 1.07, and 1.06 in the first, second, and third 5-year
periods before the survey. In NFHS-2, it was 1.07, 1.08, and 1.09 in these three
periods. Overall, for the 15-year period before each survey, the sex ratio at birth was
1.06 in NFHS-1 and 1.08 in NFHS-2. These results suggest that both surveys omitted
some births, and that omissions were more frequent in NFHS-2 than in NFHS-1.
Sex-selective abortion has entered the picture since NFHS-1, however, so that the
increase in the sex ratio at birth between NFHS-1 and NFHS-2 might be due mainly
or even entirely to an emerging pattern of disproportionate abortion of female fe￾tuses. Thus, the increase in the sex ratio at birth between the two surveys does not
necessarily indicate a higher proportion of omitted births in NFHS-2.
Data quality in the SRS
In the SRS, the extent of age misreporting cannot be ascertained easily, because the
SRS does not publish age distributions by single years of age. Thus, one cannot exam￾ine heaping on ages ending in 0 and 5 or compute a value of Myers’ Index. With the
exception of children below ten years of age whose births were registered, however,
age misreporting is likely to be much worse in the SRS than in NFHS-1 and NFHS-2,
because in the training and supervision of interviewers, the two surveys devoted
much more effort to obtaining accurate information on age than is typically done in
the SRS.
Despite the lack of published single-year age data from the SRS, it is possible to
infer that the pattern of age misreporting in the SRS changes over time, due to the
phasing in of a new sample of registration areas (villages and urban blocks) every ten
years. Since the early 1980s, the SRS sample has been replaced twice, once during
1983–85 using the 1981 Census as the sampling frame and again during 1993–95
using the 1991 Census as the sampling frame. The impact of the 1993–95 changeover
on age misreporting and on SRS fertility estimates will be analyzed later.
The SRS does not normally report sex ratios at birth, but a special report did
release, on a one-time basis, national and state-level sex ratios at birth from the SRS
for the period 1981–90 (Registrar General, India 1996). The SRS sex ratio at birth for
all India for this 10-year period was 1.10, indicating substantial underregistration of
female births and undoubtedly a good deal of underregistration of male births as well,
implying that the SRS estimates of the TFR for all-India are too low. Sex-selective
abortion was probably negligible in India during the period 1981–90 and therefore
cannot explain the high sex ratio at birth during that period. Because this is the only
reported sex ratio at birth ever released by the SRS for all India, it is not possible to
ascertain the trend in the sex ratio at birth in the SRS.
11
National Family Health Survey Subject Reports, No. 19
Although the SRS suffers from underregistration of births, it does not suffer
from displacement of births to earlier years, because births are registered in the year
in which they occur. Later in this report we estimate that, in all India, the proportion of
births that are not registered in the SRS is in the range of 7 to 10 percent.
THE BIRTH-HISTORY METHOD AND THE OWN-CHILDREN METHOD
The fertility measures estimated in this report are age-specific fertility rates (ASFRs),
the general fertility rate (GFR), and the total fertility rate (TFR). For a particular
calendar year, an ASFR for a 5-year age group is calculated as the number of births
to women in the age group during the year divided by the mid-year number of women
in the age group. Such a rate has the dimensions of births per woman per year. The
GFR is defined as births during the year to women age 15–49 divided by the mid-year
population of women age 15–49. The GFR also has the dimensions of births per
woman per year. The TFR is defined as the number of children a woman would have
if she lived through her reproductive years experiencing current age-specific birth
rates. It is calculated by adding ASFRs in 5-year age groups from 15–19 to 45–49
and multiplying the sum by five. The TFR has the dimensions of births per woman
over her entire reproductive lifetime.
In this report, annual fertility estimates from the SRS are taken from published
SRS reports. Fertility estimates from NFHS-1 and NFHS-2 are derived using two
methods: (1) the birth-history method and (2) the own-children method. In the present
analysis, the own-children method is preferred over the birth-history method, for rea￾sons explained below.
The birth-history method of estimating fertility
The birth-history method is straightforward. One simply counts births by age of mother
as reported in the birth histories for each year up to the 15th year before the survey.
One similarly counts woman-years of exposure to the risk of birth by woman’s age.
Births by age of mother are then divided by woman-years of exposure in each age
group in each calendar year or group of calendar years to obtain estimates of ASFRs
for the same period. TFRs are derived from the ASFRs, by summing the ASFRs in
five-year age groups from 15–19 to 45–49 and multiplying the sum by five. In the
calculation of these various fertility rates, which pertain to all women, not just
ever-married women, it is assumed that never-married women have had no births.
In the present context, the principal disadvantage of the birth-history method is
that it cannot be used to calculate TFRs for each of the 15 years before the survey
(either NFHS-1 or NFHS-2). This is so because, following usual practice, NFHS-1
12
National Family Health Survey Subject Reports, No. 19
and NFHS-2 collected birth histories from ever-married women only up to age 50.
Thus, the oldest women covered were only 44 years old five years before each
survey, so that an age-specific fertility rate (ASFR) for women age 45–49 cannot be
calculated for years earlier than five years before the survey. Similarly, 15 years
before each survey, the oldest women in the sample were only 34 years old. Thus,
comparable measures of overall fertility for each of the 15 years before the two
surveys can be based only on fertility up to age 35. A suitable summary measure of
fertility that is comparable over the entire period is CFR(35), the cumulative fertility
rate up to age 35. This measure is calculated by adding ASFRs in 5-year age groups
from 15–19 to 30–34 and multiplying the sum by five.
The own-children method of estimating fertility
Because the own-children method (Cho et al. 1986) is applied to the household sample
(with ages of ever-married women of reproductive age copied over from the indi￾vidual sample), covering all persons in the household with no age restrictions, it allows
computation of ASFRs and the TFR for each of the 15 years before each survey. For
that reason it is our preferred fertility estimation method.
In the own-children method, enumerated children are first matched to mothers
within households, based on answers to questions on age, sex, marital status, and
relation to head of household. A computer algorithm is used for matching. The matched
(i.e., own) children, classified by their own age and mother’s age, are then reverse￾survived to estimate numbers of births by age of mother in previous years. Reverse￾survival is similarly used to estimate numbers of women by age in previous years.
After adjustments are made for unmatched (i.e., non-own) children, ASFRs are cal￾culated by dividing the number of reverse-survived births by the number of reverse￾survived women. TFRs are then calculated from the ASFRs.
The own-children method may be viewed as fertility estimation from incomplete
birth histories, where the missing births correspond to children under age 15 who are
either dead or no longer living in the household at the time of the survey. The own￾children method uses reverse-survival and other approximation procedures to add
these missing births back in. Life tables for the reverse-survival calculations are
taken from the SRS (Registrar General, India 1986; 1990; 1995; 1998).
In applying the own-children method, calculations are initially done by single
years of age and time. Estimates of ASFRs for grouped ages or grouped calendar
years or both are obtained by aggregating single-year numerators (births) and single￾year denominators (women) separately over the appropriate ages and calendar years
and then dividing the aggregated numerator by the aggregated denominator. Such
aggregation is useful for minimizing the distorting effects of age misreporting on the
13
National Family Health Survey Subject Reports, No. 19
fertility estimates. Because numerators and denominators are aggregated separately
before dividing to get ASFRs, the ASFRs take into account population growth over
the estimation period and are therefore implicitly weighted by population size. Further
details of the calculations have been described by Retherford et al. (2001).
In order to validate the use of the own-children method instead of the birth￾history method for deriving fertility estimates for all India, we compare fertility esti￾mates derived by the two methods. Because of the problem of age truncation in the
birth-history method as one goes back in time, comparable estimates of fertility for
each of the 15 years preceding NFHS-1 or NFHS-2 can be generated only up to age
35. For this reason, an appropriate summary measure for comparing the two methods
is cumulative fertility up to age 35, or CFR(35). Figure 2 shows 15-year trends in
CFR(35) for all India derived from the birth-history and own-children methods for
each of the two surveys. Figure 2a compares fertility trends derived by the own￾children method and the birth-history method applied to NFHS-1, and Figure 2b com￾pares trends derived by the two methods applied to NFHS-2. In each case, the two
sets of estimates are in close agreement, justifying the use of the own-children method
in the remainder of this report.
EARLIER WORK ON FERTILITY ESTIMATES FOR UTTAR PRADESH
In the case of the state of Uttar Pradesh, discrepancies in TFR estimates between
NFHS-1, NFHS-2, and the SRS have been investigated in some depth in a recent
study by Retherford et al. (2001). NFHS-1 and NFHS-2 indicate that in Uttar Pradesh
the TFR for the 3-year period before each survey fell from 4.82 children per woman
in 1992–93 to 3.99 children per woman in 1996–98 (IIPS and ORC Macro 2000). In
contrast, the SRS indicates that the TFR fell from 5.17 to 4.78 over the same period.
Retherford et al. find that, because of underregistration of births in the SRS, the true
level of fertility is even higher. They estimate that the TFR in Uttar Pradesh fell from
5.55 to 5.19 between 1990–92 and 1996–98.
Briefly, the methodology used by Retherford et al. is the following: In order to
minimize bias from displacement of births to earlier years before NFHS-2, the au￾thors first calculated a TFR for the 15-year period before the survey— i.e., for 1984–
98—using the own-children method. Based on a post-enumeration survey that was
conducted for NFHS-2 in Uttar Pradesh (but not in other states), this 15-year TFR
was adjusted upward by 4.72 percent to adjust for births that were missed in the main
survey, yielding a TFR value of 5.5503. Next the authors calculated a correction
factor for TFRs from the SRS as the ratio of 5.5503 (from NFHS-2, adjusted) to the
unweighted average of annual TFRs published by the SRS for the same 15-year time
period. The value of this ratio is 1.0639. Under the somewhat heroic assumption that
14
National Family Health Survey Subject Reports, No. 19
Figure 2a Birth-history and own-children estimates of cumulative fertility rates,
CFR(35), from NFHS-1: India, 1978–92
Figure 2b Birth-history and own-children estimates of cumulative fertility rates,
CFR(35), from NFHS-2: India, 1984–98
0
1
2
3
4
5
6
78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93
Year
CFR(35)
Birth History Own Children
0
1
2
3
4
5
6
84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
CFR(35)
Birth History Own Children
15
National Family Health Survey Subject Reports, No. 19
the degree of completeness of birth registration in the SRS did not change in Uttar
Pradesh between 1978 and 1998, the correction factor of 1.0639 was then used to
correct each annual SRS estimate of the TFR between 1978 and 1998. This was
done by multiplying each unadjusted annual SRS estimate of the TFR by 1.0639 to
yield a set of adjusted annual estimates. A line was then fitted through the adjusted
annual estimates. The equation of the line is TFR = 11.0639 – .0603 t, where t is years
since 1900. Estimates of the TFR for the 3-year period preceding each survey were
then obtained by substituting t = 91.5 and t = 97.5 into the right side of this equation,
yielding TFRs of 5.55 and 5.19.
This procedure cannot be used for all India and other major states because a
post-enumeration survey was conducted in NFHS-2 only in Uttar Pradesh and not in
other states. In the present analysis for all-India and other major states, the method￾ology used for Uttar Pradesh must therefore be modified, as explained in the next
section.
FERTILITY ESTIMATES FOR ALL INDIA
Table 1 shows the discrepancies in TFR estimates for all India that we wish to ex￾plain. The NFHS estimate of 3.39 for 1990–92 is from the NFHS-1 report for all
India (IIPS 1995), and the NFHS estimate of 2.85 for 1996–98 is from the NFHS-2
report for all India (IIPS and ORC Macro 2000). These estimates are derived by the
birth-history method. The estimates from the SRS are simple 3-year averages of
values of the TFR published by the SRS for single calendar years. The SRS TFRs are
one-third to one-half child higher than the NFHS TFRs. The discrepancy is greater
for NFHS-2 (1996–98) than for NFHS-1 (1990–92). The two NFHS surveys indicate
a larger absolute decline and percentage decline in the TFR than does the SRS.
The problem of displacement in NFHS-1 and NFHS-2
As a first step in understanding the source of the discrepancies between the two
surveys and the SRS, Figure 3 shows three overlapping trends in the TFR, derived
from the three data sources. In Figure 3, the trends from NFHS-1 and NFHS-2,
which are derived by the own-children method, each show sharp peaks in the 9th, 11th,
and 13th years before the survey, reflecting heaping of children’s ages on 8, 10, and 12
years, as explained earlier. The trend based on the SRS is much smoother because in
the SRS births are recorded in the year in which they occur so that children’s ages do
not enter into the calculation of the fertility estimates. The trends derived from the
two NFHS surveys show evidence of displacement of births out of the first three
years before each survey to earlier years, suggesting that displacement is a major
16
National Family Health Survey Subject Reports, No. 19
Figure 3 Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
India
Table 1 Comparison of total fertility rates (TFR) estimated from NFHS-1, NFHS-2,
and the Sample Registration System: India
Sources: IIPS (1995) for NFHS-1; IIPS and ORC Macro (2000) for NFHS-2; Registrar General, India (various
years) for the SRS.
Note: NFHS estimates for 1990–92 are birth-history estimates from NFHS-1, and NFHS estimates for 1996–98
are birth-history estimates from NFHS-2. The SRS values are simple averages of annual estimates published
by the SRS. Time periods are calendar years for the SRS and years before the survey for NFHS-1 and NFHS-2,
so time periods differ slightly for each data source.
NA: Not applicable.
1990–92 1996–98 (1996–98) – (1990–92) (1996–98)/(1990–92)
NFHS 3.39 2.85 -0.54 0.84
SRS 3.68 3.32 -0.35 0.90
NFHS/SRS 0.92 0.86 NA NA
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
17
National Family Health Survey Subject Reports, No. 19
part of the explanation of why the TFR estimates from the two NFHS surveys are
lower than the TFR estimates from the SRS for the 3-year period preceding each
survey in Table 1. Figure 3 also suggests that displacement out of the 3-year period
preceding the survey may be worse in NFHS-2 than in NFHS-1, a possibility that was
discussed earlier in connection with the large block of questions asked of children
below a cut-off age, which was younger in NFHS-2 than in NFHS-1.
Table 2 demonstrates displacement in another way. The table shows fertility
estimates for the three 5-year periods and for the entire 15-year period before NFHS￾1 and NFHS-2. Estimates for the 15-year periods are shown in boldface. In the case
of the SRS, the table shows estimates for four 5-year periods. The fertility estimates
from NFHS-1 and NFHS-2 are generated by the own-children method, while the
estimates for 5-year periods from the SRS are unweighted averages of annual esti￾mates.
Because of displacement of births out of the first five years before each survey
and a bunching of births in the second and third five years before each survey, the
within-survey trend of the TFR derived from each survey separately is more steeply
downward than the between-survey trend based on the boldfaced 15-year-aggre￾gated TFRs. The average annual within-survey decline in the TFR between the third
and first 5-year periods before the survey is 0.17 child per year in NFHS-1 and 0.16
child per year in NFHS-2, compared with an average annual between-survey decline
of 0.10 child per year. The average annual decline in the SRS estimates between
1983–87 and 1993–97 was lower yet, at 0.08 child per year. Quite clearly, each of the
within-survey estimates of trend in the TFR, based on 5-year time periods, starts too
high and ends too low. The between-survey estimate of trend, based on two 15-year
Table 2 Trends in age-specific fertility rates (ASFRs) and total fertility rates (TFRs) from NFHS-1, NFHS-2,
and the SRS: India
Note: The NFHS-1 and NFHS-2 fertility estimates are derived by the own-children method. The SRS estimates for 5-year time periods are
simple averages of estimates for single calendar years published in annual SRS reports. For all India, the 15-year aggregations pertain
approximately to 1978–92 and 1984–98. Because the surveys were taken at different times in different states, these reference periods are
offset by a year in some states.
Fertility
measure
ASFRs
 15–19 173 158 110 143 176 152 104 140 90 89 81 60
 20–24 301 282 219 262 288 261 206 247 246 255 240 234
 25–29 255 235 182 220 222 195 159 188 231 223 198 190
 30–34 167 143 111 137 132 109 84 105 165 150 122 115
 35–39 95 79 58 75 67 53 38 51 99 86 71 58
 40–44 47 33 28 36 29 21 14 21 45 42 32 28
 45–49 18 12 12 14 12 7 6 8 20 19 12 9
TFR 5.28 4.71 3.60 4.44 4.62 3.99 3.05 3.81 4.48 4.31 3.78 3.47
NFHS-1 NFHS-2 SRS
1978–82 1983–87 1988–92 1978–92 1984–88 1978–82 1989–93 1994–98 1984–98 1983–87 1988–92 1993–97
18
National Family Health Survey Subject Reports, No. 19
time periods, looks much more reasonable and is much closer to the trend estimated
from the SRS. The even slower rate of TFR decline in the SRS could be due to
improvements in birth registration completeness over time (although we do not know
whether completeness actually did improve over this period) or to greater omission of
births in NFHS-2 than in NFHS-1.
Estimating underregistration of births in the SRS
Our strategy for obtaining a more accurate set of TFR estimates for 1990–92 and
1996–98 is to adjust the SRS estimates of the TFR in Table 1 for underregistration of
births by making use of the 15-year-aggregated estimates of the TFR derived by the
own-children method from NFHS-1 and NFHS-2. Fifteen-year-aggregated estimates
of the TFR are used because they minimize bias due to displacement of births to earlier
years. The first step is to estimate the extent of underregistration of births in the SRS.
To estimate underregistration of births in the SRS, we use the GFR rather than
the TFR. The GFR, it will be recalled, is defined as births to women age 15–49 during
the time period under consideration divided by woman-years of exposure during that
period. A 15-year-aggregated estimate of the GFR for 1978–92 is calculated by ap￾plying the own-children method to NFHS-1, and a roughly comparable estimate from
the SRS is calculated by taking a weighted average of annual estimates of the GFR
published by the SRS for 1978, 1979, ..., 1992, where the weights are based on the
estimated number of women age 15–49 in each calendar year during the 15-year
period.1 A correction factor for underregistration is then calculated by dividing the
NFHS-1 estimate of the GFR for 1978–92 by the comparable SRS estimate for the
same period. In the division, the denominators of the two GFRs cancel out inasmuch
as the proportion of females who are age 15–49 is close to the same in the NFHS
1
Weights must be used, because the own-children fertility estimates for the 15-year period are implicitly
weighted by virtue of the way they are calculated. In the calculation of 15-year weighted averages of
GFRs and TFRs from the SRS, weights were calculated as the estimated number of women age 15–49 at
the midpoint of a particular calendar year divided by the sum of the midpoint values over 15 calendar
years. Thus the weights sum to unity. The number of women age 15–49 at the midpoint of each calendar
year was estimated by linear interpolation of values from the 1971, 1981, 1991, and 2001 censuses. A
value for Assam is not available from the 1981 Census, requiring interpolation between the 1971, 1991,
and 2001 censuses. Preliminary estimates of total female population by state were available from the
2001 Census, but estimates of female population by age were not yet available. To obtain female
population age 15–49 for calendar years between 1991 and 1998, we first interpolated total female
population between 1991 and 2001 to estimate total female population at the time of NFHS-2 (taken as
1999.17, the average date of interview in NFHS-2). We then multiplied this value by the proportion of
females in the NFHS-2 household sample who were age 15–49 to estimate the number of women age 15–
49 in the general population in 1999.17. We then interpolated between this value and the number of
women age 15–49 in the 1991 Census to obtain estimates of the number of women age 15–49 at the
midpoints of intermediate calendar years.
19
National Family Health Survey Subject Reports, No. 19
surveys and the SRS, so that the correction factor specifies the ratio of NFHS-1
births to SRS births over the 15-year time period. This correction factor assumes that
no births were missed in NFHS-1. A correction factor for the period 1984–98 is
similarly calculated by dividing the NFHS-2 estimate of the GFR for this 15-year
period by the comparable SRS estimate of the GFR for the same period.
The correction factors calculated in this way for all India are 1.107 for the 15-
year period preceding NFHS-1 and 1.073 for the 15-year period preceding NFHS-2.
These values probably are underestimates, because some births are undoubtedly missed
in NFHS-1 and NFHS-2.
The correction factors of 1.107 and 1.073 are not the same as rates of
underregistration of births in the SRS, which may be calculated from a correction
factor k (1.107 or 1.073) as 1–1/k. The rates of underregistration so calculated are
9.6 percent for the period 1978–92 and 6.8 percent for the period 1984–98.
Correcting the SRS TFRs for underregistration of births
One might think that the GFR ratios could be used to adjust upward the SRS TFRs,
but actually 15-year-aggregated TFR ratios (calculated in the same way as the 15-
year-aggregated GFR ratios) are more appropriate, because TFRs are what are be￾ing corrected. The 15-year-aggregated TFR ratios for all India are 1.067 for NFHS-1
and 1.020 for NFHS-2, which are considerably lower than the corresponding GFR
ratios of 1.107 and 1.073. The reason why the TFR ratios are lower than the GFR
ratios will be explained later, when estimates of ASFRs are discussed. In what fol￾lows, we use the TFR ratios to correct the SRS estimates of the TFR.
Table 3 shows the NFHS and SRS raw estimates of the TFR, taken from Table
1, as well as two alternative sets of adjusted estimates of the TFR. The first set of
adjusted estimates adjusts the SRS estimate for 1990–92 upward by multiplying it by
the 15-year NFHS-1/SRS ratio of TFRs, and it adjusts the SRS estimate for 1996–98
upward by multiplying it by the 15-year NFHS-2/SRS ratio of TFRs. With these
adjustments, the TFR is estimated to have declined from 3.92 in 1990–92 to 3.39 in
1996–98. In the second set of adjusted estimates, the SRS TFRs for both 1990–92
and 1996–98 are adjusted upward by the higher of the two 15-year NFHS/SRS ratios
of TFRs, namely the NFHS-1/SRS ratio of TFRs. The rationale is that the extent of
birth underregistration in the SRS probably did not change much between 1978–92
and 1984–98 (periods that overlap by nine years), so that the higher of the two 15-
year NFHS/SRS ratios of TFRs probably does a better job of capturing missed births.
Thus the second set of adjusted estimates assumes that omission of births is greater
in NFHS-2 than in NFHS-1. With this adjustment, the TFR declines from 3.92 in
1990–92 to 3.55 in 1996–98.
20
National Family Health Survey Subject Reports, No. 19
In applying each 15-year-aggregated adjustment factor to a 3-year period at the
end of the 15-year period, we assume that birth registration completeness in the SRS
did not change during the 15-year period. If it improved during the 15-year period,
then the adjustment for the three most recent years of the period may be too large.
On the other hand, since NFHS-1 and NFHS-2 both missed some births, the adjust￾ment factors may be too small. These offsetting errors probably cancel each other to
some extent.
Another approach for estimating the TFR is based on the widely used “interna￾tional regression line” that relates the TFR to the contraceptive prevalence rate (CPR).
The CPR is calculated as the percentage of currently married women age 15–49
who are currently using any method of contraception. The international regression
line is based on data on national-level TFRs and CPRs from more than 90 countries
around the world (Ross and Frankenberg 1993). The equation of this line is:
TFR = 7.2931 – 0.0700 CPR (1)
Despite the fact that other variables known to affect fertility are omitted from
the right side of this equation, the equation fits the data rather well. R2
 = 0.88, indicat￾Table 3 Estimates of the total fertility rate (TFR) for 1990–92 and 1996–98: India
a
The values for 1990–92 are birth-history estimates from NFHS-1, and the values for 1996–98 are birth-history
estimates from NFHS-2 (published reports).
bThe 1990–92 SRS value is a simple average of published estimates of the TFR for 1990, 1991, and 1992, and
the 1996–98 SRS value is a simple average of published estimates of the TFR for 1996, 1997, and 1998.
c
The 1990–92 SRS TFR is adjusted upward by the 15-year NFHS-1/SRS ratio of TFRs, and the 1996–98 SRS
TFR is adjusted upward by the 15-year NFHS-2/SRS ratio of TFRs.
dThe higher of the two 15-year NFHS/SRS ratios of TFRs is used to adjust the SRS TFRs for both 1990–92
and 1996–98.
eThe equation of the international regression line is TFR = 7.2931 – 0.0700 CPR, where CPR denotes the
contraceptive prevalence rate (the percentage of currently married women age 15–49 who are currently
using any contraceptive method, including traditional methods).
Data source and method 1990–92 1996–98 Estimated decline
NFHS raw valuesa 3.39 2.85 0.54
SRS raw valuesb 3.68 3.32 0.35
SRS TFRs adjusted upward by the two
15-year NFHS/SRS ratios of TFRsc 3.92 3.39 0.53
SRS TFRs adjusted upward by the higher of
the two 15-year NFHS/SRS ratios of TFRsd 3.92 3.55 0.37
International regression line
estimates of TFRe 4.44 3.92 0.52
21
National Family Health Survey Subject Reports, No. 19
ing that the regression line explains 88 percent of the variation in the TFR.2 There is
still a good deal of scatter around the fitted line, however.
Results from NFHS-1 and NFHS-2 indicate that the CPR in all India increased
from 40.6 to 48.2 percent between the two surveys. Age misreporting has little or no
effect on these estimates, which we consider to be reasonably accurate. Substitution
of these values into equation (1) yields predicted TFRs of 4.44 for 1990–92 and 3.92
for 1996–98. These estimates are higher than any of the other estimates in Table 3,
but they are also less reliable, based as they are in part on a regression line that
reflects the experience of countries other than India. Moreover, because Indian women
rely heavily on female sterilization, which is a very effective method, and because
sterilization often occurs at a young age in India, the international regression line may
overestimate India’s TFR. Nevertheless, the estimates based on the international
regression line establish that even the higher of our two alternative sets of adjusted
estimates is plausible.
Overall, the analysis embodied in Table 3 shows that both the NFHS and the
SRS estimates of the TFR for 1990–92 and 1996–98 for all India are too low, al￾though by exactly how much is difficult to say. The true TFR for 1990–92 is probably
around 3.92, and the true TFR for 1996–98 is probably between 3.39 and 3.55.
Age-specific fertility rates (ASFRs)
Fertility rates calculated for 5-year age groups of women are much more susceptible
to bias from misreporting of women’s ages and displacement of births than are total
fertility rates. Figure 4 shows estimates of ASFRs for women in 5-year age groups
during the 3-year periods before NFHS-1 and NFHS-2 along with comparable SRS
estimates for the whole country. None of the estimates in the figure incorporate any
correction factors. The estimates from NFHS-1 and NFHS-2 are derived by the
own-children method. The estimates from the SRS are simple averages of published
ASFRs for single calendar years, averaged over three years.
Figure 4a suggests that the comparatively small fertility decline between 1990–
92 and 1996–98 estimated from the SRS stemmed from fertility reduction at all repro￾ductive ages, but especially 15–19. Inasmuch as the proportion currently using
contraception among currently married women age 15–19 increased only marginally
between 1990–92 and 1996–98 (from 7 to 8 percent, as estimated from the two
NFHS surveys), most of the fertility decline in this age group, if real, would have to
have occurred because of a decline in the proportion currently married at 15–19.
According to the SRS, the proportion currently married at 15–19 for all India de￾2
Other versions of the international regression line yield almost identical results to those shown here.
22
National Family Health Survey Subject Reports, No. 19
Figure 4a Age-specific fertility rates from the SRS for 1990–92 and 1996–98: India
Figure 4b Age-specific fertility rates from NFHS-1 (1990–92) and NFHS-2 (1996–98): India
0.000
0.050
0.100
0.150
0.200
0.250
0.300
15–19 20–24 25–29 30–34 35–39 40–44 45–49
Age
Age-specific fertility rate
SRS (1990–92) SRS (1996–98)
0.000
0.050
0.100
0.150
0.200
0.250
0.300
15–19 20–24 25–29 30–34 35–39 40–44 45–49
Age
Age-specific fertility rate
NFHS-1 (1990–92) NFHS-2 (1996–98)
23
National Family Health Survey Subject Reports, No. 19
creased from 31 to 25 percent between 1991 and 1997,3 not enough to explain the
approximately one-third decline in the ASFR at 15–19 shown in Figure 4a. In reality,
much of the apparent decline in ASFR(15–19) as estimated by the SRS is not real.
As will be explained shortly, it results from changes in the pattern of age misreporting
caused by the phasing in of a new SRS sample over the period 1993–95.
In contrast to the ASFR estimates from the SRS, the ASFR estimates for 1990–
92 from NFHS-1 and 1996–98 from NFHS-2, shown in Figure 4b, indicate virtually
no fertility decline at age 15–19 but substantial declines at older ages. The age pattern
of fertility decline shown in Figure 4b, based on NFHS-1 and NFHS-2, may be fairly
accurate, given that age misreporting, as indicated by Myers’ Index, is about the
same in the two surveys, so that the biasing effects of age misreporting on ASFRs
may mostly cancel out when calculating changes over time. Myers’ Index, however,
only covers ages 10–69. The extent of misreporting of children’s ages below 10,
which can also bias estimates of ASFRs from the two surveys, is not necessarily the
same in the two surveys.
In NFHS-1 and NFHS-2, children’s ages tend to be moved up for various rea￾sons. One reason, discussed earlier, is that some interviewers may wish to lighten
their workload by recording a child’s age above the cut-off age for a large block of
questions asked about young children. Another reason is simple rounding. For ex￾ample, a child age 2 years 9 months might be reported as age 3, in which case it is not
reported in the birth histories as a birth during the first three years before the survey.
Such exaggeration of children’s ages—equivalent to displacement of births to earlier
years—does not affect all ASFRs equally. The effect is unequal because displace￾ment of births to earlier years tends to shift the entire age curve of fertility to the left
(i.e., to younger ages of women), raising estimated fertility among younger women
and lowering estimated fertility among older women. This shift occurs because a
birth that is displaced further into the past is erroneously reported to have occurred
when the mother was younger. As a consequence of this mechanism, ASFRs below
the peak age of fertility are overestimated, and ASFRs above the peak age of fertility
are underestimated. At the same time, the net displacement of births out of the 3-year
period before the survey to earlier years tends to lower ASFRs at all ages. These two
effects on the estimated ASFR at 15–19 are offsetting to some extent, and the net
3
According to NFHS-1 and NFHS-2, the proportion currently married at 15–19 decreased from 38 to 33
percent between the two surveys. The higher estimates of the proportion currently married at 15–19 in
the two surveys than in the SRS stem from greater misreporting of women’s ages in the SRS, leading to
more moving up of married women from age 15–19 to age 20–24. (When calculating the proportion
currently married at 15–19 from NFHS-2, we first copied over the ages of ever-married women from the
individual file to the household file. This was already done in the NFHS-1 published reports, but it was
not done in the NFHS-2 published reports.)
24
National Family Health Survey Subject Reports, No. 19
effect is not clear. Above the peak age of fertility, the two effects are in the same
direction, so that the net effect on estimated ASFRs in this age range is clearly to bias
them downward.
Misreporting of women’s ages also tends to distort the age pattern of fertility. In
India as well as elsewhere in South Asia (except Sri Lanka), most women do not
know their age precisely, so that there is considerable potential for misreporting of
age. Ages tend to be shifted upward in the case of women who are married and who
have more than the typical number of children for their age (Retherford and Mirza
1982; Narasimhan et al. 1997; Retherford and Thapa 2000; Retherford et al. 2001).
Conversely, ages tend to be shifted downward in the case of never-married women
and married women with fewer than the typical number of children for their age. The
net effect of this age misreporting is to shift the age curve of fertility to the right,
increasing fertility among older women and decreasing fertility among younger women.
The extent to which the leftward shift associated with displacement of births and the
rightward shift associated with misreporting of women’s ages cancel each other is
not clear.
In the SRS there is no displacement of births because births are recorded in the
year in which they occur. There is every reason to believe, however, that women’s
ages are misreported to an even greater extent in the SRS than in NFHS-1 and
NFHS-2, because the two surveys devoted considerably more effort to collecting
accurate age data than does the SRS (Narasimhan et al. 1997). This means that the
ASFR curve in the SRS is undoubtedly shifted to the right. Because of this rightward
shift, the absolute levels of ASFRs estimated from the SRS are biased downward at
ages below the peak age of fertility and biased upward at ages above the peak age of
fertility. The downward bias at younger ages is aggravated by underregistration of
births, and the upward bias at older ages is offset to some extent by underregistration
of births. However, most underregistration of births in the SRS probably occurs at
younger ages of women in newly formed marital unions that have not yet been cap￾tured by the SRS, as well as among the substantial numbers of recently married
women who return to their parental home (usually outside the registration area) for
the birth of their first child.4 Thus the SRS estimate of fertility at 15–19 is biased
downward both by misreporting of women’s ages and by underregistration of births.
As noted by Retherford et al. (2001), the rightward shift of the age curve of
fertility in the SRS is reduced to some extent by an offsetting effect that stems from
4
In NFHS-1 in Uttar Pradesh, for example, the proportion of births that occurred in the home of the
mother’s parents during the four years immediately preceding the survey was 21 percent for women
below age 20, 10 percent for women age 20–34, and 2 percent for women age 35 and above (Narasimhan
et al. 1997). It is possible that a substantial proportion of births occurring in the mother’s parents’ home,
which occur mainly at age 15–19, are missed by the SRS.
25
National Family Health Survey Subject Reports, No. 19
the way that the SRS records ages. The SRS tabulates ASFRs by age at the begin￾ning of the calendar year. This means that a reported ASFR for a given 5-year age
group of women actually pertains to a 5-year age group that is on average six months
older. Thus the age groups that the SRS reports as 15–19, 20–24, ..., 45–49 are in
reality age groups 15.5–20.4, 20.5–24.4, ..., 45.5–50.4. This feature of the SRS has
the effect of shifting the age curve of fertility one-half year to the left (Narasimhan et
al. 1997). Retherford et al.’s (2001) results for Uttar Pradesh suggest that this left￾ward shift is more than offset by the rightward shift just described.
Figure 4b indicates that, for all India, ASFR(15–19) hardly changed between
NFHS-1 and NFHS-2. Given that the proportion currently married at 15–19 fell from
38 to 33 percent5 between the two surveys, the constancy of ASFR(15–19) suggests
that marital fertility at 15–19 may have increased slightly. This is possible, inasmuch
as age at marriage rose slightly between the two surveys (so that average fecundability
within the 15–19 age group increased) while contraceptive use at 15–19 hardly
changed, from 7 percent in NFHS-1 to 8 percent in NFHS-2.
Figure 5 depicts the same data as Figure 4 but arranged somewhat differently.
Figure 5a compares the age curves of fertility for 1990–92 derived from the SRS and
from NFHS-1, and Figure 5b compares the age curves of fertility for 1996–98 de￾rived from the SRS and from NFHS-2. In Figure 5a, for the period 1990–92, NFHS￾1 yields somewhat higher fertility than the SRS at age 15–19 and lower fertility than
the SRS at older ages. The discrepancies reflect errors in the SRS (the rightward
shift of the SRS-derived curve) as well as errors in NFHS-1 (a leftward shift and a
rightward shift, the net effect being unclear).
The discrepancies in Figure 5b (for 1996–98) are much larger than those in
Figure 5a (for 1990–92). Indeed, in Figure 5b the estimate of ASFR(15–19) is twice
as high in NFHS-2 as in the SRS. As explained below, the larger discrepancies in
Figure 5b probably occur because of greater age misreporting in 1996–98 than in
1990–92 in the SRS as result of the phasing in of a new SRS sample during 1993–95.
The period 1990–92 immediately precedes the sample changeover, whereas the pe￾riod 1996–98 immediately follows the changeover. The urban sample of urban blocks
was replaced in 1993, half of the rural sample of villages was replaced in 1994, and
the other half of the rural sample of villages was replaced in 1995.
The increase in age misreporting that occurs as a result of sample changeover
causes the age curve of fertility to be shifted more to the right. To see how the
phasing in of the new sample during 1993–95 creates this additional bias, consider
women age 15–19 in the SRS sample in 1993, just before the phasing in of the new
sample. Some of these women were recent brides from outside the registration area
5
See footnote 3.
26
National Family Health Survey Subject Reports, No. 19
Figure 5a Age-specific fertility rates from the SRS (1990–92) and NFHS-1 (1990–92): India
Figure 5b Age-specific fertility rates from the SRS (1996–98) and NFHS-2 (1996–98): India
0.000
0.050
0.100
0.150
0.200
0.250
0.300
15–19 20–24 25–29 30–34 35–39 40–44 45–49
Age
Age-specific fertility rate
SRS (1990–92) NFHS-1 (1990–92)
0.000
0.050
0.100
0.150
0.200
0.250
0.300
15–19 20–24 25–29 30–34 35–39 40–44 45–49
Age
Age-specific fertility rate
SRS (1996–98) NFHS-2 (1996–98)
27
National Family Health Survey Subject Reports, No. 19
(rural village or urban block) whose ages had to be estimated by the local part-time
SRS enumerator when they moved into the registration area. Among these women,
many who did not know their ages were undoubtedly shifted upward into the 20–24
age group by virtue of being married. Most other women age 15–19 in 1993, whether
single or married, had resided in the registration area since 1983–85 and were enu￾merated in the age group 5–9 (approximately) in the baseline survey that was taken
at the time of the 1983–85 sample changeover. In the 1983–85 baseline survey, no
one in the 5–9 age group was shifted upward or downward by virtue of marital status
because everyone was still single. In each subsequent year between 1984 and 1993,
the local enumerator simply increased the ages of these girls, as initially ascertained
in the baseline survey, by one year. By 1993, the single girls and some of the married
girls in the registration area constituted a large group of young women age 15–19
whose ages had been recorded earlier and could not be shifted.
During 1993–95, the situation changed with the phasing in of a new sample. By
1996, all women age 15–19 were new in the sample. Their ages had just been deter￾mined by the new baseline survey, and all of those who were married—and espe￾cially those who were both married and had children—were at risk of being shifted
upward in age. Since many (and in many areas of India, most) married women age
15–19 come from outside the village or urban block, however, the proportion at risk of
being rounded up to 20–24 by virtue of being married may not have increased much
during the 3-year changeover. Perhaps a more important factor is that the proportion
of single women age 20–24 who were at risk of being shifted downward into the 15–
19 age group increased greatly over the three years, the reason being that parents
anxious to marry off older single daughters tend to minimize their age. No doubt both
types of shifts occurred and help to explain the substantial decline in fertility at 15–19
and increases at 20–24 and older ages that occurred in the SRS between 1993 and
1996. It is also possible that underregistration of births temporarily increased with the
phasing in of a new sample.
According to the SRS, between 1993 and 1996 the age-specific proportion cur￾rently married at age 15–19, or ASPM(15–19), fell from 29 to 27 percent,6 ASFR(15–
19) fell from 70 to 55 births per 1,000 women per year, and the age-specific marital
fertility rate, or ASMFR(15–19), fell from 236 to 209 births per 1,000 currently mar￾ried women per year in the country as a whole. If our reasoning is correct, the
declines in ASFR(15–19) and ASMFR(15–19) in the SRS are partly spurious, espe￾cially the decline in ASMFR(15–19), which is probably entirely spurious, given that
contraceptive use at 15–19 increased by only one percentage point, from 7 to 8 per￾6
Proportions married from the SRS were calculated by dividing ASFR(15–19) from the SRS by ASMFR(15–
19) from the SRS.
28
National Family Health Survey Subject Reports, No. 19
cent, between NFHS-1 and NFHS-2. Contrary to expectation, however, ASPM(15–
19) did not decline much between 1993 and 1996 in the SRS, suggesting that down￾ward shifting of 20–24-year-old single women to 15–19 is not an important mechanism.
If the reasoning about the effect of phasing in a new SRS sample is correct, one
should see even steeper declines in SRS estimates of ASMFR(15–19) between 1993
and 1996 in India’s high-fertility states, where the proportion of women who do not
know their year of birth is even higher than in India as a whole. Retherford et al.
(2001) report relevant findings for the states of Uttar Pradesh, Madhya Pradesh,
Bihar, and Rajasthan. The SRS indicates that , between 1993 and 1996, ASMFR(15–
19) fell from 214 to 161 births per 1,000 currently married women per year in Uttar
Pradesh, from 256 to 228 in Madhya Pradesh, from 196 to 139 in Bihar, and from 212
to 136 in Rajasthan.7 Yet, between 1992–93 and 1998–99, NFHS-1 and NFHS-2
indicate that the proportion of currently married women age 15–19 who were cur￾rently using contraception changed from 3 to 5 percent in Uttar Pradesh, from 7 to 5
percent in Madhya Pradesh, from 3 to 2 percent in Bihar, and from 2 to 4 percent in
Rajasthan. The large but highly implausible declines in ASMFR(15–19) in all four
states provide additional evidence for the argument that the large decline in ASFR(15–
19) in the SRS for all India, as shown in Figures 4 and 5, is partly an artifact of the
phasing in of a new SRS sample during 1993–95.
As in the case of all India, however, ASPM(15–19) did not fall as much as
expected in these four states, except in Uttar Pradesh. Between 1993 and 1996,
ASPM(15–19) fell from 34 to 25 percent in Uttar Pradesh, from 41 to 39 percent in
Madhya Pradesh, from 39 to 37 percent in Bihar, and from 43 to 40 percent in
Rajasthan. The large declines in ASMFR(15–19) together with the rather small de￾clines in ASPM(15–19) suggest that the main effect of the SRS sample changeover,
in terms of its effect on fertility, was to increase the likelihood of shifting 15–19-year￾old married women with children into the 20–24 age group. As already mentioned, it
is also possible that underregistration of births temporarily increased with the phasing
in of a new sample.
To minimize bias from displacement of births and misreporting of women’s ages
in NFHS-1 and NFHS-2 and from the phasing in of a new SRS sample during 1993–
95, Figure 6 shows the age curves of fertility for 15-year periods instead of 3-year
periods before NFHS-1 and NFHS-2. Figure 6a compares SRS estimates of the age
curve of fertility for 1978–92 and 1984–98. In contrast with Figure 4a, Figure 6a
shows more fertility decline at the older ages than at the younger ages. Figure 6b,
which compares the NFHS-1 estimate of the age curve of fertility for 1978–92 with
7
Marital fertility changed little between 1991 and 1993. Almost all of the decline in ASMFR(15–19) that
occurred in these four states between 1991 and 1996 occurred between 1993 and 1996.
29
National Family Health Survey Subject Reports, No. 19
Figure 6a Age-specific fertility rates from the SRS for 1978–92 and 1984–98: India
Figure 6b Age-specific fertility rates from NFHS-1 (1978–92) and NFHS-2 (1984–98): India
0.000
0.050
0.100
0.150
0.200
0.250
0.300
15–19 20–24 25–29 30–34 35–39 40–44 45–49
Age
Age-specific fertility rate
SRS (1978–92) SRS (1984–98)
0.000
0.050
0.100
0.150
0.200
0.250
0.300
15–19 20–24 25–29 30–34 35–39 40–44 45–49
Age
Age-specific fertility rate
NFHS-1 (1978–92) NFHS-2 (1984–98)
30
National Family Health Survey Subject Reports, No. 19
the NFHS-2 estimate of the age curve of fertility for 1984–98, shows a rather similar
pattern, but with less fertility decline at 15–19 and more fertility decline at older ages.
Overall, the two NFHS surveys indicate a larger overall fertility decline than does the
SRS in Figure 6, again suggesting more omission of births in NFHS-2 than in NFHS￾1 or some improvement in birth registration completeness in the SRS between 1978–
92 and 1984–98.
Figure 7 depicts the same data as Figure 6 but arranged differently. Figure 7a
compares the age curves of fertility for 1978–92 derived from the SRS and from
NFHS-1, and Figure 7b compares the age curves of fertility for 1984–98 derived
from the SRS and from NFHS-2. The NFHS curves are higher, relative to the SRS
curves, in Figure 7 than in Figure 5, reflecting the smaller proportion of NFHS births
displaced out of the 15-year estimation period used in Figure 7 than the 3-year esti￾mation period used in Figure 5. The upward shift in Figure 7 is especially marked at
age 15–19. Because of the near-elimination of displacement (displacement shifts the
NFHS curves to the left), the NFHS curves are now more shifted to the right, at the
same time that their overall levels are shifted upward. As in the SRS, the rightward
shift in the two NFHS curves occurs because of a tendency to exaggerate the ages
of women who are married and who have more than the number of children than is
typical for their age and to reduce the ages of women who are single or who are
married with fewer than the number of children than is typical for their age. Fifteen￾year aggregations do not eliminate these latter sources of distortion.
The discrepancies in Figure 7 are especially large at age 15–19. In Figure 7b,
the NFHS-2 estimate of ASFR(15–19) is again almost twice as high as the SRS
estimate. The discrepancies in Figure 7b may be somewhat greater than in Figure 7a
because the two extremes of the period 1984–98 in Figure 7b both occur either in the
latter stages of an SRS sample changeover (1983–85) or right after an SRS sample
changeover (1993–95), resulting in more age misreporting and therefore more shift￾ing of the SRS curve of fertility to the right. The period 1978–92, in contrast, encom￾passes only one SRS sample changeover that occurred during 1983–85.
Figure 7 also clarifies why the 15-year GFR ratio is higher than the 15-year
TFR ratio, as seen earlier. The GFR is a weighted sum of ASFRs, where the weights
are the proportion of women age 15–49 falling in each 5-year age group within the
15–49 age range, whereas the TFR is an unweighted sum of ASFRs in which each
ASFR is weighted equally. In the case of the GFR, the weight is larger at 15–19 than
at older ages. At 15–19 the NFHS ASFR is higher than the SRS ASFR, whereas the
reverse is true at older ages, as seen in Figure 7. Because of the difference in weight￾ing, the 15-year GFR ratio (which is the ratio of the NFHS GFR to the SRS GFR) is
higher than the corresponding 15-year TFR ratio.
31
National Family Health Survey Subject Reports, No. 19
Figure 7a Age-specific fertility rates from the SRS (1978–92) and NFHS-1 (1978–92): India
Figure 7b Age-specific fertility rates from the SRS (1984–98) and NFHS-2 (1984–98): India
0.000
0.050
0.100
0.150
0.200
0.250
0.300
15–19 20–24 25–29 30–34 35–39 40–44 45–49
Age
Age-specific fertility rate
SRS (1978–92) NFHS-1 (1978–92)
0.000
0.050
0.100
0.150
0.200
0.250
0.300
15–19 20–24 25–29 30–34 35–39 40–44 45–49
Age
Age-specific fertility rate
SRS (1984–98) NFHS-2 (1984–98)
32
National Family Health Survey Subject Reports, No. 19
Although the age pattern of fertility is substantially distorted by age misreporting
in both NFHS surveys and in the SRS (although in different ways), the shifting of
ages and the displacement of births should not make nearly so much difference in
estimates of the TFR, because fertility is low at the lower and upper tails of the
reproductive age range, so that very few births are shifted outside this range. Thus
the shifted births get picked up one way or another when summing ASFRs over the entire
reproductive age range when calculating to the TFR. We therefore consider the 15-year
TFR estimates to be considerably more accurate than the 15-year ASFR estimates.
Despite some loose ends requiring further investigation, it is fairly clear that,
because age misreporting did not change much at the all-India level between NFHS￾1 and NFHS-2, 15-year-aggregated estimates from these two surveys probably pro￾vide a fairly accurate picture of how age-specific fertility changed in the country as a
whole between the two surveys, even though the estimates of ASFRs for specific 5-
year age groups in each survey are likely to be substantially biased by misreporting of
women’s ages and by displacement of births. The pattern of age-specific fertility
change over the same period derived from the SRS, on the other hand, is substantially
distorted by changes in the extent of age misreporting caused by the phasing in of a
new SRS sample over the period 1993–95.
FERTILITY ESTIMATES FOR MAJOR STATES
Our analysis of fertility estimates is necessarily briefer for major states than for all
India. Table 4 shows how values of Myers’ Index of age heaping changed for major
states between NFHS-1 and NFHS-2, with and without ever-married women’s ages
copied over from the individual questionnaire to the household questionnaire. Copying
over ages reduces Myers’ Index considerably, because the ages of ever-married
women age 15–49 (13–49 in NFHS-1) were probed extensively in the individual
interview but not in the household interview. Moreover, the ever-married woman
herself reported her age in the individual interview, whereas in the household inter￾view the household head (or other household respondent in the absence of the head)
reported it on her behalf. As noted earlier, in all India Myers’ Index changed little
between the two surveys. The lack of change is to be expected, because there is little
reason to believe that knowledge of age among respondents changed much in the
approximately six years between the two surveys. In most individual states, however,
Myers’ Index differs between the two surveys. Large changes in Myers’ Index—as
in Uttar Pradesh (where Myers’ Index doubled) and Andhra Pradesh (where it de￾clined by half)—are due to differences between the two surveys in interviewer train￾ing and supervision relating to the recording of age in a context where most respondents
do not know their age accurately.
33
National Family Health Survey Subject Reports, No. 19
Table 4 Myers’ Index for females from NFHS-1 and NFHS-2: India and selected states
a
Ages 15–49 (13–49 for NFHS-1) for ever-married women are copied over from the individual (ever-married woman)
file to the household file. The ages of all other women are from the household file. Myers’ Index is calculated for all
women age 10–69, irrespective of marital status.
India or state
India 10.4 10.0 18.0 18.4
Andhra Pradesh 14.2 6.9 25.5 10.0
Assam 10.0 13.5 27.8 28.1
Bihar 6.3 5.4 11.8 11.4
Gujarat 11.1 6.6 17.3 10.5
Haryana 9.4 11.1 27.4 25.1
Himachal Pradesh 9.9 8.1 19.8 18.5
Karnataka 7.1 10.6 11.3 19.0
Kerala 8.6 7.9 9.7 11.1
Madhya Pradesh 16.1 9.7 24.4 30.9
Maharashtra 8.8 7.6 14.5 14.4
Orissa 11.9 8.0 25.0 20.6
Punjab 11.3 10.1 25.4 20.9
Rajasthan 24.3 25.3 31.4 27.4
Tamil Nadu 8.1 5.0 11.1 5.3
Uttar Pradesh 7.5 15.3 16.3 31.8
West Bengal 10.7 12.9 16.9 16.3
With ages copied over from With ages taken from
NFHS-1 NFHS-2 NFHS-1 NFHS-2
the individual questionnaire the household questionnaire a
It is also of interest to look at the proportions of women who know their year of
birth. If women with inconsistent age and year of birth are counted as not knowing
year of birth, the proportion knowing year of birth increases from 17 to 29 percent for
all India (Table 5). The near-doubling of this proportion in only six years seems im￾plausible, however. Although instructions on how to collect age data were the same in
both surveys and in all states, it appears that the survey organizations (which in most
states were not the same in the two surveys) did not always follow these instructions
to the letter. According to the instructions, an age had to be entered by the inter￾viewer, even if it required an educated guess. But this was not so with the earlier
question on year of birth, which was supposed to be coded “don’t know” if the woman
did not know her year of birth. It is clear that in some states, such as Andhra Pradesh,
where the proportion who know year of birth increased from 12 to 70 percent be￾tween the two surveys (from 13 to 100 percent, if one counts women with inconsis￾tent age and year of birth as knowing year of birth), interviewers in NFHS-2 calculated
year of birth from age and entered it when the woman could not initially provide a
year of birth. Rajasthan shows a similar pattern. Kerala, on the other hand, shows the
opposite pattern, where the proportion knowing year of birth declined between the
two surveys, from 75 to 50 percent, which is even more implausible. These inconsis￾tencies in responses to the question on year of birth probably do not affect the ulti-
34
National Family Health Survey Subject Reports, No. 19
mate reporting of age, but they do illustrate how crucial interviewer training and
supervision are in the collection of accurate information on year of birth.
Table 6 shows sex ratios at birth, which vary greatly by state. The boldfaced
sex ratios pertaining to the 15-year period before each of the two NFHS surveys are
of particular interest, because they minimize the distorting effect of sex differentials
in displacement of births between successive 5-year periods before each survey. In
NFHS-1, these 15-year sex ratios at birth range from 1.02 in Andhra Pradesh and
Tamil Nadu to 1.15 in Punjab. In NFHS-2, they range from 1.04 in Tamil Nadu to 1.20
in Punjab. Sex ratios increased between the two surveys in every state except
Karnataka, Madhya Pradesh, Orissa, and Rajasthan. The declines in Karnataka,
Madhya Pradesh, and Rajasthan may reflect greater omission of births in NFHS-1
than in NFHS-2 in these states. In most states the sex ratio at birth did not increase or
decrease much between the two surveys. However, it increased from 1.09 to 1.14 in
Haryana, from 1.04 to 1.10 in Maharashtra, and from 1.15 to 1.20 in Punjab, perhaps
reflecting increases in sex-selective abortion in these states. The sex ratio at birth
increased from 1.01 to 1.06 in West Bengal, but the value of 1.06 is still close to the
expected value of about 1.05 and is not abnormally high.
The state-level sex ratios at birth from the two NFHS surveys suggest that the
rate of omission of births may differ between the two surveys, and that the extent of
Table 5 Percentage of ever-married women of reproductive age who know their year
of birth in NFHS-1 and NFHS-2: India and selected states
India or state NFHS-1 NFHS-2 NFHS-1 NFHS-2
India 18.3 33.9 17.1 29.4
Andhra Pradesh 12.9 99.8 12.3 69.9
Assam 6.6 13.8 6.4 11.3
Bihar 8.8 7.3 8.5 6.8
Gujarat 21.5 27.3 20.8 25.7
Haryana 18.1 17.6 17.8 17.2
Himachal Pradesh 22.7 25.5 22.0 25.4
Karnataka 17.1 22.5 16.6 22.3
Kerala 80.4 50.9 75.4 49.5
Madhya Pradesh 14.6 13.0 12.5 12.7
Maharashtra 25.4 36.1 23.2 32.3
Orissa 6.7 11.0 6.6 11.0
Punjab 20.2 26.5 18.5 24.5
Rajasthan 23.6 99.2 20.7 85.4
Tamil Nadu 31.9 39.6 29.8 37.9
Uttar Pradesh 7.9 8.5 7.6 8.0
West Bengal 10.1 35.6 10.0 33.5
Women with inconsistent 
age and year of birth counted 
as knowing year of birth
Women with inconsistent
age and year of birth counted
as not knowing year of birth
35
National Family Health Survey Subject Reports, No. 19
Table 6 Sex ratio at birth from NFHS-1, NFHS-2, and the SRS: India and selected states
Note: The sex ratio at birth is defined as the ratio of male births to female births. For all India, the 15-year aggregations pertain
approximately to 1978–92 and 1984–98. Because the surveys were taken at different times in different states, these reference
periods are offset by a year in some states.
SRS
India or state 1978–82 1983–87 1988–92 1978–92 1984–88 1989–93 1994–98 1984–98 1981–90
India 1.06 1.07 1.05 1.06 1.09 1.08 1.07 1.08 1.10
Andhra Pradesh 1.05 1.00 1.04 1.02 1.03 1.14 1.02 1.07 1.05
Assam 1.08 1.14 0.98 1.07 1.05 1.08 1.11 1.08 1.06
Bihar 1.08 1.02 1.03 1.04 1.02 1.06 1.06 1.05 1.12
Gujarat 1.13 1.03 1.02 1.06 1.09 1.05 1.06 1.07 1.11
Haryana 1.05 1.10 1.12 1.09 1.08 1.19 1.17 1.14 1.15
Himachal Pradesh 1.09 1.06 1.09 1.08 0.98 1.13 1.16 1.09 NA
Karnataka 1.02 1.08 1.06 1.06 1.06 1.06 1.04 1.05 1.07
Kerala 0.96 1.13 0.99 1.03 0.97 1.10 1.11 1.05 1.06
Madhya Pradesh 1.04 1.11 1.08 1.08 1.05 1.08 1.06 1.06 1.08
Maharashtra 1.03 1.04 1.07 1.04 1.17 1.02 1.10 1.10 1.09
Orissa 1.06 1.01 1.11 1.06 1.11 1.02 1.05 1.06 1.06
Punjab 1.10 1.19 1.16 1.15 1.20 1.23 1.16 1.20 1.13
Rajasthan 1.14 1.14 1.11 1.13 1.14 1.10 1.08 1.10 1.14
Tamil Nadu 0.99 1.11 0.96 1.02 1.04 1.04 1.05 1.04 1.05
Uttar Pradesh 1.08 1.10 1.05 1.08 1.16 1.10 1.04 1.09 1.12
West Bengal 1.02 0.96 1.05 1.01 1.08 1.02 1.08 1.06 1.06
NFHS-1 NFHS-2
differential omission varies by state. This is not surprising, given the number of different
survey organizations involved in data collection in the two surveys. Because of the growth of
sex-selective abortion between the two surveys, however, the trend in the sex ratio at birth
between the two surveys is not necessarily a good indicator of the trend in omissions.
In the SRS for the period 1981–90 (the only period for which a sex ratio at birth is
available from the SRS), the sex ratio at birth ranges from 1.05 in Andhra Pradesh and
Tamil Nadu to 1.15 in Haryana. During this earlier period there was little or no sex￾selective abortion, so that the high values of 1.11 in Gujarat, 1.12 in Bihar and Uttar
Pradesh, 1.13 in Punjab, 1.14 in Rajasthan, and 1.15 in Haryana probably reflect
underregistration of births that is greater for females than for males, reflecting the strong
preference for sons in these states. (There may also have been some sex-selective
infanticide that was undetected by the SRS, but it seems unlikely that it could account
for such high sex ratios at birth.) Overall, the high state-level sex ratios at birth from the
SRS suggest that underregistration of births in the SRS is substantial, that it varies con￾siderably from state to state, and that in most states it exceeds the rate of omission of
births in NFHS-1 and NFHS-2.
As mentioned earlier, we do not have any data on sex ratios at birth at more than
one time point that would indicate how the extent of underregistration of births in the
SRS has changed over time. However, several earlier studies based on SRS evaluation
36
National Family Health Survey Subject Reports, No. 19
checks of subsamples of the SRS sample units indicate that birth registration in the
SRS has improved over time. One study, based on the 1972 Fertility Survey con￾ducted in a 25-percent subsample of the SRS, concluded that the SRS underregistered
births in 1972 by about 8 percent (Mishra 1988; see also Registrar General, India
1976; 1983). Two subsequent studies concluded that the SRS underregistered births
by 3.2 percent in 1980–81 and by 1.8 percent in 1985 (Registrar General, India 1984;
1988). The analysis in the present report suggests much higher levels of
underregistration of births.
Table 7 shows state-level estimates of within-survey and between-survey trends
in the TFR derived from the two NFHS surveys, as well as trends estimated by the
SRS. In all states the within-survey declines in the TFR based on the trend over three
5-year time periods are steeper than the between-survey declines based on the trend
over two 15-year time periods, reflecting displacement of births backward in time
when 5-year periods from a single survey are used. In all states except Kerala and
Tamil Nadu, the SRS estimate of the TFR for 1993–97 is substantially higher than the
NFHS-2 estimate for 1994–98, reflecting displacement of births in NFHS-2 from
1994–98 to earlier years as well as some real decline in fertility between 1993–97 and
1994–98. The results for Kerala and Tamil Nadu suggest substantial underregistration
of births in the SRS in these two states.
Table 7 Unadjusted estimates of trend in the TFR from NFHS-1, NFHS-2, and the SRS: India and selected states
Note: For all India, the 15-year aggregations pertain approximately to 1978–92 and 1984–98. Because the surveys were taken at different times
in different states, these reference periods are offset by a year in some states. In the SRS in Bihar and West Bengal, TFRs are not available
for 1977, 1978, 1979, and 1980. In each of these two states, the 1981 SRS TFRs were used for 1977, 1978, 1979, and 1980. In the SRS in
Himachal Pradesh, the estimate of the TFR is not available for 1990. In this case, the average of the 1989 and 1991 SRS TFRs was used for
1990. In cases where an SRS TFR for 1999 was needed, the SRS TFR for 1998 was used for 1999.
India or state 1978–82 1983–87 1988–92 1978–92 1984–88 1989–93 1994–98 1984–98 1978–82 1983–87 1988–92 1993–97
India 5.28 4.71 3.60 4.44 4.62 3.99 3.05 3.81 4.48 4.31 3.78 3.47
Andhra Pradesh 4.26 3.91 2.79 3.57 3.70 3.23 2.37 3.04 4.00 3.82 3.04 2.62
Assam 5.91 5.64 3.82 4.99 5.05 4.16 2.68 3.81 4.12 4.15 3.51 3.40
Bihar 6.11 5.72 4.17 5.22 5.90 5.54 3.81 4.96 5.65 5.48 4.86 4.50
Gujarat 4.79 4.05 3.04 3.86 3.83 3.30 2.78 3.25 4.52 3.89 3.34 3.10
Haryana 5.61 4.76 4.25 4.79 5.00 4.08 3.11 3.97 5.00 4.61 4.01 3.61
Himachal Pradesh 4.57 3.79 3.20 3.78 4.04 3.09 2.42 3.11 3.81 3.75 3.24 2.65
Karnataka 4.72 4.14 3.40 4.00 4.06 3.13 2.39 3.10 3.62 3.60 3.19 2.70
Kerala 3.50 2.56 2.19 2.68 2.77 2.06 2.02 2.26 2.93 2.36 1.89 1.78
Madhya Pradesh 5.98 5.43 3.96 5.00 5.60 5.03 3.66 4.66 4.79 5.06 4.62 4.13
Maharashtra 4.47 3.81 3.05 3.70 4.02 3.05 2.54 3.13 3.64 3.74 3.21 2.86
Orissa 5.01 4.57 2.97 4.08 4.22 3.80 2.73 3.50 4.24 4.10 3.47 3.17
Punjab 4.80 4.01 2.87 3.79 4.01 3.26 2.52 3.19 3.95 3.61 3.22 2.86
Rajasthan 6.29 5.80 3.93 5.23 5.38 5.10 4.02 4.75 5.28 5.37 4.58 4.37
Tamil Nadu 3.88 3.29 2.47 3.15 3.03 2.54 2.27 2.58 3.45 2.95 2.36 2.08
Uttar Pradesh 6.74 6.03 5.00 5.84 6.27 5.66 4.32 5.30 5.82 5.63 5.22 5.01
West Bengal 4.76 4.25 3.12 3.95 4.05 3.38 2.43 3.20 4.19 3.83 3.26 2.80
NFHS-1 (own children) NFHS-2 (own children) SRS
37
National Family Health Survey Subject Reports, No. 19
Table 8 shows state-level 15-year-aggregated NFHS/SRS ratios of the GFR.
The NFHS-1/SRS ratios of the GFR range from 1.00 in Gujarat to 1.34 in Assam.
The NFHS-2/SRS ratios of the GFR range from 0.99 in Gujarat to 1.13 in Assam. As
already explained, underregistration of births is more accurately reflected in the GFR
ratios than in the TFR ratios. Rates of underregistration of births, derived from the
GFR ratios, are shown in Table 9. The states of Gujarat and Himachal Pradesh stand
out as states with very low GFR ratios and rates of underregistration in both surveys,
and the states of Assam, Karnataka, Kerala, Maharashtra, Punjab, Tamil Nadu, and
Uttar Pradesh are states with relatively high GFR ratios and rates of underregistration
in at least one of the two surveys.
The GFR ratios in Table 8 and the derived estimates of underregistration in
Table 9 assume that no births were missed in NFHS-1 and NFHS-2. Because the
two surveys undoubtedly did miss some births, the estimates of underregistration of
births in the SRS in Table 9 are probably underestimates. (In the case of Uttar Pradesh,
the one state where there was a post-enumeration check in NFHS-2, it is estimated
that 5 percent of births were missed, so that the estimate of underregistration of births for
Uttar Pradesh in Table 9 is definitely an underestimate. See Retherford et al. (2001).)
Table 8 Fifteen-year-aggregated NFHS-1/SRS and NFHS-2/SRS ratios of the GFR: India and selected states
Notes: For all India, the 15-year aggregations pertain approximately to 1978–92 and 1984–98. Because the surveys were taken at different
times in different states, these reference periods are offset by a year in some states. In the SRS in Bihar and West Bengal, GFRs are not
available for 1977, 1978, 1979, and 1980. In each of these two states, the 1981 SRS GFRs were used for 1977, 1978, 1979, and 1980 when
calculating 15-year-aggregated GFRs from the SRS. In the SRS in Himachal Pradesh, an estimate of the GFR is not available for 1990. In
this case, the average of the 1989 and 1991 SRS GFRs was used for 1990. In cases where an SRS GFR for 1999 was needed, the SRS
GFR for 1998 was used for 1999.
India or state NFHS-1 SRS NFHS-2 SRS
India 147.85 133.62 1.107 131.53 122.54 1.073
Andhra Pradesh 124.79 120.77 1.033 108.97 103.21 1.056
Assam 171.86 128.18 1.341 136.94 121.30 1.129
Bihar 172.89 161.88 1.068 166.58 149.56 1.114
Gujarat 130.37 129.79 1.004 112.23 113.36 0.990
Haryana 164.25 151.10 1.087 140.53 136.55 1.029
Himachal Pradesh 125.29 121.27 1.033 105.31 105.35 1.000
Karnataka 135.76 114.35 1.187 110.35 103.24 1.069
Kerala 89.05 81.79 1.089 75.39 68.36 1.103
Madhya Pradesh 168.48 161.16 1.045 161.68 148.62 1.088
Maharashtra 129.29 114.94 1.125 115.95 105.70 1.097
Orissa 135.36 125.24 1.081 121.91 114.19 1.068
Punjab 129.99 117.38 1.107 109.27 106.44 1.027
Rajasthan 167.46 160.50 1.043 160.21 150.23 1.066
Tamil Nadu 105.43 98.35 1.072 88.78 80.41 1.104
Uttar Pradesh 185.34 169.13 1.096 175.25 160.73 1.090
West Bengal 133.18 126.22 1.055 113.96 105.42 1.081
NFHS-1/SRS 
15-year GFRs for 1978–92 15-year GFRs for 1984–98
NFHS-2/SRS 
38
National Family Health Survey Subject Reports, No. 19
The pattern of variation by state in the NFHS/SRS ratios of the TFR in Table 10
is rather similar to the pattern for GFR ratios in Table 8, but the TFR ratios are usually
somewhat lower than the GFR ratios, for reasons explained earlier. In Himachal
Pradesh and Kerala, however, they are higher. The reason for this reversal is that,
because of previous fertility decline, the proportion of women age 15–49 who are 15–
19 is relatively low in Himachal Pradesh and Kerala, so that ASFR(15–19) is weighted
relatively little compared with ASFR(20–24) and ASFR(25–29) in the calculation of
the 15-year GFR ratio.
Table 11 shows state-level estimates of the TFR for 1990–92 and 1996–98
derived by various methods. The raw NFHS and SRS values are shown, as well as
two sets of adjusted SRS estimates that use 15-year-aggregated TFR ratios in differ￾ent ways, plus estimates derived from contraceptive prevalence rates using the inter￾national regression line discussed earlier. The contraceptive prevalence rates from
NFHS-1 and NFHS-2 are shown in Table 12. In Table 11, Gujarat is the only state
requiring no adjustment of the SRS TFRs, inasmuch as the TFR ratios for Gujarat are
either one or slightly less than one and then reset to one. In all other states, the SRS
estimates of the TFR have been adjusted upward to varying degrees, depending on
the size of the TFR ratios in Table 10. In most cases the TFR estimates derived using the
international regression line are higher than the other TFR estimates, but not always.
Table 9 Estimates of underregistration of births in the SRS: India and selected states
Note: Rates of underregistration of births in the SRS are calculated from the GFR ratios in Table 8. The rates of
underregistration for 1978–92 are calculated from the NFHS-1/SRS ratios, and the rates of underregistration for
1984–88 are cacluated from the NFHS-2/SRS ratios. Denoting a GFR ratio by k, the rate of underregistration is
calculated as 1–1/k.
India or state
India 0.09630 0.06838
Andhra Pradesh 0.03221 0.05292
Assam 0.25415 0.11423
Bihar 0.06369 0.10215
Gujarat 0.00442 -0.01006
Haryana 0.08004 0.02833
Himachal Pradesh 0.03205 -0.00047
Karnataka 0.15770 0.06445
Kerala 0.08149 0.09324
Madhya Pradesh 0.04349 0.08077
Maharashtra 0.11096 0.08837
Orissa 0.07470 0.06336
Punjab 0.09699 0.02585
Rajasthan 0.04161 0.06226
Tamil Nadu 0.06711 0.09426
Uttar Pradesh 0.08745 0.08283
West Bengal 0.05223 0.07496
Rate of 
underregistration for 
1978–92
Rate of 
underregistration for 
1984–98
39
National Family Health Survey Subject Reports, No. 19
It is of interest in this regard to compare TFR estimates in the last two columns
of Table 11, based on the international regression line, with the previous two columns,
in which the higher of the two TFR ratios (NFHS-1/SRS or NFHS-2/SRS) was used
to adjust the SRS estimates for both 1990–92 and 1996–98. The two sets of estimates
agree quite well for Assam, Himachal Pradesh, Madhya Pradesh, and West Bengal.
The agreement is also fairly close in the case of Punjab. The two sets of estimates
are quite different, however, for Andhra Pradesh, Bihar, Kerala, Orissa, Tamil Nadu,
and for the country as a whole. In these cases, the TFRs derived from the interna￾tional regression line are much higher than the TFRs obtained by using the higher of
the two 15-year TFR ratios to adjust the SRS estimates. The reasons for this state￾level variation are not clear. In the case of Assam, the results suggest high rates of
omission of births, especially in NFHS-2.
In Andhra Pradesh and Gujarat, the raw NFHS-1 and NFHS-2 estimates are
less than the corresponding SRS estimates for the 3-year period before each survey
(because of displacement of births in the two NFHS surveys but not in the SRS),
whereas the 15-year-aggregated TFR ratios are either exactly one or very close to
one, indicating close agreement between each of the two NFHS surveys and the SRS
when the effects of displacement are minimized.
Table 10 Fifteen-year-aggregated NFHS-1/SRS and NFHS-2/SRS ratios of the TFR: India and selected states
Notes: For all India, the 15-year aggregations pertain approximately to 1978–92 and 1984–98. Because the surveys were taken at different
times in different states, these reference periods are offset by a year in some states. In the SRS in Bihar and West Bengal, TFRs are not
available for 1977, 1978, 1979, and 1980. In each of these two states, the 1981 SRS TFRs were used for 1977, 1978, 1979, and 1980 when
calculating 15-year-aggregated TFRs from the SRS. In the SRS in Himachal Pradesh, an estimate of the TFR is not available for 1990. In
this case, the average of the 1989 and 1991 SRS TFRs was used for 1990. In cases where an SRS TFR for 1999 was needed, the SRS
TFR for 1998 was used for 1999.
India or state NFHS-1 SRS NFHS-1/SRS NFHS-2 SRS NFHS-2/SRS 
India 4.438 4.159 1.067 3.807 3.732 1.020
Andhra Pradesh 3.572 3.670 0.973 3.037 3.020 1.006
Assam 4.991 3.897 1.281 3.809 3.574 1.066
Bihar 5.224 5.278 0.990 4.960 4.810 1.031
Gujarat 3.864 3.864 1.000 3.252 3.335 0.975
Haryana 4.790 4.491 1.067 3.968 3.925 1.011
Himachal Pradesh 3.780 3.576 1.057 3.110 3.056 1.018
Karnataka 3.998 3.448 1.160 3.101 3.037 1.021
Kerala 2.681 2.358 1.137 2.257 1.950 1.157
Madhya Pradesh 4.997 4.885 1.023 4.660 4.475 1.041
Maharashtra 3.696 3.508 1.054 3.130 3.155 0.992
Orissa 4.076 3.904 1.044 3.503 3.444 1.017
Punjab 3.788 3.494 1.084 3.188 3.104 1.027
Rajasthan 5.233 5.041 1.038 4.751 4.611 1.030
Tamil Nadu 3.151 2.991 1.053 2.576 2.354 1.095
Uttar Pradesh 5.838 5.529 1.056 5.300 5.174 1.024
West Bengal 3.955 3.804 1.040 3.202 3.136 1.021
15-year TFRs for 1978–92 15-year TFRs for 1984–98
40
National Family Health Survey Subject Reports, No. 19
a
Derived by the birth-history method. Estimates for 1990–92 are from the published NFHS-1 report for all India, and estimates for 1996–98
are from the published NFHS-2 report for all India.
b
Simple average of published estimates of the TFR for 1990, 1991, and 1992.
c
Simple average of published estimates of the TFR for 1996, 1997, and 1998.
d
The 1990–92 SRS TFR is adjusted upward by the 15-year NFHS-1/SRS ratio of TFRs, and the 1996–98 SRS TFR is adjusted upward by the
15-year NFHS-2/SRS ratio of TFRs. However, if the adjustment factor is less than one, it is reset to one (i.e., no adjustment).
e
The higher of the two 15-year NFHS/SRS ratios of TFRs is used to adjust the SRS TFRs for both 1990–92 and 1996–98.
f
The equation of the international regression line is TFR = 7.2931 – 0.0700 CPR, where CPR denotes the contraceptive prevalence rate (the
percentage of currently married women age 15–49 who are currently using any contraceptive method, including traditional methods).
g
In an earlier study, when omission of births in NFHS-2 in Uttar Pradesh was taken into account (estimated from a post-enumeration survey
that was conducted in Uttar Pradesh but not in other states), the TFR for Uttar Pradesh was estimated to be 5.55 in 1990–92 and 5.19 in
1996–98 (Retherford et al. 2001). This earlier study did not, however, use population weights to calculate the 15-year SRS TFR that was used
in turn to calculate the 15-year NFHS-2/SRS ratio of TFRs. When weights are used, the corrected TFR estimates are still 5.55 for 1990–92
and 5.19 for 1996–98. Also in the earlier study, because SRS estimates of the TFR for 1998 were not yet available, the 1998 value of the
TFR was assumed to be the same as the 1997 value when calculating the average 15-year SRS TFR for 1984–98. SRS estimates for 1998
have since become available. When, in addition to using weights, the actual SRS TFR for 1998 is used in the calculations, the corrected TFR
estimates for Uttar Pradesh are 5.55 for 1990–92 and 5.18 for 1996–98. The estimates of 5.55 for 1990–92 and 5.18 for 1996–98, rather
than any of the values shown in the body of Table 11 for Uttar Pradesh, are our best estimates for Uttar Pradesh.
Table 11 Estimates of the total fertility rate (TFR) for 1990–92 and 1996–98 derived by various methods: India
and selected states
India or state
India 3.39 2.85 3.68 3.32 3.92 3.39 3.92 3.55 4.44 3.92
Andhra Pradesh 2.59 2.25 2.95 2.49 2.95 2.50 2.97 2.50 3.97 3.12
Assam 3.53 2.31 3.44 3.22 4.40 3.43 4.40 4.13 4.28 4.26
Bihar 4.00 3.49 4.61 4.38 4.61 4.51 4.75 4.51 5.67 5.58
Gujarat 2.99 2.72 3.23 3.00 3.23 3.00 3.23 3.00 3.84 3.16
Haryana 3.99 2.88 3.85 3.42 4.11 3.45 4.11 3.64 3.81 2.93
Himachal Pradesh 2.97 2.14 3.11 2.39 3.29 2.43 3.29 2.53 3.21 2.55
Karnataka 2.85 2.13 3.08 2.48 3.57 2.53 3.57 2.88 3.84 3.21
Kerala 2.00 1.96 1.81 1.83 2.05 2.12 2.09 2.12 2.86 2.83
Madhya Pradesh 3.90 3.31 4.58 3.98 4.68 4.14 4.76 4.14 4.72 4.19
Maharashtra 2.86 2.52 3.07 2.74 3.23 2.74 3.23 2.89 3.50 3.03
Orissa 2.92 2.46 3.32 3.03 3.46 3.08 3.46 3.17 4.75 4.02
Punjab 2.92 2.21 3.13 2.68 3.39 2.75 3.39 2.91 3.18 2.63
Rajasthan 3.63 3.77 4.56 4.17 4.73 4.30 4.73 4.33 5.06 4.47
Tamil Nadu 2.48 2.19 2.26 2.01 2.38 2.20 2.47 2.20 3.81 3.64
Uttar Pradesh 4.82 3.99 5.17 4.77 5.45g 4.88g 5.45g 5.03g 5.91 5.33
West Bengal 2.92 2.29 3.18 2.53 3.30 2.59 3.30 2.64 3.25 2.63
SRS TFRs adjusted
upward by the higher
of the two NFHS/SRS
ratios of TFRse NFHS raw valuesa SRS raw values
1990–92 1996–98 1990–92b 1990–92
SRS TFRs adjusted
upward by the two
1996–98c
15-year NFHS/SRS
ratios of TFRsd
from international
TFRs estimated
regression linef
1996–98 1996–98 1990–92 1996–98 1990–92
41
National Family Health Survey Subject Reports, No. 19
It is also of interest to compare Uttar Pradesh, where Myers’ Index doubled
between the two surveys, and Andhra Pradesh, where Myers’ Index was halved
between the two surveys. In Uttar Pradesh, because of greater displacement in the
second survey than in the first, the raw NFHS estimates show too great a fertility
decline, compared with the SRS estimates. In Andhra Pradesh, because of less dis￾placement in the second survey than in the first, the raw NFHS estimates show too
small a fertility decline, compared with the SRS estimates. In each case, the SRS
estimates provide a more accurate estimate of change over the six years between the
two surveys, assuming that the rate of underregistration of births in the SRS did not
change much in the approximately six years between the two surveys. We do not
know, however, how much the phasing in of a new SRS sample during 1993–95
affected the rate of underregistration of births.
Figure 8 rounds out the picture by showing overlapping trends in the TFR, esti￾mated from NFHS-1, NFHS-2, and the SRS, without any adjustments. The three sets
of estimates are fairly consistent in the case of Andhra Pradesh, Gujarat, Haryana,
Himachal Pradesh, Tamil Nadu, and West Bengal. Discrepancies are greatest in Assam,
and other states are in between.
Table 12 Contraceptive prevalence rates from NFHS-1 and NFHS-2: India and
selected states
Note: The contraceptive prevalence rate (CPR) is defined as the percentage of currently married
women age 15–49 who are currently using any contraceptive method, includingtraditional methods.
More exact (i.e., less rounded) estimates of the CPR than shown here were used to estimate the
TFRs from the international regression line in Table 11.
India or state NFHS-1 NFHS-2
India 41 48
Andhra Pradesh 47 60
Assam 43 43
Bihar 23 25
Gujarat 49 59
Haryana 50 62
Himachal Pradesh 58 68
Karnataka 49 58
Kerala 63 64
Madhya Pradesh 37 44
Maharashtra 54 61
Orissa 36 47
Punjab 59 67
Rajasthan 32 40
Tamil Nadu 50 52
Uttar Pradesh 20 28
West Bengal 58 67
Contraceptive prevalence rate
42
National Family Health Survey Subject Reports, No. 19
Figure 8a Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Andhra Pradesh
Figure 8b Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Assam
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
43
National Family Health Survey Subject Reports, No. 19
Figure 8c Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Bihar
Figure 8d Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Gujarat
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
44
National Family Health Survey Subject Reports, No. 19
Figure 8e Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Haryana
Figure 8f Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Himachal Pradesh
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
45
National Family Health Survey Subject Reports, No. 19
Figure 8g Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Karnataka
Figure 8h Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Kerala
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
46
National Family Health Survey Subject Reports, No. 19
Figure 8i Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Madhya Pradesh
Figure 8j Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Maharashtra
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
47
National Family Health Survey Subject Reports, No. 19
Figure 8k Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Orissa
Figure 8l Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Punjab
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
48
National Family Health Survey Subject Reports, No. 19
Figure 8m Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Rajasthan
Figure 8n Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Tamil Nadu
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
49
National Family Health Survey Subject Reports, No. 19
Figure 8o Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
Uttar Pradesh
Figure 8p Overlapping trends in the TFR, estimated from NFHS-1, NFHS-2, and the SRS:
West Bengal
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
0
1
2
3
4
5
6
7
8
9
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
Year
TFR
SRS NFHS-1 (Own Children) NFHS-2 (Own Children)
50
National Family Health Survey Subject Reports, No. 19
CONCLUSION
The analysis indicates that, for all India, the true TFR for 1990–92 was probably
around 3.92, and the true TFR for 1996–98 was probably between 3.39 and 3.55,
somewhat higher than the raw SRS estimates and considerably higher than the raw
NFHS-1 and NFHS-2 estimates for the same time periods. Even these new esti￾mates may be a bit low, however, because we have no estimates of omitted births in
NFHS-1 and NFHS-2, except for Uttar Pradesh in NFHS-2, where a post-enumera￾tion survey indicated that about 5 percent of births were missed. The raw NFHS
estimates are too low primarily because of displacement of births out of the 3-year
period preceding each survey to earlier years, and the raw SRS estimates are too low
because of underregistration of births. The extent of displacement in NFHS-1 and NFHS￾2 and the extent of underregistration of births in the SRS vary considerably by state.
Because the extent of age misreporting did not change much between the two
surveys in India as a whole, our estimates of change in age-specific fertility rates in
all India, based on the own-children estimates of ASFRs from NFHS-1 and NFHS-2,
are probably fairly accurate, even though the estimated levels of the ASFRs are
biased. In most individual states, however, the extent of age misreporting changed
enough between the two surveys that we are unable to arrive at accurate estimates
of the age pattern of fertility change between the two surveys. In the case of the
SRS, the age pattern of fertility change is distorted both at the national level and the
state level by changes in the pattern of age misreporting caused by the phasing in of
a new SRS sample during the period 1993–95.
The analysis also illustrates that the standard strategy in demographic and health
surveys of providing TFR estimates for the 3-year period immediately preceding the
survey does not work very well in India, where there is widespread ignorance of age,
resulting in substantial displacement of births to earlier years and therefore to under￾estimation of the TFR for this 3-year period. The analysis also shows that application
of the own-children method of fertility estimation to the household samples of DHS
surveys can usefully supplement the usual birth-history approach and contribute to
deriving more accurate fertility estimates. For such analyses, the ages of ever-mar￾ried women in the individual sample should be copied over to the household sample.
This was done in NFHS-1 but not in NFHS-2, and it is generally not done in demo￾graphic and health surveys in other countries.
Attainment of the goal of more accurate fertility estimates from the SRS would
be greatly enhanced if the Sample Registration System would publish single-year
age-sex distributions of the SRS sample as well as sex ratios at birth on an annual
basis. Such information would be useful for demographic analyses of quality of data
and bias in fertility estimates derived from the SRS.
51
National Family Health Survey Subject Reports, No. 19
ACKNOWLEDGMENTS
We thank Fred Arnold for helpful comments, Gayle Yamashita for computer pro￾gramming assistance, Sally Dai for typesetting assistance, and O. P. Sharma for
assistance in printing and distribution. The United States Agency for International
Development, via a subcontract from ORC Macro to the East-West Center, provided
support for the preparation and publication of this report.
REFERENCES
Cho, L.-J., R. D. Retherford, and M. K. Choe. 1986. The Own-Children Method
of Fertility Estimation. An East-West Center Book. Honolulu: East-West
Center.
IIPS (International Institute for Population Sciences). 1995. National Family Health
Survey (MCH and Family Planning): India, 1992-93. Bombay: International
Institute for Population Sciences.
IIPS (International Institute for Population Sciences) and ORC Macro. 2000. Na￾tional Family Health Survey (NFHS-2): India. Mumbai: IIPS.
Keyfitz, N., and W. Flieger. 1968. World Population: An Analysis of Vital Data.
Chicago: University of Chicago Press.
Mishra, V. K. 1988. India’s stagnated birth rate: A matter of serious concern.
Unpublished paper. Delhi: Institute of Economic Growth.
Narasimhan, R. L., R. D. Retherford, V. Mishra, F. Arnold, and T. K. Roy. 1997.
Comparison of Fertility Estimates from India’s Sample Registration System
and National Family Health Survey. National Family Health Survey Subject
Reports No. 4. Mumbai: International Institute for Population Sciences. Hono￾lulu: East-West Center.
Registrar General, India. Various years. Sample Registration System annual reports.
New Delhi: Office of the Registrar General.
Registrar General, India. 1976. Fertility Differentials in India, 1972: Results of
the Fertility Survey in a Sub-sample of SRS (1972). New Delhi: Office of the
Registrar General.
Registrar General, India. 1983. Sample Registration System, 1976–78. New Delhi:
Office of the Registrar General.
Registrar General, India. 1984. Report on the Intensive Enquiry Conducted in a
Sub-sample of Sample Registration System Units (1980–81). Occasional
Paper No. 2 of 1983. New Delhi: Office of the Registrar General.
Registrar General, India. 1986. SRS Based Abridged Life-Tables 1976–80. Occa￾sional Paper No. 1 of 1985. New Delhi: Office of the Registrar General.
Registrar General, India. 1988. Report on the Intensive Enquiry Conducted in a
52
National Family Health Survey Subject Reports, No. 19
Sub-sample of Sample Registration System Units (1985). Occasional Paper
No. 1 of 1988. New Delhi: Office of the Registrar General.
Registrar General, India. 1990. SRS Based Abridged Life-Tables 1981–85. Occa￾sional Paper No. 1 of 1989. New Delhi: Office of the Registrar General.
Registrar General, India. 1995. SRS Based Abridged Life-Tables 1988-92. Occa￾sional Paper No. 4 of 1995. New Delhi: Office of the Registrar General.
Registrar General, India. 1998. SRS Based Abridged Life Tables 1990–94 and
1991–95. SRS Analytical Studies Report No. 1 of 1998. New Delhi: Office of
the Registrar General.
Registrar General, India. 1996. Report of the Technical Group on Population Pro￾jections. New Delhi: Office of the Registrar General.
Registrar General, India. 1999. Compendium of India’s Fertility and Mortality
Indicators, 1971–1997: Based on the Sample Registration System (SRS).
New Delhi: Office of the Registrar General.
Retherford, R. D., and M. Mirza. 1982. Evidence of age exaggeration in demographic
estimates for Pakistan. Population Studies 36: 257–70.
Retherford, R. D., V. K. Mishra, and G. Prakasam. 2001. How Much Has Fertility
Declined in Uttar Pradesh? National Family Health Survey Subject Reports,
No. 17. Mumbai: International Institute for Population Sciences; Honolulu: East￾West Center.
Retherford, R. D., and S. Thapa. 2000. The trend of fertility in Nepal, 1961-95.
Genus 55: 61–98.
Shryock, H. S., and J. S. Siegel. 1973. The Methods and Materials of Demogra￾phy. 2 volumes. Washington, D. C.: U. S. Bureau of the Census.