Reverse Survival Method

Concept and Need

• To estimate fertility levels for last 5 years or 10 years
prior to census year then,
– We need to add those children who died during the
past five years (or 10 years) in order to get total births
in this period
– In other words, children enumerated in the age group
5-9 years in the census need to be reverse survived
using appropriate survival ratio (adjustment mortality
i.e. adding those who died)
• This method needs very minimal data; it is extremely
useful to obtain fertility levels at the lower
administrative units such as
– Districts/blocks where usually direct estimates are not
available
 Assumptions and Data Requirement
Assumptions
– There is not much child or female migration in population
– Level and age pattern of mortality in early childhood are
known
– Child population is free from age reporting errors
Data Requirement
– Population by sex for ages 0-4, 5-9 and 0-6 on the
enumerated Date and Total Population
– Annual exponential growth rate of population for
reference period preceding the date of enumeration
– Life table survival ratios (nLx values) for ages 0-4 and 5-9. If
life table for area under study is not available then q(2),
q(3) and q(5) can be estimated using CEB and CS data
 Methodology
Step I: Obtain combined survival probability from birth
to age group 0-4 by sex from a suitable life
table
L0 5 L0
S B 0  4  5

5 * l0 5
Where,
L is life table population aged 0-4
5 o
lo is the initial population in the life table=1

Continued…
 Methodology
Step II: Estimate average annual births during last five
years preceding the census date as follows:
t
5 P0
t

B(t 5,t )  1 4 P0
* 
5 S B 0  4 L / l
5 0 0

Where,
4 0 is population aged 0-4 years at census year
Pt

‘t’

Continued…
 Methodology
Step III: Obtain birth rate during the last five years
preceding census as follows:
B (t  5, t )
BR(t  5, t )  *1000
P (t  2.5)
Now, we need to know the mid-year population P (t-2.5)
i.e. (population 2.5 years prior to the census), which is
obtained as
Pt - 2.5 = Pt * e – 2.5 * r
Where,
Pt = Total population in census year ‘t’
r = inter-censal average exponential annual growth rate

Continued…
 Methodology
t
P
4 0
BR ( t  5, t )   2 .5 * r
*1000
5 L0 * Pt * e

t 2.5 * r
P 0 e
BR ( t 5, t )  4 * *1000
Pt 5 L0

t 2.5 * r
p (0  4)*e
BR ( t 5, t ) 
5 L0

Where, pt(0-4) is the proportion of population in the age

group 0-4 at time ‘t’
 Limitations of Method
• Under/over enumeration of children distorts results
• The level and pattern of mortality selected to obtain
survival ratio may not be available and is often
assumed
• Validity of estimates obviously depends on how close
the assumptions are to the reality
Regression Methods Based on
Birth Order Statistics
 Concept
• Most of the sources that collect total births also
collect information on birth order. In other words,
total births are distributed by order of birth
• Empirically, it has been observed that there is very
high correlation between fertility rates (CBR,TFR) and
birth order distribution. Proportion of births of First
and Second Order is negatively related to TFR
 Relation between PFSB and TFR
6

5
Total Fertility Rate

1
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Proportion of First and Second Order Births

 Curve Fitting
6

5
Total Fertility Rate

Observed
1
Linear

0 Exponential
.2 .3 .4 .5 .6 .7 .8 .9
Proportion of First and Second Order Births
 Data Requirement and Method
Data Requirement
(i) Total Births
(ii) First and Second Order Births (or 3+)

Method

TFR = exp(α + β* PFSB + e)

where, ‘α’ is a constant and ‘β’ is the regression coefficient, ‘PFSB’ is the
percentage of the first and the second order births to total births, and ‘e’
is the error term.
Using information from three consecutive censuses from 1981 to 2001,
an estimate of TFR can be made directly, using the following
formula:
TFR=12.265*exp (-0.0221*PFSB)
 Rele Technique: Description

 Given the information on the number of children 0-4 and 5-9, the
number of women in reproductive ages 15-54, and an estimate of
the life expectancy at birth, Rele technique estimates the GRR and
TFR of a population for one or two 5-year periods prior to the
census date
 Rele Technique: Data Required
 Child population (both sexes combined) ages 0-4 years and 5-9
years

 Female population ages 15-49 years and 20-54 years

 Life expectancy at birth (e0) for both sexes combined

Note: The values of e0 used for the purpose of generating the

estimates should refer to the corresponding periods of the
CWRs
 Rele Technique: Assumptions
 The population is stable or quasi-stable

 There is no migration

 Completeness of enumeration is the same in all the required age

groups

 Age reporting of children is correct within each 5-year age group

 Age misreporting of women within the groups of childbearing ages

does not affect the total number of women in those ages

 The mean length of generation is 28-29 years.

Rele Technique: Procedure: Estimation of GRR from CWR
 In 1967 Rele proved that in both stable and non-stable populations at a given
level of mortality there is an almost linear relationship between the Child-
woman ratio (CWR) and the gross reproduction rate (GRR).

 Since the relationship is linear, Rele said it can be expressed as GRR = a n + bn

(CWR). Or in specific:
C (0  4)
GRR  a n  bn 
W (15  49)
- - - (1)

C (5  9)
GRR  a n  bn 
W (20  54)
   (2)

Eqn. (1) provides estimates of GRR that refer to the period up to 5 years before the census
or survey, its midpoint being 2.5 years before the enumeration
Eqn. (2) provides estimates of GRR that refer to the period 5 - 10 years before the census
or survey, its midpoint being 7.5 years before the enumeration
Rele Technique: Procedure: Estimation of GRR from CWR
In the above equations the coefficients an and bn were derived by selecting a
sample of 12 stable populations at each of six levels of mortality

In the above equation,

n = level of mortality (e00),

GRR = Gross Reproduction Rate,

C = Children, W = Woman, CWR=Child-Woman Ratio

a and b are published regression coefficients at fixed values of e 00 (readily

available)

Note: Rele’s 1967 discovery of a direct link between GRR and CWR is the
outcome of an earlier observation by of a theoretical link between CWR and
a fertility rate based on an observation made by Lotka in 1936 that in a
stable population the NRR can be approximated by the “replacement Index”
which is the CWR divided by the CWR of the life table of the same
population.
Rele Technique: Procedure: Estimation of TFR from CWR

• The total fertility rate (TFR) which equals approximately 2.05 times the
GRR can be expressed as:

TFR = 2.05 X GRR = 2.05 X [an +bn CWR]

Table 1 (in the next slide) provides a and b values for estimating TFR
straight away as:

TFR = an +bn CWR

from the given CWRs and the life expectancy at birth of both sexes
combined.

Note: Rele provided coefficients for four kinds of CWR, but Table 1
provides coefficients here for two CWRs only.
 Rele Technique: Procedure: Estimation of TFR from CWR
(Cont.)

Table1 : Coefficients for the estimation of TFR from CWR

CWR Regression Expectation of life at birth (e0)
Coefficient
20 30 40 50 60 70

C(0-4) a 0.1121 0.0582 0.0264 -0.0121 -0.0373 -0.0633

W(15-49)
b 9.7744 8.8751 8.3265 7.9107 7.5087 7.1399
C(5-9) a 0.0502 0.0217 0.0043 -0.0226 -0.0463 -0.0707
W(20-54)
b 11.2158 9.5116 8.6637 8.0934 7.5879 7.1383
(Source: Given by Rele (1989) deriving from the coefficients for the estimation of GRR from CWR. See Rele (1976,P.22)

Note: The TFR of a population whose level of mortality is in between

two of the six levels can be approximated closely by computing the
TFRs at the adjacent levels and then interpolating the results for the
exact level.
 Rele Technique: Advantages
(1) Requires little information (only age distribution of population and a
LEB)

(2) Is less sensitive to errors in the data on life expectancy at birth

(3) Is quite accurate even in populations which are far from stable
 Rele Technique: Limitations

(1) Is sensitive to differential under enumeration of children and women

(2) It does not work well where the mean age at childbearing is high
 P/F Ratio Technique: Description
 Originally developed by William Brass (Brass et al., 1968)

 Produces a factor for adjusting reported age-specific fertility rates

(based on vital registration or births in the 12 months prior to a census
or survey) to the “actual” level of fertility

 The adjustment factor is determined by comparing data on children ever

born (CEB), by age of women, with a set of cumulated age-specific
fertility rates.
 P/F Ratio Technique: Data required
1) Average number of children ever born alive, per woman, by 5-year
age groups (15 to 19 years to 45-49 years),
[as collected in a census or survey]

2) Age-specific fertility rates for the same 5-year age groups of women.
[These rates may be based on reported births in the 12 months prior
to the census or on independent registration for a time period close to
the date of the census or survey]

3) Information to determine whether the age of women reported births in

the last 12 months was reported as of the date of the childbirth or as
of the date of the census or survey
 P/F Ratio Technique: Assumptions
1) The reporting of the average number of children ever born is complete
(at least for younger women, under 30 years or 35 years of age), and
represents the level of cumulative fertility up to these ages

2) The completeness of reporting of births used to estimate the age-

specific fertility rates is the same for all age groups of women

3) The pattern and level of fertility have note changed in the recent past
(15 to 20 years period to the census or survey)
 P/F Ratio Technique: Procedure

 The basis of this technique is a comparison of the average number of

CEB by age of women (P) with cumulative ASFRs from registers or
from reported births during the last year (F)

 This comparison requires some adjustments

 After such adjustments, the ratios P/F produce a factor for adjusting the
ASFRs to the “actual” level of fertility
 P/F Ratio Technique: Procedure (Cont.)

 Why the adjustments ?

 P pertains to the mean age of each age interval (17.5 years,

22.5 years,…,47.5 years)

 F pertain to the end of each age interval (20years, 25 years, …,

50 years)
 There are 2.5 years difference between them

 This difference would produce higher values for F than for P

 P/F Ratio Technique: Procedure (Cont.)
 How to adjust the data for this age discrepancy?
 Three methods have been developed

 Brass, the author of the technique, simulated the P and F by using a

third degree polynomial

 Trussell and Coale developed two alternatives:

• One uses second degree polynomials for each three consecutive

observations
• Other uses the least square technique and model ASFRs to
obtain a set of coefficients for the adjustment
 This difference would produce higher values for F than for P
 P/F Ratio Technique: Procedure (Cont.)

 Note: Some times an additional adjustment is

required
 Once the adjustments are made, the ratios of P to F can be taken and
applied to the ASFRs to proportionally increase or decrease them to the
“proper” level of fertility represented by the average number of children
per woman at young adult ages
 The ages most frequently used are 20 to 24 years and 25 to 29

years
 Occasionally 30 to 34 years

 Note: If the data is accurate and fertility has been decreasing in recent
years, the P/F ratios may show a trend that rises with the age group of
women
 P/F Ratio Technique: Computational Steps
 Step1: Calculate the reported average parities
(Pi)
 Step2: Calculate the preliminary fertility
schedule f(i)
 Step3: Calculate the cumulated fertility schedule
Ø(i)
 Step4 : Estimate the current average parity equivalents for a period F(i)
 Step5: Calculate a fertility schedule for conventional five-year age
groups f+(i)
 Step 6: Calculate the adjusted period fertility schedule from f+(i):
Formula
 Step 7: Calculate Total Fertility (TF)
 Computational procedure for estimating TF

Step1: Calculate the reported average parities (Pi): Total

number of children ever born to women in age
group ‘i’ / Total number of women in that age group
Age Number of Children Index Average parity
group women ever born ‘i’ per woman P(i)
(1) (2) (3) (4) (5) = (3)/(2)
15-19 3014706 1160919 1 0.385
20-24 2653155 4901382 2 1.847
25-29 2607009 9085852 3 3.485
30-34 2015663 9910256 4 4.917
35-39 1771680 10384001 5 5.861
40-44 1479575 9164329 6 6.194
45-49 1135129 6905673 7 6.084
 Computational procedure for estimating TF

Step2: Calculate the preliminary fertility schedule f(i): Number of births

occurring to women in age group ‘i’ during the year preceding the
interview / Total number of women (whether childless or not, ever
married or not) in that age group
Age group Number of Births in past Index Period fertility
women year ‘i’ rate f(i)
(1) (2) (3) (4) (5) = (3)/(2)
15-19 3014706 320406 1 0.1063
20-24 2653155 609269 2 0.2296
25-29 2607009 561494 3 0.2154
30-34 2015663 367833 4 0.1825
35-39 1771680 237297 5 0.1339
40-44 1479575 95357 6 0.0644
45-49 1135129 38125 7 0.0336
 Computational procedure for estimating TF

Step3: Calculate the cumulated fertility schedule Ø(i): Add the values of f(j),
beginning with j=1 and ending with j=I, and multiply this sum by five (this
number is used because five-year age groups are being considered .
Formula:  i 
 (i )  5   f (i ) 
 j 0 
Age group Index Period fertility rate Cumulated fertility
‘i’ f(i) Ø(i)
(1) (2) (3) (4)
15-19 1 0.1063 0.5315
20-24 2 0.2296 1.6795
25-29 3 0.2154 2.7565
30-34 4 0.1825 3.6690
35-39 5 0.1339 4.3385
40-44 6 0.0644 4.6605
45-49 7 0.0336 4.8285
 Computational procedure for estimating TF

Step4: Estimate the current average parity equivalents for a period

F(i): These are estimated by interpolation using the f(i) and Ø(i)
calculated in previous steps. Formula:
F (i )   (i  1)  a (i ) f (i )  b(i ) f (i  1)  c (i ) (7)
Ex : F ( 4)   (3)  a ( 4) f ( 4)  b( 4) f (5)  c ( 4) (7)
• Note: Period fertility rates were calculated from births in the 12 months preceding
the survey, tabulated by age of mother at the time of the survey, therefore,
coefficients should be used are as follows:
Coefficients for computing F(i)
a(i) b(i) c(i)
2.531 -0.188 0.0024
3.321 -0.754 0.0161
3.265 -0.627 0.0145
3.442 -0.563 0.0029
3.518 -0.763 0.0006
3.862 -2.481 -0.0001
3.828 0.016 -0.0002
 Computational procedure for estimating TF

Step4 (Continued): Estimate the current average

parity equivalents for a period F(i):
F (i )   (i  1)  a (i ) f (i )  b(i ) f (i  1)  c (i ) (7)
Ex : F ( 4)   (3)  a ( 4) f ( 4)  b( 4) f (5)  c ( 4) (7)
Age group Index Period fertility Cumulated fertility Estimated parity
‘i’ rate f(i) Ø(i) equivalent F(i)
(1) (2) (3) (4) (5)

15-19 1 0.1063 0.5315 0.237

20-24 2 0.2296 1.6795 1.209
25-29 3 0.2154 2.7565 2.338
30-34 4 0.1825 3.6690 3.323
35-39 5 0.1339 4.3385 4.094
40-44 6 0.0644 4.6605 4.503
45-49 7 0.0336 4.8285 4.789
 Computational procedure for estimating TF

Step5: Calculate a fertility schedule for conventional five-year age groups f+(i): It
can be estimated by weighting the rates referring to unorthodox age groups
according to equations (B.5) and (B.6) given below and using the coefficients
displayed in the below table.

f (i )  (1  w(i  1)) f (i )  w(i ) f (i  1)    ( B.5)
w(i )  x(i )  y (i ) f (i ) /  (7)  z (i ) f (i  1) /  (7)    ( B.6)
Coefficients for computing corrected
ASFR's
x(i) y(i) z(i)
For f+(i)
0.031 2.287 0.114 values see
0.068 0.999 -0.233 the table
0.094 1.219 -0.977
given in
0.120 1.139 -1.531
0.162 1.739 -3.592 the next
0.270 3.454 -21.497 slide
 Computational procedure for estimating TF

Step 6: Calculate the adjusted period fertility schedule from f+(i): Formula:
* 
f (i )  K  f (i ), Where K  adjustment factor
Age Index Reported Fertility rate for P/F ratio Adjusted fertility
group ‘i’ period fertility conventional age P(i)/F(i) rate F*(i) = K.f+(i)
rate f(i) groups f+(i): (Here K= 1.500)
(1) (2) (3) (4) (5) (6)

15-19 1 0.1063 0.1263 1.624 0.1895

20-24 2 0.2296 0.2323 1.528 0.3485
25-29 3 0.2154 0.2131 1.491 0.3197
30-34 4 0.1825 0.1784 1.480 0.2676
35-39 5 0.1339 0.1282 1.432 0.1923
40-44 6 0.0644 0.0595 1.376 0.0893
45-49 7 0.0336 0.0280 1.270 0.0420

Total 0.9657 0.9658 K=((P2/F2)+(P3/F3)+(P4/ 1.4489

F4))/3 =
TF 4.83 4.83 (1.528+1.491+1.480)/3 =1.500 7.24
 Calculation of TF
 Step 7: Calculate Total Fertility (TF):
 Once all the f*(i) values are available, one may calculate total fertility
(TF), as follows:

 7

 
*
TF  5 f (i ) 
 i 1 
• Note: One can also calculate the adjusted birth
rate, and the adjusted general fertility rate, once
we have the adjusted ASFRs of all age groups
 How to Select an adjustment factor K ?
 As a first step calculate the P/F ratios
 Ideally these ratios should be fairly similar for different values of i
 If children ever born are increasingly omitted by older women, the ratios will
tend to decrease as age increase (especially over ages 30 or 35)
 They are often far from being constant, even below age 35, in practice
 If P(2)/F(2) and P(3)/F(3) are reasonably consistent , either one of them can
be used as an adjustment factor for the period fertility rates
 If not very similar, take a weighted average of the two (using as weights the
umber of women in age groups 20-24 and 25-29 as a proportion of all
women aged 20-29)
 If there is any evidence suggesting that the population is experiencing a
fertility decline affecting mainly women in the older age groups, the value of
P(2)/F(2) is recommended as an adjustment factor because it is less likely
to be affected by the decline
 In general, P(1)/F(1) should be disregarded because of the intrinsic difficulty
in estimating F(1) and the P/F ratios for age groups over 30 cannot be
regarded as reliable due to the possible omission of children ever born
 Naturally, the more consistent the set of ratios obtained, the more
confidence one cane have in the adjustment factor selected
 Computation of adjustment factor ‘K’
Adjusted ASFR's
((P2/F2)+
((P3/F3)+ (P3/F3)+
P2/F2 P3/F3 P4/F4 (P4/F4))/2 (P4/F4))/3

ASFR *
[or f+
Age (i)] 1.527 1.490 1.480 1.485 1.500
15-19 0.1262 0.1928 0.1881 0.1868 0.1874 0.1895
20-24 0.2323 0.3548 0.3462 0.3437 0.3450 0.3485
25-29 0.2131 0.3255 0.3176 0.3153 0.3164 0.3197
30-34 0.1783 0.2723 0.2657 0.2638 0.2648 0.2676
35-39 0.1282 0.1958 0.1911 0.1897 0.1904 0.1923
40-44 0.0595 0.0909 0.0887 0.0881 0.0884 0.0893
45-49 0.0280 0.0428 0.0417 0.0414 0.0416 0.0420
TFR 4.8285 7.3745 7.1962 7.1441 7.1701 7.2445
* Pattern corrected for one-half year between birth and
reporting; ASFR: Age-specific fertility rate.
 P/F Ratio Technique: Advantages

1) The method is not affected by the failure of some women to

report the number of children ever born if the women who did
not report have the same number of children as those who
reported

2) The method uses a relatively small amount of information,

although it requires special questions in the survey or census
 P/F Ratio Technique: Limitations
1) If fertility has not been constant, the results of the technique may be
biased upward

2)Errors in the data on children ever born will affect the results
 A) Age misreporting of women providing these data will have an
unpredictable effect

 B) Underreporting of children ever born will cause a downward bias

in the adjusted estimates. Children who died in infancy (especially in
very early infancy), as well as those living away from home, are the
births most likely to be omitted, especially by older women.

 C) Over reporting of children ever born will cause an upward bias in

the adjusted estimates. Over reporting of children can sometimes
occur when still births, late foetal deaths, or adopted children are
mistakenly included.
P/F Ratio Technique: Limitations(Cont.)
3) Errors in the age-specific fertility rates will affect the results:
• A) Age misreporting of women providing these data will have
an unpredictable effect

• B) If the pattern of fertility taken as the “actual” pattern

contains errors, the estimated ASFRs will be incorrect. This
may also affect the level of the TFR.
 P/F Ratio Technique: Software
 The Bureau of the Census has developed a spreadsheet to apply this
technique called: PFRATIO

 This spreadsheet estimates fertility based on the Trussell variant of the

Brass P/F ratio technique

 Users can select TFR and ASFRs corresponding to the P2/F2, P3/F3, or
P4/F4 ratio or another adjustment (e.g.; Some combination of P/F ratios)
 P/F Ratio Technique: References
 United Nations (1983) Indirect Techniques for Demographic
Estimation (Manual X). Sales no. [Link] York : United
Nations, Chapter II, Pp. 32-37

 E. E. Arriaga (1994) Population Analysis With Microcomputers, Vol.I

and [Link], New York: UNFPA, USAID, Bureau of the Census
 Example: Input
Table

COUNTRY: YEAR
Trussell P/F Ratio Technique
Cumulativ
Reported Average e
ASFR CEB Fertility P/F
Age f(i) P(i) Phi(i) F(i) ratio
15-19 0.053 0.133 0.265 0.113 1.176
20-24 0.140 0.901 0.965 0.674 1.337
25-29 0.121 1.902 1.570 1.345 1.414
30-34 0.075 2.850 1.945 1.818 1.568
35-39 0.030 3.650 2.095 2.040 1.789
40-44 0.015 4.250 2.170 2.140 1.986
45-49 0.005 4.810 2.195 2.189 2.197
Age code * 0
TFR 2.1950
Thank you