LECTURE THREE

MEASUREMENT OF AGE, CAUSES OF ERROR AND THEIR DETECTION;

ADJUSTMENT OF AGE DATA
3.1 Introduction

Age is an important demographic variable. It is the primary basis of demographic classification in

vital statistics, census and survey work. Much of the demographic information in use is expressed in terms of the
age at which something happens; for example, when an individual leaves school, enters employment or dies. Age
data provide an indispensable basis for studies of a nation’s manpower potential and its requirements for schools,
housing, food and various other kinds of goods and services. In conjunction with the appropriate vital statistics,
the fundamental measures of population growth, age-specific fertility and mortality rates, reproduction rates and
life table functions can be estimated. It is the starting point for population projections. Therefore, the age-structure
of a population is considered basic to the study of population problems. In view of the importance of age in
demographic analysis, this Lesson deals with the measurement of or in data.
Objectives

By the end of this lesson you should be able to:

1. Analyze the population composition by sex and age structure

2. Identify different forms of pyramids
3. Identify methods of measuring age in africa.

3.2 Population composition by sex and Age Structure

A normal age structure derived from a complete and relatively accurate census enumeration follows a typical
pattern. It begins with a large number (or proportion) in the first age group (0-4) of each sex. This gradually
diminishes in subsequent age groups until the final group at around 100 years when the number becomes
negligible or nil. Within the two sexes, males predominate in the first few quinquennial age groups and then fall
below the numbers females, the difference gradually widening at the advanced ages.

We therefore start this section by looking at the essential features of an African age distribution as presented in
table 3.1. The table shows the African population of Uganda (1969) obtained from a national census. The
structure, both in numerical (absolute) and proportional forms, classified by sex and five-year age groups are
shown. The same information can also be presented in single years of age. The African composite age distribution
in percentage form for thirty countries, reflects the major features of the age distribution in Africa. It is not a
model, but serves as an overall pattern against which the reliability of the age and sex distribution of individual
African countries can be compared. Any assessment or comparison has to take into consideration the magnitude
and the complexities of demographic patterns and relationships in Africa. We can also examine the statistical
1
information presented in table 3.1 in a graphical form. This graphical representation is shown in figures 3.1 and
3.2.

A feature of the age structure which is immediately apparent on inspection is that it is distorted. Examining the
age structure from the youngest to the oldest we find that Table 3.1 shows that relatively large numbers have been
reported as aged 0-9 followed by a sharp cutback (trough) in the 10-14 age group. This feature has been noted in
many African censuses and surveys. It could be a genuine feature of African age distributions, and could be due
to drastic reductions in the levels of infant and child mortality, which can cause an increase in the proportions
surviving to these ages. This, however, does not explain satisfactorily this underlying characteristic in African
censuses for a number of reasons. First, the large numbers reported as aged 0-9 appear to be a feature of almost
all the reported age distributions, irrespective of the date of the census or survey. Secondly, in countries where
data for more than one census are available, there is no tendency for the bulge to move up into the 10-14 and 15-
19 age groups. Finally, this type of distortion is common not only in African but also in those Asian countries
with longer histories of census taking. It was, for example a prominent feature of age distributions obtained from
the Indian census of 1901 and 1911, when relatively little improvement in the levels of infant and childmortality
had been effected. These features have also been very much evident in the 1969 Uganda census and are largely
the result of age mis-statement which is probably the principal cause of these phenomena.

Sex differentials in the age structure are also atypical, departing from the usual pattern in which males
predominate in the early stages but females form a greater percentage of the population in the later part of life.
The analysis of sex ratios can suggest the nature of age reporting errors. All these deviations in the age data have
to be accounted for or properly rectified while at the same time keeping the total numbers and the basic
characteristics of the population identical, or very close, to those recorded by the census.

3.3 Population Pyramid

A population pyramid shows the total picture of a population by age and sex. It is a useful way of illustrating the
age structure of the population.

Table 3.1 Age and Sex Distribution of the African Population (Uganda, 1969) and Composite African Age
Data for 30 African Countries

Uganda (1969) Composite

Absolute (numbers) Percentage Percentage
Age group M F M F M F

2
 0–4 904,392 925,080 18.96 19.74 17.50 17.50
5–9 731,289 729,787 15.33 15.57 15.30 14.60
10 – 14 568,079 517,416 11.91 11.04 10.80 9.40
15 – 19 411,381 410,920 8.63 8.77 8.20 8.30
20 – 24 334,705 378,689 7.02 8.07 7.20 8.70
25 – 29 348,866 377,535 7.31 8.06 7.70 8.90
30 – 34 302,073 297,211 6.33 6.34 6.50 7.40
35 – 39 256,454 236,022 5.38 5.04 6.10 6.10
40 – 44 196,729 190,591 4.12 4.07 4.80 4.60
45 – 49 167,982 148,746 3.52 3.17 4.30 3.80
50 – 54 147,735 143,904 3.10 3.07 3.30 3.00
55 – 59 96,717 80,870 2.03 1.73 2.30 2.00
60 – 64 98,455 90,731 2.06 1.94 2.00 2.00
65 + 205,003 159,101 4.30 3.39 4.00 3.70

Total 4,769,860 4,686,603 100.00 100.00 100.00 100.00

The Construction of a Pyramid

In a pyramid, the absolute number in each group can be represented by a rectangle. Ages are located on the
vertical axis and the totals at each of the different ages (or age groups) are shown on the horizontal axis. The
rectangles are placed on top of each other, the youngest age group at the bottom and the oldest at the top. The
rectangles for males are shown on the left side of the diagram and those for females on the right side. The open-
ended age group can be represented only approximately. For example, if the open-ended age group is 75 + (75
years and over) we can treat this as a 75 – 80 age group and draw the rectangles in this age group.

Forms of Pyramids
Pyramids are of different shapes. The shapes are determined by three factors: fertility, mortality and migration.
Therefore, each rectangle representing an age, or a group of ages, has a length that depends on:
a) The total at birth of the corresponding cohort or group of cohorts
b) The amount of reduction by mortality and
c) The amount of migration

An example of a pyramid with a broad base which tapes towards the older age groups is shown. Such a pyramid
is typical of a country in an early stage of demographic development with high birth rates and death rates and a
youthful age structure. This shape is representative of the age/sex pyramids of African countries.
If population distribution shows a narrow pyramid at the foot and bulges out in the middle, this suggests that
there has been a fall in the birth rate and that more recent generations are smaller than their predecessors. Age/sex
pyramids of developed countries tend to show this shape
.

Young and Old Populations

A familiar and recurrent problem in demographic analysis is to describe and to distinguish between populations;
for example, populations can be described as young or old populations. A young population contains a relatively
3
large proportion of children (persons under 15 years). It also tends to have a low median age (the age which
divides the total population into two halves, i.e. the age at which 50 per cent of the population are above and 50
per cent below). On the other hand, an old population tends to have a relatively large proportion of its number
aged 65 and above. It has a high median age.

1. The proportion of children in the 0-14 age group for all the African countries ranges between 43 and 48
per cent.
2. The proportion of old people aged 65 years and over ranges between 2 and 6 per cent
3. The median age is low and varies between 17 and 19 years.
4. The proportion of children under 15 years in Switzerland was about 23 per cent.
5. The proportion of old people aged 65 and over was about 11.0 per cent for Switzerland.
6. The median age was 33 years for Switzerland and much higher than those found in African countries.
The characteristics of the age structure embodied in (1) – (3) are typical of the age structures of African
countries and most developing countries; the characteristics (4) – (6) are typical age structures of developed
countries. These patterns of age structures have different implications for population growth in the different parts
of the world.

3.4 Methods of Measuring Age in Africa

As a result of difficulties in the estimation of age, a number of techniques have been evolved and used in an
effort to improve the quality of age reporting in censuses or surveys. Two of the methods are described: historical
calendars and the cohort identification method.

Historical Calendar Method

In this method, interviewers are provided with calendars of historical events of national or local significance
which can be used to help place the life events of an individual on the standard time scale. The technique has
often been applied in a systematic way where the interviewer is instructed to go through the list of historical
events with each respondent in order to identify the events the respondent recalls.
Limitations.
1. only national events, mainly of a political nature, can be used in the calendars, and these events often
mean very little to rural populations. W
2. hile it may be possible to prepare historical calendars specific to local regions, the task can be time-
consuming and expensive.
3. Unless the interviewer is given specific instructions on how these calendars are to be used, as well as
sufficient motivation and time to carry out these procedures properly, the results obtained can be of
questionable value, even if some effort has been expended to develop an adequate calendar.
The effective use of the calendars is a complex operation which cannot be applied blindly. The method must be
varied according to the nature of the respondent: for example, children for whom the answers are being given by
the parents, adults who are present and are answering for themselves and adults who are absent or who cannot or
will not answer themselves.

Cohort Identification Method

It involves the identification of age cohorts in the population, linking these age cohorts to the standard time scale,
and then in the actual interview asking about membership in the age cohort. For example, the method was used
among the Kikuyu and all those circumcised in the same year are regarded as belonging to the same age group,
each group being given a special name. These groups then form age cohorts from which age data can be obtained.
4
The usefulness of this procedure is limited because of the wide variation between different localities and peoples
in the nature, size and the very existence of such culturally generated age cohorts. In addition, migration of
persons from different cultural traditions can make the procedure difficult to apply. Despite these limitations, the
procedure can produce an appreciably more accurate age distribution than had been previously obtained.

A related procedure involves obtaining from each respondent the names of persons in the local area considered
to be of the same ages. Age cohorts are then formed of all those reporting themselves as contemporaries. Ages
are determined for each such cohort and then assigned to each cohort member.
Advantages
1. No elaborate materials, event calendars or conversion tables need be prepared in advance; the procedure
can be used in population lacking widely known historical events or without culturally derived age
cohorts.
2. It makes use of the fact that relatively accurate age information may be known for some persons in a
community. The procedure, however, is time-consuming and expensive.

3.5 Causes of Errors in Age data

Age reporting suffers from a number of errors due to a variety of causes. The reasons vary but the causes can be
classified as follows:
1. Ignorance of correct age
2. Carelessness in reporting and recording
3. A general tendency to state age in figures ending in certain ‘preferred’ digits
4. A tendency to exaggerate length of life at advanced ages
5. A possibly subconscious aversion to certain numbers
6. Mis-statements arising from motives of an economic, social, political or purely personal character.

Mis-statements of age can be classified in two groups: net age mis-statement and gross age mis-statements. Net
age mis-statement is the number of persons reporting at a particular age, minus the true number of persons at that
age. It can be distinguished from gross age mis-statement by the fact that some of the persons reporting
themselves as aged x who were not aged x are counter-balanced by persons aged x reporting themselves at other
ages. A net shift occurs when there is a systematic tendency for persons in a particular age range to report
themselves as either younger or older than the true age, which is not counter-balanced by errors in the opposite
direction. Age heaping errors are unbiased errors, i.e. the errors cancel in either direction. On the other hand,
age shifting or net shift errors are biased errors.

The causes of age mis-statements are common to most demographic investigations. In tropical Africa these causes
are reinforced by the fact that in almost all African cultures, numerical age has had no importance over the years.
A good part of the errors in age data in African censuses and surveys arise from this factor. As a result, the age-
sex distributions of populations in many African countries show the following irregularities:
1. Deficiency in the number of infants and young children
2. Heaping at ages ending in 0 and 5, so a relatively large concentration of persons enumerated with ages
ending in 0 and 5.
3. A preference for even ages over odd ages, so a relatively large concentration of persons enumerated
with even-numbered ages
4. Unexpectedly large differences between the frequency of males and females at certain ages
5. Unaccountably large differences between the frequencies in adjacent age groups
5
 6. Non-stated or unknown ages

Hence the reliability of age data obtained from censuses and surveys in countries of tropical Africa is generally
poor. Until recently, however, differentiation had been made only between broad groups of the population, the
simplest such division being between children and adults, since efforts to obtain age reports in greater detail than
broad groups were thought unlikely to succeed among a population where birth data are not readily known.
However, there has been remarkable progress in the collection of demographic statistics through field surveys
and censuses during the past two decades. This progress is the result of new national and international interest in
all aspects of economic and social development.
The fact that age in completed years has been asked in more of the recent censuses and sample survey
indicates a growing optimism that efforts to obtain detailed age reports are worthwhile. Attempts to obtain more
accurate age reports by reference calendars of local events and by taking into account indigenous methods of
reckoning age have drawn attention to the possibilities for improvement. Blacker (1967), for instance, has
outlined how age grades, which have great significance among the Kikuyu of Kenya, can be converted to age as
customarily calculated with a fair degree of accuracy.
While age reporting can be expected to improve with social, educational and economic advancement of
the population, for the time being special techniques for assisting respondents in determining age will need to be
developed in the light of prevailing cultural patterns. More important, perhaps is the need for a continuous
appraisal of the quality of any data generated, through a systematic analysis of resulting age distributions.

3.6 Methods of Detecting Errors in Age Data

Age Ratios, Sex Ratios and Age/ Sex Accuracy Index

Age ratios and sex ratios can be used either separately or jointly in evaluating the quality of the census returns of
surveys by age groups. Age ratio is usually defined as the ratio of the population in the given age group to one
half the population in the two adjacent age groups. Mathematically, this can be defined as follows:
Let 5 Px be the age group from age x to age x + 5, 5 Px −5 and 5 Px + 5 be the preceding and the following age
group respectively, then
5 Px
Age ratio = 1 x 100 3.1
2
( 5 Px − 5 + 5 Px + 5 )

The computed age ratios are then compared with the expected value, which is usually 100.0. The discrepancy at
each group is a measure of net age misreporting. An overall measure of the accuracy of an age distribution, called
an age accuracy index, is derived by taking then average deviation (regardless of sign) from 100.0 of the age
ratios and summing over all age groups. The lower this index, the more adequate the census data on age. The
age ratios are usually calculated for males and females separately and can be calculated for each age group (except
the youngest and the oldest) provided the intervals are equal. An age ratio under 100 implies either that members
of the group were selectively under-enumerated or that errors in age group. A ratio of more than 100 suggests
the opposite of one or the other, or both, of these conditions. Generally, age ratios should be studied for a series
of age groups, preferably for the entire span of age for which they can be calculated.

Age specific sex ratios (number of males per 100 females in each age group) can also be used in evaluating census
age data. The general pattern of the age-specific sex ratios is such that they approximate to the sex ratio at birth

6
 in the younger ages, and fall gradually with advancing age. For countries with a similar level of sex ratio at birth,
the patterns of the age-specific sex ratios are quite similar from country to country if there is no migration.

The United Nations proposed an age/sex accuracy index. This employs the age ratios and the sex ratios
simultaneously. The method consists essentially in the computation of sex ratios and age ratios, for five-year
groups of ages, up to about age 70. In the case of sex ratios, successive differences between one age group and
the next are noted and the average taken, irrespective of sign. In the case of age ratios, for either sex, deviations
from 100 are noted and averaged, irrespective of sign. Three times the average of sex ratio differences is then
added to the two averages of deviations of age ratios from 100, to compute the index. The resulting index is not
very exact and is usually regarded as an ‘order of magnitude’ rather than as a precise measurement. An example
of the application of the method is shown in table 3.3. In table 3.5, model sex ratios (sex ratios not disturbed by
migration), and the composite African sex ratio for thirty African countries, are given as examples.

The calculation of age ratios, sex ratios and joint scores, as presented in tables 3.3 and 3.4 is shown as follows:
Example 1:
P x 100
Age ratio = 1 5 x
( P + 5 Px + 5 )
2 5 x −5

Therefore, age ratio for males aged 5 –9

Males aged 5 − 9 x 100
= 1
2
(males aged 0 − 4) + (males aged 10 − 14)
(515,520) x100 2 x (515,520) x 100
= =
(643,041 + 357,831) (643,041 + 357,831)
1
2

= 103.01

Example 2
5 Pxm
Age specific sex ratio = x 100 ,
5 Px f
Where 5 Pxm stands for males aged x to x + 5, and 5 Pxf stands for females aged x to x + 5.
Age-specific sex ratio for persons
males aged 5 − 9
aged 5 – 9= x 100
females aged 5 − 9
515,520
= x 100
503,070
= 102.47
Table 3.3 Results of Age Ratios, Sex Ratios and Joint Score (Ghana, 1960)
Age Males Females
group Number Age Deviations Number Age Deviations Sex First
ratio from 100 ratio from 100 ratio difference
0–4 643,041 654,258 98.29 -4.18
5–9 515,520 103.01 +3.01 503,070 102.91 + 2.91 102.47 -8.16
10 – 14 357,831 90.47 -9.53 323,460 84.17 -15.83 110.63 +6.86
15 – 19 275,542 88.01 -11.99 265,534 82.20 -17.80 103.77 +20.58
20 – 24 268,336 96.85 -3.15 322,576 112.82 +12.82 83.19 -7.76
25 – 29 278,601 109.07 +9.07 306,329 107.78 +7.78 90.95 -7.68
30 – 34 242,515 101.72 +1.72 245,883 101.29 +1.29 98.63 -12.00
35 – 39 198,231 96.36 -3.64 179,182 91.55 -8.45 110.63 -5.42
7
40 – 44 168,937 105.26 +5.26 145,572 105.96 +5.96 116.05 -12.37
45 – 49 122,756 92.40 -7.60 95,590 84.11 -15.89 128.42 +9.99
50 – 54 96,775 106.31 +6.31 81,715 113.49 +13.49 118.43 -4.07
55 – 59 59,307 74.02 -25.98 48,412 71.04 -28.95 122.50 +6.20
60 – 64 63,467 54,572 116.30
65 + 113,185 100,392
Total (irrespective of sign) 87.26 131.18 104.62
Mean 87.26 131.18 104.62
11 11 12
Mean 7.9 11.9 8.7
Joint score (sex ratio score) + (males and female age ratio scores) 45.9
_______________________________________________________________________________Example 3:
Joint score = 3 x (sex ratio score) + (male and female age ratio scores)
= 3x 8.7 + (11.9 + 7.9)
= 45.9

Tables 3.3 and 3.4 present detailed calculations for age and sex ratios, as well as joint scores. Note that the two
methods suffer from one serious disadvantage: fluctuations in age ratios and sex ratios can be due to irregular
demographic events (epidemics, military action, etc.), even when the population is closed to migration. Unless
the history of the population is well documented, there is a likelihood of misinterpretation of irregular age or sex
ratios.

Table 3.5 gives the composite sex ratio and model sex ratios for populations with sex ratio at birth of 103 and 104
male births per 100 females. Notice sex ratio variations by age in the model and in the observed distribution for
Ghana shown in table 3.3 and for the composite African data in table 3.5 sex ratio in Africa is poorly recorded or
reported. The patterns in the reporting of sex ratios and age ratios in Africa are shown in figures 3.9 and 3.10.

Whipple’s Index
Whipple’s index is applicable where age is reported in single years. The index is obtained by ‘summing the Age
returns between 23 and 62 years inclusive, and finding what percentage is borne by the sum of the returns of years
ending with 5 and 0 to one-fifth of the total sum’. From the above definition, Whipple’s Index can only measure
the extent of age heaping at the digits 5 and 0. It is, therefore, not an efficient method, since heaping also occurs
on digits other than 5 and 0. There are also other types of errors apart from age heaping.

The result varies between a minimum of 100, representing no concentration at all and a maximum of 500, if no
returns are recorded with any digits other than 0 and 5. Between these extremes, the following scale for estimating
the reliability of the data can be used.

Quality of the Data Whipple’s Index

Highly accurate Less than 105
Fairly accurate 105 – 109.9
Approximate 110 – 124.9
Rough 125 – 174.9
Very rough 175 +

8
3.9 Patterns of Sex Ratios: Ghana, Uganda and Composite African Age Data

9
Table 3.4 Sex and Age Ratio Score and Joint /score for /selected /countries at Given Census Dates

Table 3.4 (cont.)

Algeria Botswana Ghana Kenya Libya Morocco Tanzania

1954 1966 1964 1971 1960 1970 1962 1969 1954 1964 1960 1971 1967
Sex ratio score
Age ratio score 13.3 5.3 5.6 6.2 8.5 7.2 11.7 7.7 17.9 7.0 26.7 17.1 12.3
Males
Females 11.7 5.3 10.9 6.1 11.7 9.1 11.2 4.6 17.1 9.5 24.0 15.6 21.4
Joint Ratio score 23.5 6.9 11.6 6.4 15.6 12.0 12.3 7.0 32.2 6.1 48.4 26.9 17.3
74.1 28.1 39.3 31.3 52.8 42.7 58.6 33.7 103.0 36.6 152.5 93.8 75.6

Tunisia Ceylon India Iran Pakistani Argentina Colombia

1966 1963 1961 1966 1961 1960 1964

Sex ratio score 5.7 4.2 5.9 10.0 8.2 1.8 3.8
Age ratio score
Males 6.2 7.7 14.6 17.5 12.4 3.6 9.1
Females 4.8 7.6 17.9 20.9 21.9 3.1 10.0
Joint Ratio score 26.8 27.9 50.2 68.4 58.9 12.1 30.5

An example of the calculation of the index is shown in Table 3.6.

10
Myers’ Index
Myers’ Index is an ingenious device which reflects preferences (or dislikes) for each of the ten digits from 0 to 9.
Like Whipple’s index, it is applicable where age is given in single years. The method derives a blended population
which is essentially a weighted sum of the number of persons reporting ages ending in each of the ten terminal
digits – 0, 1, 2, 3, 4, 5, 6, 7, 8, 9. The underlying assumption of this method is that if there are no systematic
irregularities in the reporting of age, the blended sum at each terminal digit should be approximately equal to 10
per cent of the total blended population.

If the sum at any given digit exceeds 10 per cent of the total blended population, it indicates over-selection of
ages ending in that digit (digit preference). Conversely, a negative deviation (or a sum that is less than 10 per
cent of the blended total) indicates under-selection of ages ending in that digit (digit preference). Conversely, a
negative deviation (or a sum that is less than 10 per cent of the blended total) indicates under-selection of a ages
ending in that digit (digit avoidance).

An overall measure of the extent to which there is digit preference and/or avoidance in a census age distribution
is the index of Preference, which is obtained as the absolute sum of deviations for each of the ten terminal digits
(or half the total of the absolute differences). It does not really matter for comparative purposes which of these
methods of computing the indices are used. The method facilitates comparison among different populations.
Comparison between different populations can be made by noting that, theoretically, Myers’ Index can vary from
0 to 180, representing, respectively, the situations where ages are reported accurately and where all ages were
reported with the same terminal digit.

Tables 3.7 and 3.8 show a numerical example of the calculations of Myers’ Index. From table 3.8 Myers’ Index
is given as follows:
Myers’ Index = (10.8 + 3.6 + 0.1 + 3.6 + 2.0 + …+ 2.5)
= 30.8

The student should note that some that some textbooks define Myers’ Index as equal to half of the sum of absolute
values of deviations from 10.0. Such a definition will give Myers’ Index (from the example in table 3.8) as 15.4
instead of 30.8 which we obtained using the sum of the absolute values of deviations from 10.0. The comparative
values of Myers’ Index given in table 3.9 use the definition of half the sum of absolute deviations. One further
comment is necessary;

Table 3.5 Composite African Sex Ratio and Model Sex Ratios
Age group Composite African Model of 103 males Model of 104 males
sex ratio per 100 females per 100 females
0–4 99.0 100.5 101.5
5–9 103.0 99.7 100.7
10 -1 4 113.0 99.6 100.6
15 – 19 97.0 99.7 100.6
20 – 24 82.0 99.7 100.7
25 – 29 85.0 99.7 100.6
30 – 34 87.0 99.5 100.5
35 – 39 89.0 99.1 100.1
40 – 44 101.0 98.3 99.2
45 – 49 113.0 96.9 98.8
50 – 54 108.0 94.7 95.6
55 – 59 114.0 91.5 92.4
60 – 64 103.0 87.9 88.7
65 – 69 106.0 84.2 85.0
11
 70 – 74 102.0 80.3 81.1
75 – 79 103.0 75.5 76.3
80 – 84 105.0 68.6 69.3
85 + 106.0 54.5 55.0

the age at which the calculation starts or ends in the estimation of the Index may vary and different results can be
expected by choosing different starting points; usually an inspection of the data will suggest where to start and
end because the incidence of heaping will become apparent from the inspection of data given in single years of
age. It is usually not advisable to go too far into the old age years because the incidence of a heaping may be
confounded with the incidence of age shifting in this part of the age span. In table 3.9, some comparable measures
of Myers’ Index of Preference by sex for selected populations are given. Figure 3.11 gives patterns of deviations
from the expected 10 per cent for selected African countries.

Table 3.6 Calculation of Whipple’s Index for Ghana, 1960

Age Males Age Males
23 38,687 25 77,141
24 51,289
30 110,379
25 – 29 278,591
30 – 34 242,515 35 64,091
35 – 39 198,231
40 – 44 165,937 40 81,515
45 – 49 121,756
50 – 54 96,775 45 44,654
55 – 59 59,307 50 52,024
60 38,377
61 4,779 55 14,540
62 7,866
Total 1,304,110 60 38,377

Total 482,721

482,721
Whipple’s Index = 100 1
=185
5 (1,304,110 )

12
Table 3.7 Application of Myers’ Method to Age Data for Males According to the Ghana Census of 1960

Terminal Numbers at ages specified Sum of Sum of

digit 10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 ages ages
10-89 20-89
0 88,147 79,113 110,379 81,515 52,024 38,377 17,875 10,177 477,607 389,460
1 52,863 49,502 28,659 17,598 8,561 4,779 2,057 956 164,975 112,112
2 82,708 49,745 48,234 37,339 15,707 7,866 5,053 1,665 248,317 165,609
3 62,944 38,687 26,048 16,570 8,268 5,172 2,066 744 160,499 97,555
4 71,169 51,289 29,195 12,915 12,215 7,273 2,745 1,480 188,281 117,112
5 64,321 77,141 64,091 44,654 14,540 12,785 6,845 2,927 287,304 222,983
6 55,878 51,820 43,567 22,388 15,993 4,306 3,167 1,593 198,712 142,834
7 45,690 44,319 24,139 13,281 8,247 4,512 1,023 883 142,094 96,404
8 64,133 63,855 40,275 26,546 12,325 7,007 3,420 885 218,446 154,313
9 45,520 41,466 26,159 15,887 8,202 3,767 1,728 1,224 143,953 98,433

Table 3.8 Application of Myers’ Method to Age Data for Males According to the Ghana Census of 1960

1 2 3 4 5 6 7 8 9 10
Terminal Sum Ages Product Sum Ages Product Blended Percentage Deviation
digit 10 – 89 10-89 20-89 20-89 sum distribution from 10
Coefficient Coefficient
0 477,607 1 477,607 389,460 9 3,505,140 3,982,747 20.8 +10.8
1 164,975 2 329,950 112,112 8 896,896 1,226,846 6.4 -3.6
2 248,317 3 744,951 165,609 7 1,159,263 1,904,214 9.9 -0.1
3 160,499 4 641,996 97,555 6 585,330 1,227,326 6.4 -3.6
4 188,281 5 941,405 117,112 5 585,560 1,526,965 8.0 -2.0
5 287,304 6 1,723,824 222,983 4 891,932 2,615,756 13.6 +3.6
6 198,712 7 1,390,984 142,834 3 428,502 1,819,486 9.5 -0.5
7 142,094 8 1,136,752 96,404 2 192,808 1,329,560 6.9 -3.1
8 218,446 9 1,966,014 154,313 1 154,313 2,120,327 11.0 +1.0
9 143,953 10 1,439,530 98,433 0 0 1,439,530 7.5 -2.5
Sum 19,192,757 100.0 30.8

13
Table 3.9 Myers’ Indices of Preference, by Sex, for Selected Populations (c. 1960)

Indexes of PreferenceIndexes of Preference

Country Census Total Male Female Country Census Total Male Female
date date
Reunion 1961 1.4 1.7 1.3 Botswana (Bechuanaland) 1964 8.6 8.7 8.5
Mauritius 1962 2.6 2.4 2.1 Tanganyika 1957 8.8 - -
Mauritius (ex.dep.) 1962 2.7 3.2 3.1 Kenya (non-African pop) 1962 11.7 10.8 12.9
Lesotho Gabon 1960-1 12.0 10.9 13.0
(Basutoland) 1966 6.4 6.6 6.3 Ghana 1960 15.7 - -
Uganda 1959 6.5 - - Togo 1958-60 16.4 17.2 16.1
Senegal 1960 6.7 6.6 7.0 Congo(Brazzaville) 1960-1 17.3 14.3 19.6
C. A. Rep. 1959-60 7.0 7.7 7.3 Kenya (ex.Nothern Province) 1962 19.8 19.0 20.5
S.W. Africa 1960 7.4 8.0 7.4 Liberia 1962 19.9 - -
White 1960 1.1 1.1 1.3 Islamic countries
Coloured 1960 2.9 2.9 2.8 Tunisia 1966 8.0 5.8 9.7
Bantu 1960 8.1 9.8 8.8 Algeria 1966 8.1 5.7 10.7
Rep. of S.A. 1960 8.1 7.5 8.6 Turkey 1960 22.3 - -
Asiatic 1960 1.7 1.3 2.2 Iraq 1957 26.7 22.8 30.5
White 1960 0.6 0.7 0.7 Morocco 1960 38.3 33.6 42.7
Coloured 1960 3.6 3.7 3.6
Bantu 1960 11.3 10.4 12.4
Swaziland 1966 8.3 8.1 8.5

14
Best Grouping Limits
Digital preference of the kind we have been discussing is to a greater or lesser degree worldwide
and its prevalence can be regarded as a rough measure of the reliability of the recorded age
distribution. Errors in age classification are troublesome, not only because they occur frequently,
but because their presence introduces errors in tabulations of population characteristics by age;
such tabulations are fundamental to more sophisticated analysis such as the construction of life
tables, calculations of measures of fertility and mortality and population projections.

The incidence of digital preference can be reduced or diminished by grouping single-year age data
into appropriate age groups. The problem of the best grouping limits is therefore of utmost concern
to us. A method for determining the best limits or optimum breaking-points has been developed
by Myers (1954). In order to locate the most satisfactory cutting-points, Myers suggested that the
blended sum percentages be added for five consecutive digits starting with 0, 1, 2, 3 and 4. Using
the data from table 3.8, the following results emerge:

Digit grouping Percentage of total blended sum Deviations from 50%

0–4 51.5 -1.5
1–5 44.3 +5.3
2–6 47.4 +2.6
3–7 44.4 +5.6
4-8 49.0 +1.0

The results suggest that age intervals 4 – 8 or 0 – 4 might result in less irregular age groups than
other cut-off points, because their deviations from the expected 50 per cent are minimal amongst
the five grouping intervals. The suggestion, however, has one limitation: the use of non-standard
age intervals, in the case of 4 – 8 for example, may not be advisable in situations where the age
distribution is to be used as a base for calculating age-specific rates where the numerator data may
be available only for conventional five-year groups.

When the main purpose of the data on population, deaths and births by age is to construct age-
specific vital rates and these data are obtained from the same source, or when it can be assumed
that the pattern of age mis-statement is similar in the population data and the vital events data, then
the calculation of rates from unsmoothed data may be preferable to adjusting both the numerator
and the denominator data separately and the rates obtained from the unsmoothed data may be
smoothed through graduation formulae.

3.7 The Graduation or Smoothing of Census and Survey Data on Age Distribution

Most age distributions are usually corrected by methods of graduation or soothing, but the
graduated age distributions normally provide only plausible patterns or rough approximations of
the expected age distribution for two reasons:
a) Genuine irregularities in the age composition, which may exist before the graduation, may
have been obliterated in the graduated figures

15
 b) The choice of the technique or graduation is largely arbitrary.
In the graduation of age distribution three methods are usually employed:
• graphic methods,
• moving average methods and
• curve-fitting formulae.

Graphic Methods
The best-known graphical method is the ogive. A population ogive is defined as a diagram which,
against each age, plots the population under (or over) that age. The method is built on the
assumption that in the absence of distributing factors, such as migration confined to a narrow age
group or a single year’s epidemic of an infantile disease, the relationship between the number of
survivors of a given age and the births from which they sprang may be expected to be a smooth
function of age. In this case, a population ogive will also be smooth, provided the incidence of the
original births was smooth. Lack of smoothness in the ogive of a census population is taken as an
indication of the presence of error. The method is appropriate for removing large irregularities in
the age distribution. It is easy to apply, but it is a subjective method. The errors cannot be
quantified.
A number of techniques, based essentially on the mathematical graduation of age data, can
be used to derive figures for five-year age groups that are corrected primarily for net reporting
errors. The figures obtained by graduations are subsequently compared with the recorded figures
to get an idea of the extent of age misreporting. Two of these methods are now presented.

Smoothing Age Errors Due to Digital Preference by Method of Moving Averages

Given the five consecutive quinary age groups P – 2, P – 1, P0, P1, P2 where the P’s refer to
enumerated populations either in single years or in five-year groups, then it can be shown that S0,
the graduated value of P0, is given by:
− P−2 + 4 P−1 + 10 P0 + 4 P1 − P2
S0 = 3.2
16
The use of the formula assumes that the pattern of error can be represented as follows:
Sx + Px + (-1)xe 3.3
Where Sx is the true population, Px is the enumerated population, and e is the error component. If
we put
x = 0, 1, 2, - 1, -2, we have
S0 = P0 + e 3.4
S1 = P1– e 3.5
S2 = P2 + e 3.6
S-1 = P-1-e 3.7
S-2 = P-2+e 3.8
For the proof of the formula, we have to assume that  S x = 0 or that the fourth differences of Sx
4

are all zero. The form specified ensures that alternate values of P require a quantity to be added
and the intermediate values require e to be subtracted. This is a mathematical model devised to
correct for the erroneous transfer of population from one age group to the next age group on the
assumption that the true population is smooth (where a smooth population is defined as one in
which the expected age distribution is a progression such that the population at each age is slightly
larger than the next older age group and slightly smaller than the next younger age group). An
example of the application of the method is given in table 3.0
16
 There are three general limitations with moving average methods
a) data at the beginning and the end of the series are lost (see table 3.10)
b) moving averages may generate cycles or other movements which were not present in the
original data
c) moving averages are usually affected by extreme values.

Table 3.10 Graduation of Ugandan Census Population 1969 (All Races) by Method of
Moving Averages

Age group Males Females

Enumerated Graduated Enumerated Graduated
5–9 736,169 734,371
10 – 14 573,723 522,759
15 – 19 415,970 419,961.4 415,243 416,082.9
20 – 24 338,311 354,790.4 382,283 386,511.5
25 – 29 352,284 338,970.1 380,901 367,876.4
30 – 34 305,756 316,732.8 300,463 306,774.8
35 – 39 259,617 255,822.6 238,729 239,238.1
40 – 44 199,129 203,467.2 192,470 189,714.7
45 – 49 170,264 169,962.4 150,312 158,278.6
50 – 54 149,483 141,843.8 144,965 130,871.1
55 – 59 98,073 107,754.7 81,673 97,587.5
60 – 64 99,448 89,422.3 91,182 77,856.9
65 – 69 62,115 69,551.5 49,605 58,975.8
70 – 74 54,992 49,014.9 46,256 39,431.2
75 – 79 28,013 31,932.9 20,516 25,138.2
80 – 84 26,746 20,963
85 + 34,041 22,218

Some of these limitations are somewhat overcome by the weighted moving average where
the central item (or items) is given the largest weight and extreme values are given small weights
as shown in the formula used in the example given in table 3.10. it cannot, however, be too strongly
emphasized that past fluctuations in mortality, fertility or migration can cause true irregularities in
age structures and these irregularities should not be eliminated in the course of smoothing or
graduation procedures. Examination of previous trends in fertility, mortality and migration may
be helpful in identifying discontinuities caused by fluctuations in these factors.

Carrier-Farrag Method of Graduation

Where the extent of digital preference is relatively trivial, estimates can be obtained by graduation
or smoothing based on the least squares of a polynomial fitting to a range of observed values, and
accepting the regression value at the centre of the range as the graduated value at that point. In
practice, the technique or method gives satisfactory results, though basically, it must be deemed
unsound because a least-squares fit, intended for the treatment of random errors, is not necessarily

17
valid when dealing with systematic patterns of errors such as age shifting. The carrier-Farrag
method is, therefore, devised to handle the problem of massive age shifting which sometimes
occurs in certain sections of the age span.
The method consists, essentially, of splitting the total numbers enumerated in two adjacent
five-year age groups (quinary age groups) by first grouping the quinary age groups into ten-year
age groups (denary). By the manipulation of simple algebraic formulae, using well-known
theorems of elementary difference equations, the basic working formulae can be derived. The
derivation is shown and a numerical example given in table 3.11.
Let ui be the population in the quinary age group 5(x + i) to 5(x + i) + 4 and Uibe the
population in the denary age group such that:
Ui = ui + ui + 1 3.9
If it is assumed that uiis a second degree function of i:
ui + 1= ui + u i 3.10
thenu1 = u0+ u 0 (i = 0) 3.11
and U0 = u0 + u1 3.12
Therefore: U0 = 2u0 + u 0 3.13
Similarly
u 2 = u0 + 2u0 + 2 u0 3.14
u3 = u0 + 3u0 + 32 u0
u3 = u0 + 3u0 + 32 u0 3.15
Therefore: U 2 = 2u0 + 5u0 + 4 u0 = u 2 + u3
2
3.16
Similarly
U 4 = 2u0 + 9u0 + 162 u0 = u 4 + u5 3.17
thus U 2 = u0 + u0 + 2 u0
1
2
5
2
2
3.18
and U 0 − U 4 = −8uo − 162 u0 3.19

hence 12 U 2 + 161 (U 0 − U 4 ) = u0 + 2 u0 +  u0 = u 2
2
3.20

Table 3.11 Graduation of Uganda Census Population 1969 (Males, All Races) by Carrier-
Farrag Method)

Age group Enumerated Denar Graduated

quinaryu1 Ui=ui+ui+1 quinaryui
5–9 736,169
10 – 14 573,723 1,309,892
15 – 19 415,970 417,881.3
20 – 24 338,311 754,281 336,399.7
25 – 29 352,284 347,490.9
30 – 34 305,756 658,040 310,549.1

18
 35 – 39 259,617 250,516.3
40 – 44 199,129 458,746 208,229.7
45 – 49 170,264 176,200.1
50 – 54 149,483 319,747 143,546.9
55 – 59 98,073 111,425.5
50 – 64 99,448 197,521 86,095.5
65 – 69 62,115 67,476.1
70 – 74 54,992 117,107 49,630.9
75 – 79 28,013
80 – 84 26,746 54,759
85+ 34,041

Procedure
U0 = u0 + u1 = 736,169 + 573,723 = 1,309,892
U2 = u2 + u3 = 415,970 + 338,311 = 754,281
U4 = u4 + u5 = 352,284 + 305,756 = 658,040

u2 = ½U2+ 161 (u0-u4)=½(754,281)+ 161 (1,309,892 – 658,040) = 417,881.3

u3 = ½U2 - 1
16 (U0-U4)= ½ (754,281) - 1
16 (1,309,892 – 658,040) = 336,399.7

Note: u2 + u3 = U2 = 754,281 and

½U2 - 1
16 (U0 – U4) = u0 + 3 u 0 + 2 u 0 = u 3 3.21

The key formulae in the graduation method are:

1
u2 = ½U2 + 16 (U0 - U4) 3.22

u3 = ½U2 - 1
16 (U0 – U4) 3.23

An example of its application is given in table 3.11

3.8 SUMMARY

19
3.9 ACTIVITIES

Identify and discuss three methods used graduation of age distribution seeking their
applicability in Kenya.

3.10 FURTHER READING

1. Klauke, A. (2000) Coping with Changing Demographics. Webpage at ERICDigests.org. An

analysis of the effect of changing demographic patterns on school enrollments and
education.
2. Kpedekpo G.M.K (1982) Essential of demographic analysis for Africa,London
Heinemann Educational books ltd.

3. Meredith, G., Schewe, C., and Haim, A. (2002), Managing by defining moments: Innovative
strategies for motivating 5 very different generational cohorts, Hungry Minds Inc., New York.

Using a4.flow chart Lars

Weber, explain theDemographic
2010: importance ofChange
managerial economics
and Economic in the -firm.
Growth Simulation on Growth
ModelsPhysica.
State the shareholder’s wealth maximization model and explain how it can be affected by adoption
of new technology by the firm

20