Ngalinda, Innocent: Age at First Birth, Fertility, and Contraception in Tanzania

139

Chapter 5. Fertility

5.1. Introduction

During the course of 1987, a milestone in human history was reached when the world population surpassed the five billion mark (five thousand million persons). The United Nations Population Fund (UNFPA) symbolically chose 11th July of each year to be commemorated as the ‘Day of the Five Billion’. This figure has a special significance when one realises that after millennia of little growth, the world’s population reached one billion around the beginning of the nineteenth century. It took about 160 years to increase to three billion and yet it has grown by an additional two billion since 1960 (UN, 1987). The sixth billion will be reached in June 1999 (PRB, 1998). Although the average annual rate of world population growth has been falling since 1965 from an historically unprecedented rate of 1.99 percent to 1.67 percent in 1985 and 1.4 percent in 1998, the absolute size of the annual increment to world population has been rising from around 63 million persons in 1965 to 77 million in 1985, further to 93 million in 1998, and is projected to reach a peak level of 1.9 billion in 2010 (PRB, 1998). This results from the built-in population momentum of each cohort of women in the child bearing age being larger than the previous one. Hence, the addition of the next billion to the world population is expected to take only 12 years, as compared to the 13 years it took for the population to increase from four to five billion, even though the growth rate during this period will be lower.

The key element behind this change in population, particularly in developing countries, is the level and pattern of fertility. Although mortality and migration also contribute to the size, structure and growth of population, and are hence important areas of study, population dynamics are strongly moulded by fertility. Fertility is also important because it is inextricably bound up with many aspects of the economic and social milieu. A better understanding of fertility behaviour may therefore yield insights relevant to a wide range of social and economic behaviour; patterns; and changes such as labour force participation, income distribution and educational aspirations for children.


140

Furthermore, an ability to estimate the magnitude of changes in fertility and the causes of those changes are required background information for the development of policy in many areas. For example, for the projection of expenditure on primary education, the analyst must know the number of children who may be enrolled in the future, and for this such projections, data concerning fertility and child survival are very important. Even more obvious, knowledge about fertility is necessary for the designing of policies so that they are likely to have direct or indirect effect on fertility and hence population growth. Within this context, planners and policy-makers may examine the probable impact of proposed policies and programmes directed towards other social or economic development objectives on fertility. Hence, it may be desirable to have an understanding of the linkages between fertility and socio-economic factors not only to enable the development of policies and programmes for achieving desired fertility goals, but also to contribute to the attainment of other development objectives through influencing patterns and levels of fertility.

Sources of such data include vital registrations, but in most of the developing countries vital registration systems are either incomplete or non-existing. In Tanzania vital registration somehow exist in Dar es Salaam and a few regional headquarters. The exercise of vital registration is a responsibility of the Attorney General’s office. But the collection of these data does not give room for any scientific study as they are always incomplete. For this reason, determining the current level of fertility in the absence of vital data is a big challenge. In this situation, the only reliable sources of vital rates are censuses and surveys. But direct estimations of crude birth rates and age specific rates from censuses and survey data in developing countries, have a lot of errors due to poor quality. This handicap called for many scholars to try and fill this gap of knowledge by developing different techniques of estimating levels of fertility and mortality, popularly known as indirect techniques.

In this chapter, attempt is made to examine the fertility levels, differentials, determinants, and trends in Tanzania. Indirect estimates of fertility are obtained from the three post-independence population censuses of Tanzania (1967, 1978 and 1988). These estimates give the national trends in fertility for the period 1967-1988. Methodology of analysis will be fully discussed to give a reader an overview of indirect techniques. This chapter also tries to study fertility trends by using the 1991/92 and


141

1996 Tanzania Demographic and Health Survey (TDHS) data. The second task in this chapter is to examine the determinants of fertility in Tanzania. The Bongaarts model is used to investigate the proximate determinants of fertility by using the two TDHSs. Finally, the socio-demographic determinants of cumulative fertility are analysed using bivariate and multiple regression analysis.

5.2. Population Census Estimates

In all three post-independence population censuses two types of data were obtained on which fertility estimations are based. First, women were asked questions regarding the number of children they had ever born. Second, women were asked how many children they had born in the twelve months prior to the census. The answers to the first set of questions give information on lifetime fertility, and those to the second set help us to determine current fertility.

Census data from developing nations as already mentioned earlier, suffer from some limitations. These limitations can be explained as those affecting current fertility data, age specific parity data, and maternity history data. Errors that affect current fertility data include age misreporting, omission of births, reference period error, and the use of short time period, which raise uncertainty in the reported fertility levels due to sampling variability of the observed number of births.

Age specific parity data on the other hand can be affected by misclassification errors arising from misreporting of age and/or duration of marriage, errors in the reported number of births (numerator) and women of specific age group (denominator). The most serious error in the reported births is the omission of births by older women, especially of those births that ended in the early death of the child. Older women also tend to forget grown-up children, those born to another husband or man, and children not present at home for various reasons. There are also factors that may tend to inflate the number of births, for example the inclusion of step or adopted children or grandchildren, the inclusion of births, and non-inclusion of parity of a sizeable proportion of women who did not state their parities, or a dash or a space left blank (UN, 1983).


142

For maternity history data, possible sources of variation other than cohort or period changes are misstatement of the age of women especially in their earlier lifetime fertility data, under-reporting of births of women above 35 years and unmarried adolescents who would not like to be reported as mothers. Those women, who died before the interviews were conducted, might have had a different fertility pattern from those interviewed. Given the possibilities of these distortions, caution needs to be taken in interpreting the reported data. In this situation, indirect techniques cannot be avoided.

The usual method of adjusting the ASFRs is to adopt one of a number of strategies that use the P/F ratio first proposed by Brass (1964). This is the ratio between the reported number of children ever born by women at a particular age (Pi) and the sum of the ASFRs to that age (Fi). In situations where there has been no decline in fertility this ratio should be around one. However, this is not the case due to the fact that F is affected by under-reporting of births 12 months prior to the census (assumed to be constant at all ages), and P is affected by omission of children, particularly by older women. The latter error is overcome by adjusting the ASFRs by the P/F ratio for women aged 20-29, since these women are believed to report their number of children ever born most accurately. This estimation procedure is based on the assumption that fertility levels and patterns have not changed markedly in the recent past.

5.2.1. Crude Birth Rate

The crude birth rate (CBR) is normally the first step to estimate the fertility of a nation. It is defined as the ratio of the total births in a population for a specified period to the total number of person-years lived by the population during that period, with the assumption that the population is closed to migration and experiencing constant age specific fertility and mortality rates eventually attaining a constant age distribution given by:


143

(5.1)

Where

Ca = Proportion of population under age a

b = Birth rate

r = Intercensal growth rate

Pa = Probability of surviving from birth to age a

Such a population is known as a stable population. But when mortality gradually declines without any change in fertility, the population loses its stability and become what is known as a ‘quasi-stable’ population. The age distribution of the quasi-stable population is close to the age distribution of the stable population that has the same level of fertility and current mortality. A number of scholars have suggested corrections of birth rates derived from the stable population to accommodate the decline of mortality (UN, 1967; Zachariah, 1970; Abou-Gamrah, 1976). But these methods are not easy to use, as they require the knowledge of the time mortality started to decline and the magnitude of decline. Therefore it is not easy to determine the time in which mortality in Tanzania started to decline as we have so far only three reliable censuses. In this study therefore, the robust estimate of birth rate developed by Coale (1981) and simplified by Venkatacharya and Teklu (1987) will be employed but only with the assumption that it is reliable with regard to the Tanzanian context.

One advantage of using the Coale’s robust method is that the cumulated age distribution is used to reduce the errors of age misstatement within the age range considered. In countries like Tanzania where one does not have a firm knowledge of the pattern of mortality, the birth rate estimates which use only rate of growth of population (r), proportion of children of both sexes under 15 years C (15-), and the probability of surviving from birth to age 5 (l5) can be used irrespective of the true mortality pattern of the study population without any serious error in the estimation value (Venkatacharya and Teklu, 1987).

The work of Coale, Venkatacharya and Teklu indicated the survival of the population of both sexes under age 15, using the life table corresponding to the Brass type of indirect estimate l5 , gave robust estimate of birth rate in the form:


144

(5.2)

Where br is the birth rate

C(15-) = proportion of children of both sexes under 15 years.

r = rate of population growth.

7.5 = period prior to the second enumeration.

15Lo = survivorship from birth to age 15.

By using the relationship between l5 and 15Lo Venkatacharya and Teklu (1987) found that:

15Lo was equal to U + V(l5) where U and V are constants obtained by fitting a straight line for various model life table families. The crude birth rate obtained by using this relationship is assumed to be close to reality since the population under 15 is cumulated. Therefore, error of omission in age below five years and age heaping between age 5-9 as well as shift from age 10-14 is reduced.

The values of U and V are tabulated and presented in Venkatacharya and Teklu (1991:47). Egero and Henin (1973) found that the north family life table was suitable for Tanzania. The values of U and V as given by Venkatacharya and Teklu (1991:47) are 0.161 and 14.789 respectively. By substituting these values in our formulae 5.2 we obtain:

5.3

Table 39: Input and estimated crude birth rate by using formulae 5.3

Census year

1967

1978

1988

Growth rate (r) in percentage

3.1

3.2

2.8

l5

.741

.769

.808

C (15-)

.439

.461

.457

CBR per 1,000

50

51

47

Source: 1967, 1978, and 1988 Censuses

Table 39 shows that crude birth rates in Tanzania have been very high, reaching almost 50 per 1,000. But in the 1980s, they seem to have declined by 3 per 1,000. This figure


145

suggests that fertility in Tanzania has just started to decline. Hence, more effort should be encouraged to reduce fertility. On the other hand, this decline can be attributed to the quality of data. It seems that earlier censuses had a higher incidence of misreporting errors whereas the 1988 census was of better quality.

The crude birth rate is a crude measure of fertility because the denominator contains a large population not exposed to child bearing, i.e. men, children and elderly persons. It also includes sexually inactive and non-fecund women of child bearing age. Despite this shortcoming, the measure is highly useful in measuring short run changes in fertility within a particular country. This measure is less useful in comparing the fertility of two countries because of differences in age structure (Ngalinda, 1991). Another shortcoming of CBR is that it is not very sensitive to small fertility changes as it tends to minimize them. For example if the birth rate rises, there is an increase of children in the population. This swells the size of the denominator and tends to understate the fertility increase. Therefore, CBR tend to understate the extent of a genuine fall in fertility. Besides those shortcomings, the crude birth rate is an exact measure of the impact of fertility upon population growth at a particular period. Demographers on the other hand, are interested in obtaining crude birth rate as it is one of the inputs in estimation of natural increase of any nation because crude birth rate combined with the crude death rate signals the rate of natural increase.

5.2.2. The Completed Family Size

The Completed family size represents the cumulated fertility of specific women for each successive age and involves only the variability of age (Kpedekpo, 1982). The completed family size is defined as the number of children ever born by the end of the reproductive period of a woman’s life. This exhibits much more stability than do age-specific fertility rates from year to year. This is important for demographic analysis as the exercise involves following-up a group of women born in a particular year for their entire reproductive life by recording the number of children they bear. Due to time and financial constraints in developing countries, the exercise is not widely used, instead the average parity of women aged 45-49 (P7) is taken to represent the completed family size with the assumption that fertility of older cohorts are equal to the current fertility


146

experience of women in child bearing ages. If the value of P6 is greater than P7, P6 is taken to represent the Completed family size.

Table 40: Mean number of children ever born

Age Group

Pi

1967

1978

1988

1991/92

1996

15-19

1

.527

.437

.313

0.269

0.235

20-24

2

1.978

2.048

1.553

1.424

1.335

25-29

3

3.181

3.664

3.154

2.919

2.804

30-34

4

4.160

5.111

4.764

4.434

4.215

35-39

5

4.778

5.933

5.846

5.831

5.460

40-44

6

5.017

6.209

6.396

6.900

6.698

45-49

7

5.204

6.081

6.474

6.941

7.285

Source: calculated from 1967, 1978, and 1988 Censuses; 1991/92 and 1996 TDHS

Figure 17: Completed family size in Tanzania

Source: calculated from 1967, 1978, and 1988 Census

Figure 17 suggests that fertility in Tanzania has been increasing from the observed Completed family size. But perhaps the figure of 1978 does not give a true picture as we took P6 instead of P7. Choosing P6 is due to the fact that the value of P6 is smaller than P7 an indication that the mean parity of ages 45-49 is more affected by omission and age misreporting. We assume the age group 45-49 for 1988 has fewer omissions as these are cohorts that had most of their child bearing period in the post-independence


147

era or modern times compared to earlier censuses. However, mortality of women also affects the measure since not all women reach age 45-49. Women mortality inflates fertility because those who died at a younger age on average might have had fewer children (different fertility behaviours). However for a constant fertility population, this methodology has one serious shortcoming as the completed family size might not be identical with total fertility rate.

5.2.3. Fertility Patterns

The reproductive period of a woman is usually considered to extend over a span of 35 years, from about 15 to 50. Ideally natural fertility could have an upper limit of as many as 35 births per woman. This means that a woman gives birth during the 35 years (15-49) of her life at 12 months interval between two successive births (Bongaarts and Potter, 1983). The upper-limit model assumes 9 months of full pregnancy and about 3 months after the birth of each child during which a woman cannot become pregnant. In reality, it has been estimated that natural fertility can reach around 15 births per woman if the fertility inhibiting effects of delayed sexual unions, sexual interruption (abstinence, coitus interuptus etc), and breastfeeding are removed (Bongaarts, 1978).

In an ideal situation, all women are expected to be fertile during their whole reproductive period. But a small proportion of women may be sterile throughout the entire span, and most women or their partners may be sterile during some part of the reproductive age span. The proportions of women who are sterile at different ages are unknown. But this could be shown by an age specific fertility schedule of a country in the absence of contraception. Generally, a hypothetical model of age specific fertility curve shows a general low rate at 15 and then rises reaching a maximum at ages between 20 to 29, sometimes between 25 and 34. Then there is a decline, at first a gradual, and then a steep one at older ages until it reaches the lowest level at age 50.

The analysis of the shape of the age specific fertility schedule is an interesting and important part of the study of fertility. This is due to the fact that the mean age at child bearing, which is closely related to the mean of this schedule, is important in the relation between total fertility rate and population growth. The shape of the age specific fertility curve is also the link between the total fertility rate and such variables as the age


148

at first birth and the age at menopause. For example a decrease in the age at first birth will affect the early part of the age specific fertility schedule and it will affect the total fertility rate through this part of the curve.

Another importance for the study of age patterns of fertility, is the implication it has on policy formulation. For example, two countries with the same gross reproduction rate (GRR) but different mean ages of fertility schedule would produce different annual crude birth rates.<21> Therefore the age patterns of fertility have a bearing on the natural growth rate. Hence in order to reduce the natural growth rate, one policy measure in a country like Tanzania would be to raise the age at first birth in order to reduce fertility rate at young ages. Considering the broad-based age structure of these populations, such policy is likely to reduce the number of infants that would have been born and consequently reduces the crude birth rate. However during the transition period, if the average span between the generations is growing, then the population growth decreases even at constant fertility levels.

Table 41: Reported ASFR Tanzania

Age

1967

1978

1988

1991/92

1996

15-19

.169

.135

.084

.144

.135

20-24

.334

.305

.227

.282

.260

25-29

.316

.295

.241

.27

.255

30-34

.260

.239

.219

.231

.217

35-39

.201

.183

.176

.177

.167

40-44

.115

.093

.097

.108

.087

45-49

.060

.039

.050

.037

.042

15-49

3422

3267

1.097

1.249

1.163

5 x (15-49)

7.275

6.445

5.485

6.245

5.815

Sources:

For 1967:

Bureau of Statistics, 1973.

For 1978:

Bureau of Statistics, 1983.

For 1988:

Bureau of Statistics, 1994.

Otherwise the data are calculated from the TDHSs.


149

Figure 18 shows the pattern of fertility by ages as reported by respondents aged 15-49 years in each census. The shape of the curves looks similar to those observed in other less developed countries. That is, the ASFR increases from early ages of child bearing (15-19 years) and reaches its maximum value in the age group 20-29 years. It then declines steadily to the end of the child bearing ages (45-49 years). The curves for the 1967 and 1978 censuses look similar, having a sharp peak at age group 20-24 whereas the 1988 census data has a peak for women aged 25-29. This implies a change in the age-specific fertility pattern. Another observation from the curves is that the ASFRs are consistently higher for the 1967 census compared with the 1978 and 1988 censuses. Whilst the 1978 census gives higher ASFRs for younger women than that in 1988, the opposite is true for older women (aged 40 years or above). It could be concluded from this that recently women have tended to bear more children at an old age than previously. However, it could also be argued that the observed pattern is due to better reporting of live births by old women in the 1988 census compared with the 1978 census. In summary, the age pattern of fertility observed in Figure 19 suggests a decline in fertility for women less than 40 years particularly for the period 1978-88.


150

Figure 18: Patterns of fertility in Tanzania

Source: calculated from 1967, 1978, and 1988 Census

The age specific fertility distribution can be classified into three broad group (UN, 1963).

  1. Early peak type (maximum fertility in the age group 20-24)
  2. Late peak type (maximum fertility in the age group 25-29)
  3. Broad peak type ((i) and ii) differs only slightly).

From this categorisation, the shape of the fertility curve of the 1988 census suggests a broad peak in recent years for Tanzania compared with earlier censuses. The easy way to crosscheck this finding is to examine mean age at first birth. In censuses however, it is not easy to investigate age at first birth, as there is no birth history but only the


151

number of children ever born. In a situation like that, we can simply assume mean age of fertility schedule to represent the average age at first birth.

The mean age of fertility schedule can be calculated from tabulated responses to a question about births occurring in the preceding year, with due allowance for the fact that women who report a birth during the preceding year would on average have been 15<22> months younger at the time of birth than at the time of census. Mean age at fertility schedule is defined as

Where is the mean age at fertility schedule, fi is the age specific fertility rate in an ith age group (i.e. for age group 15-19, i is 1 and for the age group 45-49, i equals 7), and xi is the mean age interval of the ith age group (i.e. for the age group 15-19 xi is 17, and 47 for age group 45-49). Table 42 shows the result of mean age fertility schedule by using this formula.


152

Figure 19: Mean age at fertility schedule

Source: calculated from Population Censuses of 1967, 1978, and 1988

Figure 19 shows that, the mean fertility schedule has been rising from 28.8 years in 1967 to 29.8 years in 1988, however this observed mean age at fertility schedule contradicts the finding of age specific pattern we observed above, that the earlier censuses prior to 1988 shows an early peak, while the 1998 indicate a shift from early child bearing to broad peak where child bearing is mostly spread between age 20 to 29. The observed mean age at fertility schedule signals the decline of fertility between 1978 to 1988, although there was a steady level of fertility between 1967 to 1978 so that the difference between these two mean age at fertility schedule is negligible.

To control the observed mean age of fertility schedule so that it will not suffer from the effect of age mis-statement errors, we will compare Table 42 results with computed mean based on a model schedule. One such a schedule is the one suggested by Coale and Demeny (UN, 1967) who studied the relationships governing the mean age at child bearing and the mean parity (Pi) at ages 20-24 and ages 25-29, P2 and P3 respectively. They found that the ratio P3/P2 in a population not practising birth control, depends primarily on the ages at which women begin their child bearing. A high value of P3/P2 indicates a late start and a low value indicates early start.


153

In this model the assumption is that the decline of fertility with age in a population not practising birth control follows a fairly common pattern, so that primarily the rising portions determine the mean age of the fertility schedule (m). Coale and Demeny found that the relationship between m and P3/P2 follows a linear curve of the form of:

(5.5)

Where alpha is the y-intercept and beta is the gradient of the graph in normal notations.

Based on empirical data, the values of alpha and beta were determined by least square estimates and found to be alpha = 23.28 and beta = 2.25 respectively. Then the equation becomes:

(5.6)

By using formulae 5.6 the calculated mean age at child bearing is tabulated below:

Table 42: Mean age of distribution of fertility

Census Year

1967

1978

1988

Mean (m) (observed)

28.8

28.9

29.8

Mean age at child bearing by using the parity approach

26.0

27.3

27.8

Source: calculated from 1967, 1978, and 1988 Censuses

After controlling the observed mean age of fertility schedule to save it from the effect of age mis-statement errors by using the parity approach, we did not find a shift in peak, rather the data suggest that Tanzanian fertility can be categorised as a broad peak from 1967 to 1988. This calls for more indirect techniques to confirm if there is any shift between censuses.


154

5.2.4. Level of Fertility

In this study we will use the total fertility rate (TFR), defined as the number of children a woman would have by the end of her child bearing years if she were to pass through those years bearing children at the observed age-specific fertility rates. This measure has an advantage that it is not affected by the age structure of the population like CBR. In calculating Completed family size we took the average parity of age group 45-49 to represent fertility level of a population. But average parities calculated from data on children ever born to a total number of women in the age group, can be distorted either by errors in reporting children ever born or errors of misplacing women out of their age group. Considering this shortcoming, the work of Brass and Coale and Demeny tried to minimize these distortions (Brass, 1980).

5.2.4.1. Parity Ratio Methods of Estimating Fertility

Coale and Demeny demonstrated that period TFR could be approximated as:

Where P2 = MNCEB by women aged 20-24

P3 = MNCEB by women aged 25-29.

The relationship is an empirical relationship rather than a theoretical relationship. It depends on the following assumptions:

  1. P2 and P3 are not affected by the reference period of misplacement errors and are least biased for omission.
  2. The parities are based on women in the younger ages when reporting is assumed to be more accurate.
  3. Fertility for the ages 15-29 has remained constant in the recent and past.

This implies that the deviations from these basic assumptions are likely to result in a biased estimate of TFR.

In a similar way Brass (1980) developed a short cut method of estimating fertility from mean parity by age under the following assumptions:

  1. The pattern of fertility follows the Gompertz curve.

    155

  2. The quality of the reported mean number of children ever born extends to age group 30-34. Then by using the second, third and fourth mean parities, he found that:

where P2, P3, and P4 are defined above.

He stressed that this method will give a better estimation than the Coale and Demeny method if ages and births extending to age 35 are reported accurately.

As a check for the choice of the formula to be employed, he suggested that the following conditions must be met:

  1. If < then the Gompertz model does not give a good fit for the reported mean parities of cohorts. In that case he recommends the use of .
  2. If > then the Brass formula is likely to give better estimates (Brass, 1980). Therefore he recommend the use of .
  3. Again, if the estimates are less than P6 and P7 there is an indication that the underlying assumptions are not met. Then the formula should not be used.

Let us examine Table 43 in relation to the conditions mentioned above. Condition three is not met for two years, as Completed family sizes are greater than all estimates for 1967 and 1988 censuses. The choice of TFR 6.6 in 1978 can be plausible.

5.2.4.2. P/F Ratio Methods of Estimating Fertility

Brass (1967) developed a useful formula of comparing lifetime fertility to cumulative current fertility and the ratio of the two under the following assumptions:


156

  1. Fertility for the population under study remained constant for sometime in the past.
  2. The reported number of children ever born to women in their early ages, say 15-35 is more or less accurately reported.
  3. The reported number of births for the previous year may suffer from errors resulting from inaccurate perception by the respondents of the reference period, but these errors are invariant with age.

5.2.4.2.1. 5.2.4.2.1. Brass P/F Ratio Method

In the P/F ratio analysis Brass (1967) recommended P2/F2. His assumption was that women in the age group 20-24 remember the number of children born by them sufficiently accurately because such women will have only a small number of births that could have occurred in recent years thereby reducing recall error. But some of these women could have entered into reproductive life in recent years. This can lead to a false conclusion that fertility has been on decline in the years proceeding the enumeration date. Also the cumulative fertility F2 which is based on women in the age group 15-25 is assumed to have misstatement errors that will have no effect on the age structure of the current fertility schedule, but that is not the case. In order to minimize these and other shortcomings, other scholars have tried to use the average of various combinations (Somoza, 1981). These adjustment factors are shown in Table 43 .

In all cases the (P2/F2 + P3/F3 + P4/F4) adjustment factor is somehow more plausible. But in lieu of problems of assumptions and quality of data, other methods for estimating fertility should be taken into considerations.

5.2.4.2.2. Brass Relational Gompertz Model (BRGM)

Brass modified his traditional P/F method through the intermediary of relational Gompertz model that has some modest but useful advantages. According to Brass, the simplest specific use of the model is the estimation of fertility from the mean parities reported at a census or any enquiry, provided the level of fertility can be assumed to have remained relationally constant and parities are also the current synthetic cohort values. However, experience has shown that the parities for the older women are too low because of the omission of births and possibly selection factors. The whole idea therefore was to get a model that can fit the mean parities of the younger women.


157

The Gompertz function can also be estimated by fitting the model to cumulated age specific fertility rates of younger women such as ages 15-19, 20-24 and 25-29. In short, the function is

Where F(x) is the cumulative fertility by age x. A and b are constants related to F, which is the Total Fertility Rate by the end of reproductive life.

The above equation can be reduced to a linear function.

In the transformation, Y(x) values of various populations are related to one another. One population therefore can be chosen as a standard for comparison with others. Booth (1979) developed a standard model based on the Coale and Trussell model of fertility gammas(x), hence:

Y(x) = alpha + betagammas(x) Where alpha and beta are constants reflecting the pattern of fertility. The model also holds if F(x) is replaced by mean parity Pi.

Zaba (1981) showed that the series of partial fertility ratios F(x)/F(x+5) or Pi/Pi+1 can also be represented linearly in the form:

i=1,2,...,7

x = 20,..., 50

The estimates of alpha and beta are obtained by fitting the approximately equivalent relations:

Z(i) - ei = alpha + beta gi

X(x)-ex = alpha + beta gx

The values of ei, ex, gi, and gx are tabulated for reference in Brass (1981).


158

5.2.4.2.3. Arriaga’s Approach

Arriaga (1983) also modified the P/F approach by extending it to a case of changing fertility rather than transforming the recorded age specific fertility figures to children ever born type figures. The difference between this approach and others is that the usual P/F ratio methods transform the current data into an equivalent measure of lifetime fertility. But Arriaga’s method transforms the data on children ever born into estimates of age specific fertility rates consistent with mean parities. Steps are as follows:

  1. For a given Pi, age specific fertility rates (ASFR) consistent with the mean parity are generated.
  2. ASFR are cumulated and compared with the reported cumulated ASFR.
  3. For these two sets of ASFR, the adjustment factors are then derived from the ratio of these two cumulated ASFR.

The UN software Mort pak-Lite is used to generate ASFR in this study.

5.2.5. Summary of Census Fertility level Estimates

Table 43 shows a summary of the estimates for all three censuses, which present the fertility rate estimated by using a range of different methods, together with a final ‘best’ estimate.


159

Table 43: Summary of fertility estimates by using various methods

S/N

Technique (method)

1967

1978

1988

1 Observed/Reported<23>

7.3

6.5

5.5

2 Official documented

6.6

6.9

6.5

3 Completed family size

5.2

6.2

6.5

4 Coale and Demeny (P3)2/P2

5.1

6.6

6.4

5 Brass modified Coale and Demeny’s formulae P2(P4/P3)4

5.8

7.7

8.1

P/F Ratio Methods:

6 P2/F2 adjustment factor

7.8

8.3

7.9

7 ½(P2/F2 + P3/F3)adjustment factor

7.2

8.0

7.8

8 1/3 (P2/F2 + P3/F3 + P4/F4) adjustment factor

6.9

7.8

7.7

Brass Relational Gompertz Model

9 fitted to current data base on 15-19

7.8

6.5

4.7

10 fitted to current data base on 20-24

7.4

7.7

5.8

11 fitted to current data base on 25-29

6.5

7.5

6.4

12 fitted to current data base on 30-34

6.0

7.4

6.9

Arriaga’s Approaches based on adjustment factor:

13 20-25

7.1 (6.9)

7.8 (7.9)

7.6 (6.7)

14 25-30

6.8

7.8

7.6

15 20-30

6.9

7.8

7.6

Range

5.1-7.8

6.2-8.3

4.7-7.8

Plausible

6.9

7.4

6.5

Source: calculated from 1967, 1978 and 1988 Tanzania Population Censuses data

The reported national annual TFR for 1967 using children ever born for women age 40+ was 7.3 and the Completed family size is far below the observed TFR. We have seen above that the two parity relationships cannot be used here as conditions are not satisfied. But the use of P/F ratio to adjust for under-reporting of births of older women inflates the fertility level further to a TFR of 7.8 ( Table 43 ). Egero and Henin (1973) argue that births in the period 12 months prior to the census might have been over-reported due to a problem with the references period. It could be possible that women reported their births since 7 July 1966 instead of 28 August 1966 because enumerators were told to ask people to recall their live births since one and a half months after Saba-


160

Saba<24> day in 1966. However, there is also the possibility that children were omitted due to the way the question was framed. In the Kiswahili questionnaire the question read: ‘Una watoto wangapi?’ which could mean present children excluding those dead and living elsewhere. Moreover, women aged 12 years and above were asked a single question on the number of children they had ever had. It has been argued that a single question on the number of children ever born increases the number of omitted births particularly among older women who are likely to omit children not living at home or those who died a long time ago (United Nations, 1983). This argument is supported by the relatively small value of average children ever born for women in age group 45-49 (P7) given in Table 40 .

The official estimate given by Egero and Henin for the 1967 census was obtained by computing TFR based on the ‘child-women’ ratio method. The estimated national TFR was 6.6 (Egero and Henin, 1973). Whilst the possibility of over-reporting due to misunderstanding of the reference period cannot be ruled out, this estimate places the official figure lower than the reported TFR (0.7 births per women). This value of TFR seems implausibly low. Egero and Henin did not take into consideration other problems that are believed to cause under-reporting of births 12 months prior to the census. For instance, in addition to the fact that only a single question was used, people, particularly in Africa, are usually reluctant to talk about recently dead children, especially those who died in infancy. This can deflate the number of recent births. It can therefore be argued that the reporting of births during the 1967 census was probably subject to errors in both directions (those that inflate and those that deflate the number of births).

The three combination average of the Brass P/F ratio shows somehow a plausible figure. The Brass Relational Gompertz Model, fitted with ages below 25, gives a higher figure than reported while that fitted with ages between 25 to 35 gives lower figures. Arriaga’s approaches show an average of 6.9 too. Therefore, it seems that many methodologies shown in Table 43 give an adjusted TFR of 6.9 which is argued to be more plausible for Tanzania in 1967 than the official estimate of 6.6 children per woman.


161

The data collection process was modified for the 1978 census so as to avoid the problems faced in the 1967 census. The first modification was to ask three questions on children ever born instead of just one. Women were asked to report the number of their own children still living with them, the number living elsewhere and the number who had died. The total number of children ever born was obtained by adding the three numbers reported. This method is known to minimise the problem of omission of births (United Nations, 1983). However in Table 43 above, P6 is greater than P7, which suggest that a significant number of births to older women are omitted.

Several other procedures were also used to estimate the TFR using the 1978 census data as explained above. TFR based on parities at various ages can be calculated using the Coale-Demeny method and the Brass method as the above conditions are satisfied. But for the 1978 data, we can take into consideration the Coale-Demeny method since it is the one which satisfies all conditions explained earlier, giving a figure of 6.6 ( Table 43 ). This figure however seem to be at the lower side due to the period in discussion. The value obtained by Brass Relational Gompertz Model based on ages 30-34 seems to give a plausible estimate of 7.4 for the 1978 population census.

The computational procedure followed to obtain TFR during the workshop on the initial analysis of the 1988 census data (Chuwa et al., 1991) put the official figure for 1988 at 6.5. This figure seems to be plausible as it is supported by the Completed family size; it is the average figure for BRGM and in line with Arriaga’s approaches. The national estimates show that fertility did not begin to decline in Tanzania until late 1970s or early 1980s. A modest fertility decline has been observed in the intercensal period 1978-1988 of one birth.

5.3. TDHS Estimates of Fertility

The Tanzania Demographic and Health Surveys provide an alternative data source with which to study recent fertility levels and trends in Tanzania. The complete birth history of live births for each woman was collected by using the women's individual questionnaire. It was used then for the calculation of fertility rates for the two surveys.


162

5.3.1. Fertility Patterns

The pace of reproduction reaches a peak in the age group 20-24 years and then declines steadily to the end of the reproductive life span. This pattern is broadly similar to that shown by the 1967 and 1978 population censuses ( Figure 20 ). The distinctive feature for age pattern of fertility observed for all TDHS is that ASFR for age group 15-19 is higher than that observed in censuses. This suggests a rise in teenage child bearing although the mean age at child bearing was almost the same as observed in the censuses, namely 28.8 years.


163

Figure 20: Trends in age pattern of fertility for all censuses and TDHSs

Source: calculated from 1967, 1978, and 1988 Censuses; 1991/92 and 1996 TDHS

5.3.2. Summary of TDHS Fertility level Estimates

Table 44 gives the TFRs computed for the two TDHS by using the same methodologies explained above. The observed TFR was 6.2 and 5.8 for the 1991/92 TDHS and 1996 TDHS respectively. Some results of 1996 TDHS were dropped as they either did not satisfy the conditions or are not in line with the plausible estimates. The Completed family size and the BRGM fitted to current data, based on 30-34, are on the highest levels. An estimated TFR of 3.9 for BRGM fitted to current data, based on 15-19, is at the lowest level in the Tanzanian context. The 1996 TDHS did satisfy the conditions for using parity ratio in the estimation of the fertility level.


164

We can conclude that the TFR ranged between 5.9 to 6.2 in 1991/92, between 5.4 to 5.8 in 1996. This shows a slight decline of fertility in four years between the two TDHSs. The plausible estimate for the 1991/92 TDHS is 6.1 live births per woman and for the 1996 TDHS is 5.6 live births per woman. The summary of findings is presented in Table 44 .

Table 44: Summary of fertility estimates by using various methods

S/N

Technique (Method)

1991/92

1996

1 Observed/Reported

6.2

5.8

2 Officially documented

6.3

5.8

3 Completed family size

6.9

7.3***

4 Coale and Demeny (P3)2/P2

6.0

5.9*

5 Brass modified Coale and Demeny’s formulae P2(P4/P3)4

7.5

6.8*

P/F Ratio Methods:

6 P2/F2 adjustment factor

6.7

5.4

7 ½(P2/F2 + P3/F3)adjustment factor

5.9

5.6

8 1/3 (P2/F2 + P3/F3 + P4/F4) adjustment factor

6.2

5.8

BRGM fitted to current data based on:

9 15-19

4.2

3.9**

10 20-24

5.5

5.2

11 25-29

6.1

5.8

12 30-34

6.5

6.1***

Range

5.9-6.2

5.4-5.8

Plausible

6.1

5.6

*

Does not satisfy criteria

**

On the lowest side

**

On the highest side

Source: calculated from 1991/92 and 1996 TDHS

5.4. Fertility Trends

It seems there has been some decline in fertility during the recent past. The TFR was 7.0 in the early 1980s and decreased to 6.5 in the mid-1980s before reaching 6.1 at the end of the 1980s. In the beginning of the 1990s it declined to 5.6 during the two and a half years prior to the 1996 TDHS ( Figure 21 ). Therefore Tanzania has experienced a reduction in fertility by about 1.6 births per woman (a 24 percent reduction) during a decade prior to the survey. This can be regarded as a substantial decline in fertility during the recent past.


165

Figure 21: Fertility trends in Tanzania 1967-1996

Source: calculated from 1967, 1978, and 1988 Censuses; 1991/92 and 1996 TDHS

The trend in sub-Saharan fertility has recently been the subject of an intense debate. Most of the countries that participated in the WFS conducted in the 1970s and early 1980s exhibited a rising trend in fertility (Cochrane and Farid, 1989). More recently, however, the DHS data obtained since 1986 show a declining trend in fertility in several countries, including Zimbabwe, Botswana and Kenya (Arnold and Blanc, 1990; Van de Walle and Foster, 1990; Freedman and Blanc, 1991; Rutenberg and Diamond, 1993).

This study has examined fertility trends in Tanzania using data from the three censuses and two TDHS. TFR was used as a measure of fertility. TFR estimated from the 1967 census data was 6.9 births per woman ( Figure 21 ). There appears to have been an increase in fertility between 1967 and 1978 since the TFR estimated from the 1978 census data was 7.4. This is due to the reasons explained earlier. The 1988 population census produced a lower estimate of 6.5. This was the first indication of a decline in fertility in Tanzania, as these estimates are of the same type of data and used the same method at different dates.


166

The data obtained from the TDHS provide another opportunity to examine fertility trends in Tanzania. The estimated TFR from 1991/92 TDHS was 6.1 births per woman; it decreased to 5.6 births per woman for the 1996 TDHS. These results show a declining trend in fertility and are consistent with those results obtained using the census data. From this study, the current level of fertility in Tanzania is estimated to be 5.6 births per woman. Possible reasons for this decline will be discussed in chapter 8. However, the TFR of 5.6 is moderate according to sub-Sahara African standards but very high in comparison with other countries like Egypt 3.6, the Dominican Republic 3.2, and Brazil (2.5). The TFRs given in Table 45 support this statement.

Table 45: Observed TFR in sub-Saharan African countries and other selected developing countries

Country

Year

TFR

Benin

1996

6.3

Central African Republic

1994

5.1

Comores

1996

5.1

Côte d’Ivoire

1994

5.7

Ghana

1993

5.5

Kenya

1993

5.4

Mali

1995

6.7

Tanzania

1996

5.8

Uganda

1995

6.9

Zambia

1996

6.1

Zimbabwe

1994

4.3

Egypt

1995

3.6

Bangladesh

1996

3.3

Brazil

1996

2.5

Dominican Republic

1996

3.2

Source: Macro International Website

In order to understand the causes of the fertility decline in Tanzania, it is necessary to study the determinants of fertility. The next section therefore deals with the proximate determinants of fertility.

5.5. Proximate Determinants of Fertility

The causes of the fertility decline in Tanzania are explored by analysing the proximate determinants of fertility using the 1996 TDHS. In Chapter 2 we discussed intermediate variables thought of as providing the link between social, cultural and economic factors,


167

on the one hand, and the physiological processes which ultimately determine fertility on the other - the proximate determinants. In Chapter 3 we defined and discussed the methodology to compute indices for Bongaarts’ model. Table 46 therefore shows the data used to compute indices for the Bongaarts' model and Table 47 gives a summary of the results. The estimate of Cm is 0.637. The estimate of Cm for Kibaha, Tanzania, calculated by Komba and Kamuzora (1988) was 0.72. This shows that the proportion of married women in Tanzania is declining; this has been fully discussed in Chapter 4.

Table 46: Data used to determine proximate determinants for the 1996 TDHS

a) Estimation of Cc

Methods

U(m)

e(m)

e(m)*u(m)

Pill

0.055

0.09

0.0050

IUD

0.006

0.95

0.0057

Injections

0.045

0.9

0.0405

Condom

0.008

0.62

0.0050

Sterilisation

0.019

1

0.0190

Others

0.051

0.7

0.0357

Total

0.184

0.1108

b) Estimation of Cm

Age

BCMW

CMW

AW

g(a)

m(a)

g(a)*m(a)

15-19

0.83

2.73

15.93

0.304

0.171

0.052

20-24

2.63

8.01

13.39

0.328

0.598

0.196

25-29

2.47

8.35

10.59

0.296

0.788

0.233

30-34

1.73

6.71

8.1

0.258

0.828

0.214

35-39

0.9

4.92

5.95

0.183

0.827

0.151

40-44

0.32

2.99

3.64

0.107

0.821

0.088

45-49

0.14

2.24

3.1

0.063

0.723

0.045

15-49

9.02

35.95

60.7

0.251

0.592

0.149

1.538

0.980

5*

7.692

4.898

BCMW = Births Currently Married Women; CMW = Currently Married Women; AW = All Women

Source: calculated from 1996 TDHS

Note: The method specific use-effectiveness levels, e(m), were obtained from Bongaarts and Potter (1983; 84). Explanation for the computational procedure and notations is given in the text.


168

The Cm of 0.637 indicates that non-marriage alone suppresses maximum fertility by about 36 percent to a total marital fertility rate (TM). This means, if Tanzanian women were all married throughout their reproductive period, the TFR would be nearly one-sixth (1/0.637) higher than it is, with an average of 8.8 births per woman. Another way to interpret this estimate is to say that late marriage, non-marriage and divorce or widowhood together suppress fertility by about 36 percent.<25> Thus the reduction in the exposure to sexual intercourse which results from divorce and widowhood occurring during women's reproductive period seems to be contributing substantially to this suppression of fertility.

Table 47: Summary measure of the proximate determinants of fertility for 1996 TDHS

Index

Intermediate step

Estimates of the index

Percentage suppresses maximum fertility

Cm

Sigmam(a)g(a)=.980

Sigmag(a)=1.538

0.637

36
Cc

u=0.184

e=0.602

0.880

9
Ca

TA=0

1.000

 
Ci i=15.7* 0.641 39

* From: Bureau of Statistics (1997; 83)

Sources: calculated from 1996 TDHS

TF = 5.6/(0.637 x 0.880 x 0.641 x 1)=15.6

Note: see text for details of the computational procedure

g(a) = the age-specific marital fertility rates

m(a) = the age-specific proportions of females currently married

u = proportion currently using contraceptives among married women of reproductive age (15-49)

e = average use-effectiveness of contraception

TA = total abortion rate

i = mean duration of postpartum infecundability

TF = total fecundity rate


169

Among currently married women, 12.5 percent were using family planning methods at the time of the interview (Bureau of Statistics, 1997). Contraceptive prevalence is therefore very low in Tanzania. The estimated Cc of 0.88 implies that about 9 percent of maximum potential fertility have been suppressed by use of family planning methods (or that if no women were using contraception, the TFR would be only slightly higher than it is). Bongaarts (1978) states that Cc ranges between 0.8 and 1.0 for countries with TFR greater than 5.

Unfortunately, there are no data with which it would be possible to calculate the value of the index of induced abortion (Ca), which is therefore assumed to be 1.0. Although induced abortion is illegal in Tanzania (unless undertaken to save the life of the mother) illegal abortion does occur, especially amongst young unmarried Tanzanian women living in urban areas (Justesen et al., 1992). Therefore the assumption of Ca being 1.0 is likely to over-estimate fertility levels.

Table 47 shows that postpartum infecundability has the strongest fertility-inhibiting effect of all the indices in Tanzania. The estimated Ci of 0.64 means that postpartum amenorrhea and postpartum abstinence suppresses maximum potential fertility by about 39 percent to the total natural marital fertility rate (TN) of 10. This is mainly a result of universal and prolonged breast feeding. Nearly all women who have given birth (99 percent) breast feed their children for at least the first year of life and the mean duration of breast feeding is 21.2 months (Bureau of Statistics 1997). These indices for the four proximate determinants of fertility, together with a TFR of 5.6 given in the preceding section, give a total fecundity rate (TF) of 15.6. Bongaarts (1978) suggested a TF value between 13.5 and 17.0 for countries with a TFR greater than 5, and an overall mean figure of 15.3. Therefore, the TF of 15.6 estimated for all women is very close to Bongaarts' mean figure. This adds credibility to the results obtained in this study. The analysis of the proximate determinants of fertility suggests that late marriage, divorce and widowhood, and especially postpartum infecundability are the main factors reducing the prevailing levels of fertility in Tanzania compared to its biological maximum. However, the use of contraception has only a very minor fertility inhibiting effect.


170

5.6. Decomposition of Change in Fertility

It was concluded earlier that Tanzania’s fertility has started to decline. However, it is believed that any change in a population’s level of fertility is of necessity caused by a change in one or more of the proximate determinants. Bongaarts and Potter (1983) argued that the transition of fertility decline could easily be explained by the transition of a population from natural to controlled fertility. They further acknowledged that for countries where induced abortion is restricted, this control is exerted through a rise in contraceptive use. Therefore deliberate control of fertility will have a greater impact on marriage and postpartum infecundibility than on other proximate determinants.

The fertility control influences marriage and postpartum infecundibility as follows: the proportion of married women will decline because the use of contraceptives liberalises sexual activities in so far as women no longer have to worry about premarital pregnancies; the ultimate evidence of premarital sexual intercourse which is condemned by society. To avoid that women had get married to enjoy sexual intercourse. Secondly, the duration of breastfeeding will decline. Before the transition of fertility decline, breastfeeding is a means of contraception. With the use of modern contraceptive methods there is no need any more for long periods of breastfeeding. In this study we will examine the change in fertility measures by analysing not only the 1996 but also the 1991/92 TDHS indexes of proximate determinants.

Table 48: Data used to determine proximate determinants for the 1991/92 TDHS

i) Estimation of Cm

Age of women

ASFR

m(a)

g(a)

15-19

0.195

0.254

0.2924

20-24

0.267

0.685

0.3898

25-29

0.251

0.795

0.3157

30-34

0.208

0.806

0.2581

35-39

0.170

0.840

0.2024

40-44

0.103

0.842

0.1223

45-49

0.030

0.784

0.0383

Total

1.224

0.653

1.619


171

ii) Estimation of Cc

Method (m)

u(m)

e(m)

e(m) x u(m)

Pill

0.034

0.90

0.0306

Sterilisation

0.016

1.00

0.0160

Condom

0.007

0.62

0.0043

IUD

0.004

0.95

0.0038

Other

0.043

0.70

0.0301

Total

0.104

0.0848

Source: calculated from 1991/92 TDHS

Table 49: Summary measure of the proximate determinants of fertility for the 1991/92 TDHS

Index

Intermediate step

Estimates of the index

Percentage of suppression of maximum fertility

Cm

Sigmam(a)g(a)=1.224

Sigmag(a)=1.619

0.756

24

Cc

u=0.104

e=0.815

0.908

9

Ca

TA=0

1.000

0

Ci

i=15.6*

0.587

41

* From: Bureau of Statistics (1993; 101).

TF = 6.1/(0.637 x 0.880 x 0.641 x 1)=15.1

Sources: 1991/92 TDHS

The decomposition of a trend in the TFR is based on the following equation for first (1) and last year (2):

Where:

Proportional change in TFR between the first and last year (Pf) is


172

Proportional change in TFR due to a change in the index of marriage (Pm) is

Proportional change in TFR due to a change in the index of contraception (Pc) is

Proportional change in TFR due to a change in the index of induced abortion (Pa) is

Proportional change in TFR due to a change in the index of postpartum infecundability (Pi) is

Proportional change in TFR due to a change in the remaining proximate variables - natural fecundability, spontaneous intrauterine mortality, and permanent sterility (Pr) is

Then the above equation can be rearranged as:

Pf = Pm + Pc+ Pa + Pi + Pr + I where I is an interaction factor.

Table 50: Decomposition of the change in the TFR between 1991/92 and 1996

Factors responsible for fertility change in Tanzania between 1991/92 to 1996

Percentage of change in TFR

Distribution of percentage of change in TFR<26>

Absolute change in TFR<27>

Proportion of women married (Pm)

-15.7

-191.5

-0.96

Contraceptive practice (Pc)

-3.1

-37.8

-0.19

Practice of induced abortion (Pa)

0

0

0

Duration of post partum Infecundability (Pi)

9.2

112.2

0.56

Other proximate determinants (Pr)

2.6

31.7

0.16

Interaction (I)

-1.2

-14.6

-0.07

Total (Pf)

-8.2

100

-0.5

Source: calculated from 1991/92 and 1996 TDHS


173

The table above indicates that the TFR decline of 8.2 percent between 1991/92 and 1996 can be decomposed into a 15.7 percent decline due to a decrease in the proportion of women married, a 3.1 percent decline due to an increase in contraceptive practice, and a 9.2 percent increase due to shortening of the duration of postpartum infecundibility. The remaining proximate variables together contribute only 1.2 percent, and the interaction factor equals -1.2 percent.

In order to ascertain the findings above, let us investigate the decomposition of change in birth rates. The advantage of this method is that it takes the contribution of shifts in age structure within the reproductive ages into account as well as changes in the proportion of the women in reproductive age among the total population. The necessity of adopting these two approaches is due to the poor reporting of marital status and possible overstatement of marital fertility rates.

The CBR is linked to its proximate determinants by the following equations:

CBR=SxCmxCcxCaxCixTF

S is an age-sex composition factor given by

Variations in S are caused by changes in the population’s age-sex structure. Then

Pb =Pb + Pm + Pc+ Pa + Pi + Pr + I although

pb = CBR2/CBR1-1 (proportional change in the CBR between two periods).

Ps = S2/S1-1 (proportional change in CBR due to a change in the age-sex composition).

CBR in 1991/92 was found to be 42.8 and 40.8 per 1,000 population in 1996.

Table 51: Decomposition of the change in the CBR between 1991/92 and 1996

Factors responsible for CBR change in Tanzania between 1991/92 to 1996

Percentage of change in CBR

Distribution of percentage of change in CBR

Absolute change in CBR

Age-sex structure (Ps)

3.8

80.9

1.6

Proportion of women married (Pm)

-15.7

-334.0

-6.7

Contraceptive practice (pc)

-3.1

-66.0

-1.3

Practice of induced abortion (Pa)

0

0.0

0.0

Duration of post partum Infecundability (Pi)

9.2

195.7

3.9

Other proximate determinants (Pr)

2.6

55.3

1.1

Interaction (I)

-1.5

-31.9

-0.6

Total (Pb)

-4.7

100

-2.0

Source: calculated from 1991/92 and 1996 TDHS


174

Table 51 indicates that CBR has declined by 4.7 percent from 42.8 to 40.8 per 1,000 population. The contribution of the age-sex composition change contributes a 3.8 percent increase to the CBR from 1992 to 1996. This figure indicates that even if there has been a decline in fertility in Tanzania for the period between two TDHSs, the reproductive age structure still affects the decline of fertility in Tanzania. Hence, there is a need to reduce the number of births in a year so as to have a potential decline in fertility in the future. However, it confirms that the marital status can also be explained as one factor attributed to the fertility decline in Tanzania.

5.7. Socio-Demographic Determinants of Fertility

The study of proximate determinants of fertility is not enough to inform policy. The indirect determinants need to be considered in order to understand the social and economic factors which can be manipulated to change fertility levels. In the analysis that follows, we compute the variation in the mean number of children ever born (MNCEB) with various social and demographic factors. Since the number of children ever born is known to be highly associated with age, it is necessary to examine differentials in cumulative fertility in different age groups of women. The results are summarised in Table 52 .

Age at first birth in a non-contracepting society becomes an important determinant of the length of reproductive life and thus is highly and negatively correlated with fertility. Late age at first birth shortens the reproductive period of a woman, which will consequently reduce the total number of children ever born (and vice versa). This pattern holds for age at first marriage, and age at first sexual intercourse except for the pattern given for women aged less than 25 years in the table. But if we remove the current age effect ( Table 52 ), again it shows the same pattern.

The data of older women at the age of 35-49 show that women who engaged in sexual intercourse earlier than age 15 had approximately three births more than those who had first sex experience at age 25 or more. A similar pattern is repeated for women who started child bearing before age 15, they have about three births more than women married after age 24. Exposure to the risk of child bearing at very early age can be


175

associated with high fertility as compared to the late entry for all three broad age groups. On the other hand, early exposure to the risk of child bearing could have a negative effect on the level of fertility because physical maturity especially of the pelvis often lags behind the ability to conceive. As such, the pelvis and other reproductive organs may not be mature enough for delivery of the foetus when conception takes place. These obstetrical complications of young women may sometimes result in secondary infertility (Chapter 4).

Married women have much higher cumulative fertility in all three broad age groups than their counterparts who have never married. Divorced women have the lowest fertility among the three ever-married groups whereas widowed women have slightly higher fertility than married women in the age groups 15-24. In general for all women, it seems widows have higher fertility than any other group. The difference between married women and widows is surprisingly high 1.186. This may be due to the participation in sexual activities after the death of their spouses.

It is interesting to note that younger women in polygamous marriages have higher cumulative fertility compared to women in monogamous marriages. This situation may be attributed to the competition effect, as young women in polygamous union would like to have as many children as possible to either compete with older wives of the same union or to satisfy the husbands need if the other wife is infertile. Moreover, the reason for polygamy is either to have children or to have many children, hence contraception might only be practised at older ages. On the other hand, older women (35-49) in polygamous marriages have lower fertility than women in monogamous marriages. This may be attributed to the reason for the necessity of polygyny to a family as explained in the former sentence. The frequency of intercourse is expected to be higher for a woman in monogamous marriage, than a woman in a polygamous marriage keeping other things constant. However, the prevalence of polygyny is likely to rise by the failure of the first wives to bear children. Polygyny would seem to be a result of infertility in such cases and not its cause (Ahmed, 1986; UN Economic Commission for Africa, 1983). The result of all women (15-49) shows that polygamous women have higher fertility compared to women in monogamous unions.


176-177

It seems that there is no sound difference in fertility among the different religious groups in Tanzania. The differences are marginal in all age groups except that young women in the category "others" have higher mean CEB than women in other categories. The category "others" includes women with traditional faiths or no religion, who are more likely to marry at a younger age particularly if residing in rural areas. However, with regard to older women, Moslems have the lowest fertility of all. Protestants have higher fertility in comparison with Catholics. In general, Catholics have higher fertility than other religious beliefs and Moslems have the least fertility of all.

Women residing in urban areas have lower fertility than their rural counterparts. As stated in Chapter 2, this pattern is also true for all countries in sub-Saharan Africa (Cohen, 1993). The difference becomes substantial as the age of women increases.

Table 52: Mean number of childrean ever born by selected socio-demographic characteristics and current age

Characteristics

Age of women

15-24

25-34

35-49

15-49

Age at first Birth

<15

2.691

5.316

7.458

5.868

15-17

1.874

4.418

7.138

4.632

18-19

1.508

3.668

6.663

3.830

20-21

1.318

3.110

6.193

3.535

22-24

1.073

2.662

5.289

3.501

25+

1.763

4.208

3.232

Age at first Marriage

<15

1.885

4.557

6.708

5.205

15-17

1.425

3.887

6.607

4.085

18-19

1.217

3.255

6.236

3.518

20-21

1.081

3.036

5.735

3.500

22-24

.796

2.542

5.383

3.155

25+

1.962

3.985

3.237

Age at first sexual Intercourse

at first union

1.356

3.922

6.697

4.549

<15

1.177

3.596

6.386

3.449

15-17

1.018

3.539

6.290

3.268

18-19

1.086

3.083

6.204

3.004

20-21

0.746

2.762

4.780

2.580

22-24

0.777

1.863

4.382

2.346

25 +

1.272

3.617

2.088

Marital Status

Never Married

.183

1.244

3.429

0.318

Married

1.412

3.700

6.608

3.942

Widowed

1.560

3.419

6.074

5.128

Divorced

1.196

2.534

5.013

3.169

Polygyny

Monogamous

1.353

3.700

6.542

3.794

Polygamous

1.488

3.577

6.445

4.332

Religion

Moslem

.764

3.350

6.038

2.862

Catholic

.777

3.478

6.372

3.125

Protestant

.663

3.364

6.556

2.964

Other

1.167

3.643

6.623

3.817

Place of Residence

Rural

.830

3.619

6.571

3.308

Urban

.618

2.817

5.389

2.366

Women’s level of education

No Education

1.071

3.995

6.633

4.567

Primary incomplete

.584

3.876

6.693

3.159

Primary complete

.805

3.201

5.334

2.289

Secondary +

.392

2.425

4.804

1.779

Partner’s education

No Education

.771

3.417

6.315

4.763

Primary incomplete

1.341

3.625

6.492

5.201

Primary complete

1.375

3.412

5.548

2.945

Secondary +

1.364

3.588

6.613

3.819

Source: calculated from 1996 TDHS

Partner education is another factor known to have an influence on fertility. The spread of education and literacy among women is believed to lead to fundamental changes in their reproductive behaviour. Table 52 shows that cumulative fertility decreases as the mother's level of schooling increases. Older women (35-49 years old) shows that women without education have an average of 7 children per woman compared to 5 children for women with secondary education. The pattern for young women is not clear perhaps because they have just started the reproductive process. Partner’s education gives more or less a similar pattern to that observed for women's education. One possible explanation is that women's education is highly correlated with their partners' education. There is also a possibility that partner's education influences fertility in a similar way as women's education.

In summary, the descriptive analysis ( Table 52 ) shows that age at first intercourse, age at first marriage, age at first birth, women’s level of education, marital status, polygyny, and type of place of residence have a significant influence on cumulative fertility. A woman is likely to have a large number of children ever born if she had her first sexual


178-179

intercourse, marriage and birth before reaching the age of 15, lives in rural areas, and has little or no education.

5.8. Regression Analysis

After examining the differentials in fertility, we are now in a better position to pick up associated variables and subject them to a more complex analysis in order to examine their relative importance as determinants of fertility. However, it is necessary also to account for the interrelationships between the chosen independent factors.

Multiple regression analysis is the multivariate analysis undertaken to examine the determinants of cumulative fertility in Tanzania. Only ever given birth women are included in the regression analysis.

Table 53: The variables used in the second regression analysis

Variable

Status

Description and Category

Abbreviation used

Number of children ever born to a woman

Dependent variable

continuous variable

V212

Age

Independent variable

15-19

20-24

25-29

30-34

35-39

40-44

45-49

RC

A2

A3

A4

A5

A6

A7

Age at first birth

Independent variable

<15

15-17

18-19

20-21

22-24

25+

RC

B2

B3

B4

B5

B6

Age at first marriage

Independent variable

<15

15-17

18-19

20-21

22-24

25+

RC

M2

M3

M4

M5

M6

Age at first sexual intercourse

Independent variable

<15

15-17

18-19

20-21

22-24

25+

RC

I2

I3

I4

I5

I6

Marital status

Independent variable

Never married

Currently married

Widowed

Divorced

Not living together

RC

Ma2

Ma3

Ma4

Ma5

Polygyny

Independent variable

Monogamous

Polygamous

RC

Po1

Religion

Independent variable

Catholic

Moslem

Protestant

Other/none

RC

R1

R3

R4

Woman’s educational level

Independent variable

None

Primary incomplete

Primary complete

Secondary +

RC

E2

E3

E4

Woman’s partner’s educational level

Independent variable

None

Primary incomplete

Primary complete

Secondary +

RC

Pe2

Pe3

Pe4

Place of Residence

Independent variable

Rural

Urban

RC

U1

RC = Reference Category

The nine variables used in the analysis along with a brief description of their measurement are given in Table 53 . The response variable is the number of children ever born (CEB) which ranges from 0 to 15. We have two models, in the first one we treat explanatory variables as pure continuous variables. In the second model, all explanatory variables are categorised. Hence the regression equation used is given as:

CEB = f (age, age at first birth, age at first marriage, age at first intercourse, marital status, type of marriage, education, partner’s education, religion, place of residence). Hence

CEB = alpha + beta1(AGE) + beta2(AFSI) + beta3(AFM) + beta4(AFB) + beta5(EDU) + beta6(POR) + beta7(RELIGION) + beta8(MARITAL) + beta9(POLYG)

alpha is a constant and beta1,..., beta9 are unstandardised regression coefficients for each of the respective explanatory variables.


180

Table 54: The OLS regression coefficients for the determinants of children ever born

Variable

Unsta. Coeffic.

Level of Sig.

Std. Error

(Constant)

1.337

0.000

0.223

Current age - respondent

0.233

0.000

0.004

Age at first intercourse

0.015

0.208

0.012

Age at first marriage

-0.057

0.000

0.009

Age of respondent at first birth

-0.201

0.000

0.011

Education in single years

-0.023

0.017

0.010

R2 (adjusted) = .643

Durbin - Watson = 1.846

Source: calculated from 1996 TDHS

Table 54 gives the ordinary least square (OLS) regression coefficients (unstandardised) for all five explanatory variables along with their standard errors (SE) computed for unstandardised coefficients. It is interesting to note that all variables selected are highly significant and explain the variation in CEB by 64.3 percent except age at first intercourse. The apparent effect of age on CEB is that an increase of five years in the age of a woman results in an increase in the numbers of births to a woman by one birth. If a woman starts child bearing at age 15, she is statistically expected to have 7 children at the end of her child bearing period. This finding corresponds with the TFR of almost 5.6 in Tanzania we obtained in this study.

Age at first marriage seems to be a weak factor in determining fertility in Tanzania although it is significant as raising the age at first marriage by 10 years results in a reduction of 1 child. However age at first birth is a strong factor in determining fertility in Tanzania as increasing age at first birth by 5 years results in a decrease in the numbers of children expected to be born to a woman by one child. This finding corresponds with the findings in Chapter 4 that delaying child bearing after the adolescent period has an impact on fertility.

Table 55 shows a weak relationship between education and cumulative fertility, though this relationship is less significant. Generally speaking, it is expected that the higher the education of women the lesser the number of children ever born (linearity). However, this finding seems to suggest that there is a non-linearity between the education of a woman and cumulative fertility. Education may break down birth-spacing practices without lowering fertility desires or increasing age at first birth and marriage, hence it


181-182

might be associated with much higher fertility, referred to as a compensation (Cochrane, 1979; Cohen, 1993), although in bivariate analysis it was found that education signals a significant influence on cumulative fertility.

The regression analysis above was to determine the association between fertility and other variables. It is necessary to estimate the net effect of each variable when variations in the other selected factors are controlled. The independent variables interrelate with each other and their interactions can alter the effects observed in Table 54 . All variables used in the following model have been treated as dummy variables ( Table 52 ). For each background characteristic, one category has been selected as the reference category (RC) and is omitted from the equation. The unit of analysis is all women who have experienced a live birth irrespective of their marital status.

Table 55: Regression results of the relationship between fertility and some selected variables

Variable

Category

Dummy

Unstd. Coeffic.

Std. Error

Age

15-19

RC

20-24

A2

0.932***

0.106

25-29

A3

2.374***

0.108

30-34

A4

3.600***

0.111

35-39

A5

4.766***

0.116

40-44

A6

5.806***

0.123

45-49

A7

6.374***

0.129

Age at first sexual intercourse

<15

RC

At union

I2

0.253***

0.075

15-17

I4

0.017

0.073

18-19

I5

0.114

0.089

20-21

I6

-0.113

0.125

22-24

I7

0.057

0.202

25+

I8

0.080

0.289

Age at first marriage

<15

RC

15-17

M2

0.108

0.084

18-19

M3

-0.043

0.098

20-21

M4

0.087

0.110

22-24

M5

-0.166

0.128

25+

M6

-0.837*

0.140

Age at first birth

<15

RC

15-17

B2

-0.419***

0.104

18-19

B3

-0.886***

0.112

20-21

B4

-1.424***

0.121

22-24

B5

-1.989***

0.133

25+

B6

-2.877***

0.163

Marital status

Never married

RC

Currently married

MA2

0.828***

0.137

Widowed

MA3

0.472**

0.178

Divorced

MA4

0.091

0.163

Not living together

MA5

0.409*

0.190

Polygyny

Monogamous

RC

Polygamous

PO2

-0.212***

0.058

Woman’s educational level

None

RC

Primary incomplete

E2

-0.016

0.070

Primary complete

E3

-0.179**

0.065

Secondary +

E4

-0.124

0.125

Partner’s educational level

None

RC

Primary incomplete

PE2

0.083

0.076

Primary complete

PE3

-0.187**

0.072

Secondary +

PE4

-0.236**

0.091

Religion

Catholic

RC

Moslem

R1

-0.366***

0.055

Protestant

R3

-0.056

0.057

Place of Residence

Rural

RC

Urban

U1

-0.458***

0.056

Constant

1.524***

0.169

R2 = .694

R2 (adjusted) = .692

Durbin - Watson = 1.858

*** Highly significant at level .001

** Significant at level .01

* Significant at level .05

Source: calculated from 1996 TDHS

As Table 55 shows, all the independent variables included in the multiple regression analysis explain 69 percent of the variance in the number of children ever born. But they are not the only variables responsible for the variation in the number of children ever born though they play a very significant role.

The results in Table 55 show that the respondents’ number of children ever born are determined to larger extent by age; the relationship is statistically significant. In general,


183

these results show that an increase of 5 years in age of a woman results in an increase of one birth i.e. a woman increases the number by one child in five years.

The regression results suggest that there is an inverse relationship between age at first birth and fertility. The higher the age of the respondent at first birth the lower the number of children ever born. Those who start child bearing early have a longer period of exposure to the risk and are thus expected to have more children than those who start child bearing late. In general the relationship is seen to be highly significant. A women who started child bearing between the age of 18-19 will have one child less than those who started child bearing below 15 years. Women who started child bearing between age 20 and 24 bear less than 2 births compared with those who start child bearing before attaining age 15. However, women who got their first child at an age above 24 years, everything being equal, are expected to have 3 children less than those who start child bearing aged less than 15 years; the difference is highly significant. In general the study found a strong relationship between age at first birth and fertility.

It seems that women whose first sexual intercourse experience coincides with first union have the highest fertility compared to those who experience sexual intercourse out of wedlock and before attaining age 15; the difference is statistically significant. However, it seems that age at first intercourse cannot solely explain the cumulative fertility of a woman.

Marriage is a weak factor in explaining fertility. Perhaps this is due to the increase of premarital births in Tanzania as seen in Chapter 4. However, it is after attaining age 24 when age at first marriage become significant in reducing fertility in Tanzania. There is no significant difference in the number of children a woman bears if she marries before attaining age 24. In general, the table reveals that age at first marriage has an inverse relationship with motherhood provided that the age at first marriage is above 24 years. Marital status is significant in relation to the fertility of a woman. Married women had more number of CEB than never married women; the relationship is highly significant. It seems that women in polygamous unions had fewer CEB than those in monogamous unions; the relationship is highly statistically significant. In bivariate analysis it was also found that women in polygamous union above 25 years have lower mean number of children ever born than those in the monogamous union ( Table 52 ).


184

As can be seen from Table 55 , it seems there is no significant relationship between the number of children ever born and the completion of secondary education contrary to our expectations. No education was used as the reference category. However women with secondary and above levels of education are expected to have lower fertility than women with no education and even lower than those who completed their primary education. This inverse relationship between education and fertility is partly due to attitude and value changes that come with education. Unlike women without education, these women are more likely to use modern methods of contraception to space their births. In addition, they are known to start child bearing late because they stay in the school system for more years than those with primary education. On the other hand women who completed their primary education have less CEB compared to women who did not attend school; the relationship is statistically significant. Perhaps women who complete secondary education use the ‘compensation factor’ as explained in Chapter 4.

In terms of religion, Catholic was used as the reference category. Moslems have the lower fertility than Catholics in Tanzania; the difference is statistically significant. Catholics are generally expected to have high fertility compared to other religious denominations due to the Vatican anti-contraceptive use doctrine. This finding contradicts the finding by Lucas (1980) and general knowledge that Moslems have higher fertility compared to Christians. This finding proves the church’s influences on their believers. We are tempted to say so as in Chapter 4 we found that Catholics have higher age at first birth compared with others denominations in Tanzania and that they are expected to have lower fertility than other religious denominations, but they have high fertility. The only reason for this contradiction may be that while others contracept Catholics do not.

The regression results show that the type of residence influences the number of children ever born. It seems that women residing in urban areas have fewer children than those in the rural areas; the difference is very significant. Although in Chapter 4 we found rural residents to have higher age at first birth, the difference for urban residence might be due to contraceptive use and modernity in urban areas.


185

5.9. Concluding Remark

This chapter has shown that according to the 1978, 1988 population censuses and TDHSs fertility in Tanzania declined from 7.4 births per woman in the 1970s to 6.5 in the 1980s and further to 5.6 in the early 1990s. Overall, there was a 24 percent decline in fertility during a decade prior to the 1996 TDHS. In Chapter 4 we noticed a slight shift from low age at first child bearing to higher age in recent years. Although fertility is still high, we can say the Tanzanian population is in a transition stage of fertility decline. A further decline in fertility depends on the success of the family planning programme in raising contraceptive prevalence and changing people’s positive attitudes towards large families.

The analysis of proximate determinants of fertility has shown a significant contribution by breastfeeding (through its effect on postpartum infecundability) on lowering fertility levels below the biological maximum. This is due to the fact that the use of contraception is not widespread in Tanzania. The contributions of postpartum abstinence, late marriage, divorce and widowhood in lowering fertility seem to be minimal. Therefore the use of contraceptives should be encouraged if a further decline in fertility is intended. Unfortunately, the effect of abortion on fertility could not be assessed in this analysis because of non-availability of data.

The analysis of proximate determinant on the other hand confirmed that Tanzania fertility is in transition of decline. For the period between two TDHS, there was a decline of 8.2 percent in TFR due to a decrease in the proportion of women married, an increase in contraceptive practice, and the remaining proximate variables i.e. natural fecundability, spontaneous intrauterine mortality, and permanent sterility also played a significant role in fertility decline. Although breast-feeding practice contributed to the decline in fertility in Tanzania, the study found out that there is a shortening of the duration of postpartum infecundibility (breast feeding practice). This has to be more compensated by contraceptive use if further decline of fertility in Tanzania is to be achieved.

The socio-demographic determinants of fertility suggest the factors that can be manipulated in order to accelerate a decline in fertility. Discourage premarital sexual intercourse, delay marriage up to at least 22 years, improve women’s level of education,


186

delay the age at first birth to at least 22 years, and reduce infant and child mortality. Raising a woman’s education level would probably raise her age at first intercourse and birth, reduce child loss, and improve her status in general. The most important way in which education influences fertility is through delaying the first birth by using modern contraceptives.

Fußnoten:
<21> The GRR is a fertility measure related to TFR. This measure is identical to TFR except that it refers to female births only. It can simply be obtained by multiplying the TFR by the proportion of all female births in a year. This indicates how many daughters a woman will have in her lifetime. With that in mind, we can equate TFR to GRR. Then for two populations with the same fertility level but different mean age at fertility will have different CBR and hence different population growth. For a policy formulation, the level of fertility may not necessarily be compared to minimum age at birth called fertility schedule.
<22> Six months younger plus 9 months (from January to 28.08 the Tanzania census day ).
<23> Observed or reported TFR means the calculated TFR by using direct methods. The official documented TFR is that rate which is given by the Bureau of Census as a final estimate. One must use several indirect techniques to get a plausible estimate as there is no specific indirect method to use in estimation of TFR so far.
<24> Saba-Saba day (7 July) refers to the day TANU, by then the ruling and the only political party, was established. This day used to be very special and was celebrated nation-wide every year. Hence, it was easier for a person to remember this event than 28 August.
<25> The question on temporary separation for married couples was not asked in the TDHS. Therefore the effect of temporary separation on fertility is not included in the calculation of Cm. This is likely to over-estimate the total marital fertility rate.
<26> This is obtained by distributing the total change in TFR (i.e. 191.5 was obtained by dividing 15.7 by the total and multiplying by 100).
<27> The total value for this column was obtained by subtracting TFR of the 1991/92 TDHS from of the 1996 TDHS. Then the total is multiplied by the distribution of change in TFR we get the Absolute change in TFR.


[Titelseite] [Abkürzungsverzeichnis] [1] [2] [3] [4] [5] [6] [7] [8] [Bibliographie] [Anhang] [Selbständigkeitserklärung]

© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.

DiDi DTD Version 1.1
a subset from ETD-ML Version 1.1
Zertifizierter Dokumentenserver
der Humboldt-Universität zu Berlin
HTML - Version erstellt am:
Mon Jan 17 19:31:05 2000