African Population Studies Union for African Population Studies ISSN: 0850-5780 Vol. 12, Num. 1, 1997

African Population Studies/Etude de la Population Africaine, Vol. 12, No. 1, March/mars 1997

Bolaji M. FAPOHUNDA and Prosper V. POUKOUTA

African Population Policy Research Centre, A programme of the Population Council, Nairobi, Kenya

Code Number: ep97002

Abstract

The objectives of this study are to examine the trends, assess socio-economic differentials, and to determine whether declines in desired family size are associated with declines in fertility in Kenya. Using the synthetic cohort analytical method, the study found a consistent and monotonic decline in desired family size over the periods studied. These declines are highly correlated with declines in total fertility rate (TFR) over the same period, thereby suggesting that observed declines in TFR are probably driven by changes in desired family size. The analysis of the differentials reveal that women's education, women's work and income status, ownership of durable goods, husband's desired family size, knowledge of modern contraception and ethnicity account for significant variations in desired family size.

Les objectifs de cette étude consistent à examiner les tendances, à évaluer les différentiels socio-économiques, et à déterminer si les baisses de la taille désirée de la famille sont associées à celles de la fécondité au Kenya. Utilisant la méthode analytique de la cohorte synthétique, l'étude a trouvé une baisse constante et uniforme de la taille désirée de la famille au cours des périodes étudiées. Ces baisses sont en étroite corrélation avec les baisses du taux global de fécondité au cours de la même période, au point de faire croire que les baisses observées au niveau du TGF sont probablement provoquées par des changements au niveau de la taille désirée de la famille. L'analyse des différentiels révèle que l'éducation des femmes, le travail et le niveau du revenu des femmes, la possession de biens de consommation durable, la taille de la famille désirée par le mari, la connaissance de la contraception moderne et l'ethnicité expliquent les fortes variations enregistrées au niveau de la taille désirée de la famille.

Introduction

Recent data show that fertility is declining in Kenya amidst speculations that such a decline would be impossible, given the constellation of cultural forms and norms that provide support for high fertility (Caldwell and Caldwell, 1987; Van de Walle, 1990). Such norms are said to be relatively impervious to social, economic and political changes and in countries, such as Kenya, where the majority of the population live in rural areas, such norms should cause transitory increases in fertility (Ascadi et al., 1990). However, recent declines in fertility in Kenya provide no support for this speculation. Therefore, to explain the current declines in fertility, one could look into the structure of demand for children and examine how this has changed over time. By explaining trends and differentials in desired family size (DSF), one may gain insight into the causes of the fertility transition that is currently underway in Kenya.

Demand for children is a major component of completed fertility (Bongaarts, 1995). A change in demand for children should lead to a change in supply of children, ceteris paribus. Therefore, an explanation of changes in demand should yield an understanding of changes in the supply of children. This assumption forms the basis of this study. Moreover, an understanding of changes in demand for children could also provide insight into women's attitude toward future child-bearing, desired completed family size and the structure of demand for contraception. When desired fertility is high, it provides the motivation to raise large numbers of children; when low, it should motivate women to apply fertility control measures. Thus, an analysis of desired family size (DFS) could enhance our understanding of women's attitude towards fertility control and of their intention to use contraception because expressed preferences regarding desired family size should be related to behaviour. The objectives of this study, therefore, are to: (1) examine trends in desired family size; (2) assess the socio-economic and demographic differentials in desired family size; and (3) determine factors which influence desired family size.

The paper is divided into three sections. Following the introduction, Section Two presents the literary review and then discusses the conceptual framework. The results are presented in the third and final section.

Literature review

The following questions are often used in Demographic and Health Surveys (DHS) to gather information on DFS: women with children are asked: "if you could go back to the time you did not have any children and could choose exactly the number of children to have in your whole life, how many would that be?" Childless women were asked: "If you could choose exactly the number of children to have in your whole life, how many would that be?" The reliability of these questions and the validity of measures based on them have received tremendous attention in the literature (Bushan and Hill, 1995; Ascadi and Ascadi, 1990; Van de Walle, 1992; Bongaarts 1992; Ware, 1974). Two main criticisms have been levelled against the conventional measurement of DFS. One relates to questions of post-facto-rationalization bias and the other to the ability of respondents to give numeric responses to questions related to DFS, especially in settings where fertility decision-making is beyond the control of the individual women. Ascadi et al. (1990) said that the ability to respond to such queries is questionable in societies where fertility is controlled by lineage, ancestors and gods--agents who do not recognize individual desires in fertility decision-making, and where fertility control is not widespread.

Mhloy (1994) argued that the conventional approach measures DFS at a fixed point in time, which assumes that a woman adopts the same fertility behaviour, and has the same views about the social and economic value of children throughout her reproductive life, whereas fertility preferences are constantly reviewed over the life course. Similarly, Rasul (1993) argued that the conventional measure of DFS is econocentric in that individuals, as decision-makers, are assumed to carefully weigh costs and benefits of making choices to satisfy personally-defined objectives. He argued that overlapping cultural, socio-economic and physical realities define the relative power of women and men in decision-making and, therefore, that changes in the circumstances affecting women could cause them to revise their fertility preferences over the life course irrespective of market considerations. Kent and Larson (1982) argued that the existence of ex-post-rationalization complicates the relationship between actual and desired family size and that it is not clear whether the close relationship often observed between stated and actual family size is fortuitous or systematic. Bushan and Hill (1995) argued that responses to questions on DFS are characterized by a great deal of ambivalence because individuals are not sure of what might happen in the future and that is why individuals say that the number of children to have is up to God. They argued that this uncertainty accounts for the high incidence of non-numeric answers to questions on DFS. It logically follows that the lower the proportion giving non-numeric answers as in the case of Kenya, the higher the reliability of the answers, and the less the role played by ancestors and Gods in fertility desires. On the basis of this logic, the measure of ideal family size is likely to be highly reliable in Kenya because the proportion giving non-numeric answers is one of the lowest in sub-Saharan Africa.

Two alternative measures of DFS have been proposed in order to remove the ex-post-rationalization and the uncertainty biases. These are: the wanted fertility (WF) method developed by Bongaarts (1992) and the prospective desired total fertility rate (PDTFR) developed by Bushan and Hill (1995). But none of these alternatives is a better measure of DFS than the conventional approach. Both alternative measures are based on the wantedness of children which may also be rationalized for the same reason that children are given by God and that their wantedness is beyond the scope of the individual's of choice. Moreover, ideal/desired family size is affected by the ceteris paribus assumption and/or ex-post-rationalization bias no more than questions eliciting information about the future, including but not limited to questions on contraceptive intention, the preferred timing of the next birth, or the wantedness of children, etc. Responses to these questions are widely used in demographic analyses. Besides, unless there is sound argument to the contrary, biases introduced by the ceteris paribus assumption and ex-post-rationalization bias can be assumed to be randomly distributed in the population and, as such, should not lead to any systematic error in the data. Even if we assume that it is not, only a small segment of the population--mostly older women who have completed fertility--will be affected (see also McCarthy et al., 1987; Ahmed, 1990).

DFS has several advantages over the alternative measures. First, it is a measure of fertility desires; there is no measure that provides an equally effective index of the potential for change in family size in the developing countries (Ware, 1974). Secondly, it reflects the norms and culture of a place (Bankole and Westoff, 1995), particularly, those that are related to the value of children (Kent and Larson, 1982). Thirdly, the literature reports significant correlation between DFS and fertility behaviour in different contexts. For example, Farooq et al. (1987) found considerable correspondence between DFS and contraceptive use in Egypt. The study indicated that 66% of the low-income rural women who said they wanted no more children were using modern contraception compared with 2% among those who said they wanted more. DFS was found to have a stronger effect on contraception than education and place of residence. Similarly, an analysis of DHS data from 18 developing countries revealed considerable agreement between stated preferences and demographic behaviour among women in these countries (Bongaarts, 1991). The study found that 85% of the respondents whose actual fertility exceeded their DFS said they wanted no more children. The inconsistences observed among the remaining 15% were attributed to unachieved sex composition preferences and to real life constraints which inhibited the realization of stated preferences.

Using the analysis of a longitudinal data set from Thailand, Knodel and Prachnabmoh (1977), argued that the contention that respondents rationalize their DFS on the basis of their actual family size is not valid, and that if post-facto-rationalization bias existed at all, only a small segment of the population would be affected. They argued that the fact that women with a given number of children state that number as their DFS indicates that DFS is related to actual fertility. It does not indicate that DFS is determined by actual family size or vice versa.

Similarly, McCarthy et al. (1987) demonstrated with data from Nigeria that the high incidence of non-numeric responses in DFS data is not due to uncertainty about DFS nor to the misunderstanding of the questions but, rather, to a feeling of lack of control over the number of children to have. Therefore, no-numeric answers are valid answers according to these authors. For instance, a multivariate analysis of data from Bangladesh revealed significant effects of completed family size, employment status, age and education on the odds of giving non-numeric responses.

The argument of ex-post-rationalization bias could also be an artifact of the demographers' mind-set. This is because whenever the stated DFS is lower than actual fertility, responses are assumed to be valid but whenever the stated DFS is greater than actual fertility, especially for women at the tail-end of their reproductive career, the response is said to have been affected by ex-post-rationalization bias. Campbell (1994) pointed out that findings from research conducted among 3,006 men and 539 women in Sierra Leone indicated that men and women whose desired family size exceeded their actual family size actually stopped child-bearing before they reached their DFS and not because they rationalized DFS on the basis of actual fertility. In earlier studies, Palmore et al. (1981) and Pullum (1981) argued that respondents made different assumptions about their existential reality in answering questions regarding DFS. For instance, in answering questions on desire for additional children, respondents stated their preferences taking into consideration real world constraints, such as their age, financial implications of additional children, marital problems, etc. Such constraints are not considered in answering questions on the total number of children desired, according to these authors. Given this explanation, it is conceivable that there are situations in which DFS exceeds completed fertility without necessarily resulting in post-facto-rationalization bias.

Moreover, recent data from around the world are increasingly showing that questions on DFS are meaningful to respondents (Zick and Xiang, 1995; Van de Walle et al, 1992; Cleland, 1992; Hardee-Cleveland, 1982; Cochrane et al., 1990; McCarthy et al., 1987; Zehri et al., 1988). These studies found significant variation in DFS by age, sex, education, place of residence, mass media exposure, age at first marriage, expected education for children, extended family ties and number of surviving children. Other factors that have been found to be highly correlated with DFS are gender composition of children, knowledge of modern contraception, infant mortality and old age security (Smith, 1993; Pust et al., 1985; Mousa, 1987).

Using data on three West African nations, Benefo (1990) found that differentials in women's socio-economic status account for much of the variation in DFS. The effect of contextual factors such as religion and kinship ties were significant. In Nigeria, McCarthy and Oni (1987) found that the number of surviving children, women's education and sex preferences significantly affect DFS. Also the analysis of data on young people in Kenya revealed significant negative effects of age, education, mass media exposure and modern orientation on DFS. Of these variables, education and age have the strongest effects (Musyoki, 1983).

However, studies exploring the correlates of DFS and the linkages between DFS and actual fertility in Kenya are limited. Yet such studies are necessary to understand the on-going fertility transition, identify the determinants of DFS and to provide insight that could inform policies and programmes. This study is intended to fill this gap.

Conceptual framework

The theory of fertility decision-making propounded by Easterlin is applied as an organizational framework in this research. This application borrows significantly from the work of Zick and Xiang (1994) who applied this theory in explaining the correlates of desired family size, using Chinese data. In its original formulation by Easterlin, the theory postulates that demand for children is affected by three proximate factors; price, income and tastes. The price of children is composed of the direct expenditure on children and the opportunity cost of time spent bearing and rearing children. The opportunity cost refers to both economic and psychological costs of alternatives forgone. The underlying assumption is that the higher the price, the lower the demand for children, given the household income and tastes. Household income is also hypothesized to affect demand for children. An increase in income is expected to lead to an increase in demand for children. The third component of Easterlin's theory is taste for children. Taste refers to the subjective preferences for children compared to general goods. The greater this preference, the greater the demand for children (see also Caldwell and Caldwell, 1987).

All three factors are measured in relation to other goods. In other words, children are assumed to be affected by market forces just as other goods are. But one major advantage of the Easterlin model over other economic theories of fertility decision-making (see Becker, 1981; 1991) is that he integrates the economic and sociological factors in explaining fertility behaviour (see also Zick and Xiang, 1994). Although the model has been criticized for not specifying factors which are important in different contexts (see Hardee-Cleveland, 1982), the integration of economic and sociological factors makes the theory appropriate in settings (such as Kenya) where market forces are conditioned by socio-cultural factors (see also Blake, 1986).

In measuring these three components, we relied on proxies as is the practice in most studies that have applied this theory, because most data sets do not contain direct measures of the different dimensions of the theory. For example, parental aspiration for sons/daughters' education could have been an appropriate measure of expenditure on children in the context of Kenya, but this variable is not included in data analyzed in this paper. In spite of these limitations, the analysis indicates that the theory is applicable in the context of Kenya and most of the hypothesized relationships are supported by data.

Opportunity cost is measured using information on whether women work and earn income (see also Zick and Xiang, 1995). The assumption is that those who work and earn income are likely to have a higher opportunity cost than those who do not work or those who work but earn no income and, as such, will be more likely to desire fewer children. Another dimension of the opportunity cost measured in this study is the psychological dimension. Zick and Xiang measured this dimension by using data on the type of households in which women live. They hypothesized that the pressure to raise larger numbers of children is likely to be greater for women who live in extended family environments than for women who live in nucleated family environments. This logic is applied in this study. However, we used place of residence as a proxy for the psychological cost dimension. We hypothesized that women residing in rural areas are more likely to live with other family members than those located in the urban areas and, as such, the desire for large numbers of children is likely to be higher in the rural than in the urban contexts. In addition, the influence of normative values supporting high fertility are likely to be higher in the rural than in the urban areas because of the constellation of factors that help support high fertility. Therefore, the psychological cost of living in extended family environmentsin the rural areas is likely to be much higher than the psychological cost of living in extended family environments in the urban areas Those cultural norms and rules of behaviour are likely to have been altered by the forces of modernization in the urban place. Therefore, we expect women living in urban areas to have smaller desired family size than those in the rural areas.

Taste for children is measured using indicators of traditional pronatalist values. These are type of marriage and religion. Persons who are polygynously married are likely to be more pronatalist and, as such, to have higher desired family size than persons who are monogamously married. Religion also measures attachment to cultural practices that are supportive of large family sizes. Therefore, we expect Catholics and Protestants to desire smaller family size than women affiliated with traditional religion.

Two indicators of income are used in this study. These are women's educational attainment and household wealth. The latter measures physical wealth (see also Zick and Xiang, 1994). Education is used as a measure of human capital. It is also highly correlated with income. Household wealth is measured using ownership of durable goods. Six of such goods are measured: electricity, refrigerator, television, radio, bicycles, and type of toilet facility; flush or pit. Similar measures were used as indicators of household wealth by Zick et al. (1994) in China and by Speizer (1995) in a study conducted in three francophone African countries: Burkina Faso, Cameroon, and Niger. Zick and Xiang (1994) explained that the relationship between income and demand for children is not necessarily linear. An increase in income may not necessarily lead to an increase in demand for children because individuals may choose to invest in the quality of surviving children rather than have another child. Thus, it is important that researchers bear this alternative income effects in mind in applying this theory.

In addition to these dimensions, other socio-demographic controls are included in the analysis in order to understand the impact of the context and women's experiences on their fertility desires. These variables include ethnic affiliation, husband's desired family size, women's age, number of surviving children, infant/child mortality and knowledge of modern contraception. These variables have also been demonstrated by previous research to have significant effects on desired family size. For instance, women's perceptions of their husbands' fertility desires are likely to shape their own reproductive choices (Muvandi, 1995). In addition, it is important to include ethnic affiliation in a study such as this because it is a measure of the context of women's lives (Kritz, et al, 1997, Caldwell and Caldwell, 1887). It captures norms, beliefs, and perceptions which are difficult to measure in standard surveys, such as the ones analyzed in this study, but which, nonetheless, influence the individual's reproductive attitude and behaviour (Njogu, 1991). Furthermore, because ethnic groups in Kenya speak the same ethnic language, have similar cultural practices and live within the same or contiguous districts, the flow of ideas about small family sizes are likely to be faster within than across ethnic groups. Also, ethnic boundaries are co-terminus with political/administrative boundaries to a large extent in Kenya and, as such, access to education, health and infrastructural development is likely to vary by ethnicity. For example, the Kikuyu, who coincidentally live in the central province, are likely to have access to better infrastructure because they had an early access to colonial education and political power than did many other ethnic groups in the country. Therefore, ethnic affiliation also reflects, albeit indirectly, differentials in access to socio-economic development across the country.

Data

This paper uses data from the 1989 and 1993 KDHS standard recode files. Data are also obtained from published reports on the Kenya World Fertility Surveys (KWFS) and the Kenya Contraceptive Prevalence Survey (KCPS). These data sets are used in order to examine trends in the desired family size in the country. The 1993 standard recode file is used to explain the differentials and the key factors in desired family size because it is the most recent nationally representative data set on Kenya. The sample consists of 4,629 currently married women who are within the age of 15-49. The KDHS data set is weighted to correct the unequal probability of sampling among strata (National Council for Population and Development et al, 1994). That weight is applied to the estimates presented in this work.

Measurement of variables

The dependent variable analyzed in this paper is the desired family size. It is an interval variable measuring the number of children desired by individual women. Another dependent variable also used in this analysis is preference for no more children. Preference for no more children is constructed from responses to the following questions: those who were not pregnant or unsure were asked: "...Would you prefer to have a/another child or would you prefer not to have any more children?" The pregnant women were asked: "After the child you are expecting, would you like to have another child?" Desire for additional fertility is a dummy variable coded "1" if women said they wanted no more children and "0" otherwise. Although this is not a measure of desired family size, it indicates what women are willing to do in the immediate future, given their current fertility. If the result of the analysis of preference for additional fertility is consistent with findings on desired family size, it will enhance the reliability of the result. This is one major advantage of using multiple dependent variables enhanced reliability and validity of findings. However, this variable, desire for additional fertility, is not used in the multivariate section since the goal of that section is to examine differentials in desired family size.

The independent variable measured in this analysis include women's education, ownership of durable goods (used as a proxy for household wealth), women's work and income status, type of union, religion, husband's desired family size, place of residence, knowledge of contraception, ethnic affiliation, and women's age. Women's education was measured in years. Ownership of durable goods was constructed from the following questions: (1) "What kind of toilet facilities does your household have?", (2) "Does your household have: electricity, a radio, a television, a refrigerator", and (3) "Does any member of your household own a bicycle." Respondents received a score of "1" for any of these durable goods they have in their household and "0" otherwise. The final measure consists of four dummy variables: the first is coded "1" if a woman has none of these commodities; the second is coded "1" if a woman has one out of six; the third is coded "1" if a woman has two out of six and the fourth is coded "1" if a woman has 3 or more of the six items in her household and "0" otherwise. In the multivariate analysis, the first group is used as the comparison group. These six items are measured because they are luxury goods. Those who can afford them must have satisfied the first order needs: food, clothing and shelter. This measure, ownership of durable goods, in our opinion, is a reflection of household income.

Women's work and income status is a dummy variable coded "1" if a woman is currently working and earning income and "0" otherwise. Type of union is a dummy variable coded "1" if a woman is married to a polygynous husband and "0" otherwise. Religion is measured by two dummy variables: the first is coded "1" if the woman is Catholic; the second is coded "1" if the woman is Protestant and "0" if otherwise. The last group, the other category, is made up of persons who belong to other religions. This group is used as the reference category in the multivariate analysis. Knowledge of contraception is indicated by three dummy variables: "knows no method", "knows a traditional method" and "knows a modern method". A woman received code "1" for each of these variables if any of these conditions applied and "0" if otherwise.

Place of residence is coded "1" if urban and "0" if rural. Age is measured in years. Ethnic affiliation is measured using five dummy variables. The first is coded "1" if the woman is Kalenjin; the second is coded "1" if the woman is Kamba; the third is coded "1" if the woman is Kikuyu; the fourth is coded "1" if the woman is Luhya and the fifth is coded "1" if the woman is Luo. All others who do not belong to these groups are pooled into the "other category" because members in other groups are too few to constitute separate dummy categories.

Husband's desired family size indexes differed between the number of children women said their husbands desired and the number they themselves desired although it has been suggested that wives' responses to questions on husband's fertility, attitude and behaviour are likely to be less reliable than husbands' responses to questions about their own desired family size. Hence, husbands' report about their own desired family size should be used when those data are available. However, in this study, wives responses are used because our aim is to measure women's perception of their husbands' preferences and the effect of that perception on their own reproductive choices. These perceptions are likely to have significant effects on wives' behaviour since husbands' actual preferences may be unknown if spousal communication is rare. Another reason for using the wives' responses relates to data. The KDHS collected data on only 25% of the husbands. Therefore, the use of data on a couple could lead to a significant loss of information due to sample attrition. Moreover, a comparison of spousal responses to questions on fertility preferences indicates a great degree of agreement between wives' and husbands' responses. The final measures consist of three dummy variables: the first is coded "1" if the number of children desired by the husband is greater than the number desired by the wife; the second is coded "1" if the husband's desired family size is smaller than the wife's and the third is coded "1" if both husband and wife desire the same number of children. This last category is used as a reference category in the multivariate analysis. Some of the variables described above are included as controls in the multivariate context. Tables 1 and 2 present a descriptive analysis of these variables by desired family size.

The variables described above are used in the bivariate and multivariate analyses. Other variables included for the purpose of describing bivariate differentials in desired family size are age at first marriage, number of times married, literacy (used as an additional measure of education), number and composition of living children, number of children who have died and contraceptive use. Age at first marriage is measured in years. Number of times married is coded "1" if the woman has been married once and "0" if she has been married twice or more. Literacy is coded "1" if the woman says she can read and write easily and "0" otherwise. Number and composition of children are the actual number of sons and daughters alive at the time of the 1993 KDHS survey. Number of children who have died measures the actual mortality experience of women.

Two dimensions of contraceptive behaviour are measured: "ever" and "current" use. "Ever use" is indicated by three dummy variables. The first is coded "1" if the woman has never used any method; the second is coded "1" if she ever used a traditional method and the third is coded "1" if the woman ever used a modern method. Similarly, three dummy variables measure current contraceptive practices: currently using no method, currently using a traditional method and currently using a modern method. A woman receives code "1" if any of these conditions tended to be applicable. The categories are mutually exclusive. In other words, a woman who says she is currently using no method cannot say that she is also currently using a modern method. Although a woman may be using traditional and modern methods concurrently, such women are automatically coded as currently using a modern method, thus avoiding multiple scoring for the same respondent.

Method of analysis

Two goals are pursued in this paper. The first is to explain trends in desired family size and the second is to explain the differentials. The former is achieved by using a synthetic cohort analytical technique. This procedure permits a quantitative description of temporal variations in the experiences of synthetic cohorts (see also Halli and Rao, 1992). Using data from KWFS (1978), KCPS (1984), and KDHS (1989, 1993) surveys, we constructed synthetic cohort data sets for women aged between 15-19 and 30-34 in 1978.

To explain the differentials in desired family size, we used both bivariate and mutivariate procedures. The bivariate procedure consists of summary statistics such as percentages, means and standard deviations. Ordinary least square regression was used to verify differentials observed in the bivariate context. The results of these analyses are presented in Table 3.

Results Trends in desired family size

Figure 1 below traces changes in desired family size from 1978 to 1993 for cohorts 15-34. This Figure demonstrates a consistent and sharp decline in desired family size over the periods studied. For cohort 15-19 in 1978, the average desired family size declined from 6.5 in 1978 to 5.6 in 1984, with a difference of about 1 child. By 1989, the desired family size was 4.6 and it reached 4.0 by 1993. So over the 15-year period studied, we observed a decline of about 2.5 children in average desired family sizes, from a high of 6.5 in 1978 to a low of 4.0 in 1993 for this cohort of women.

For the cohort 20-24 in 1978, their desired family size declined from a high of 6.3 in 1978, 5.9 in 1984, 4.9 in 1989 to 4.1 in 1993. As in the cohort 15-19, we observed a decline of over 2 children in average desired family size for this cohort. The pattern of change was slightly different for older cohorts. For instance, the cohort 25-29 in 1978 experienced no decline in desired family size between 1978 and 1984. However, by 1989, the desired family size for this cohort had declined by 1.6 children and by 1993, the number of desired family size stood at 4.1., with a decline of about 2.2 children from the 1978 levels. The last cohort, 30-34 years old in 1978, also behaved like the previous one. The desired family size for this cohort was 7.2 in 1978. By 1984, it was 6.9, thus representing a very slight difference from the 1978 level. However, by 1989, the desired family size had declined to 5.5 children on average, with a decline of about 2 children over this interval. The next five-year period saw an additional decline of 1 child in desired family size so that by 1993, the desired family size had declined to a low of 4.4 children.

We also analyzed synthetic cohort differences in the percentage of women wanting no more children. The results are presented in Figure 2. Figure 2 presents data for cohorts 15-19, 20-24, 25-29, 30-34, and 35-39 years old in 1984. As shown in Figure 2, the percentage of women wanting no more children rose significantly from 3.8 in 1984 to 18.8 in 1989 and to 46.1 in 1993 for the cohort 15-19. By 1993, the percentage wanting no more children among this cohort had increased more than 12 times over the 1984 level.

Figure 2 also revealed very significant changes in demand for no more children for the cohort 20-24. The percentage wanting no more children increased from a low of 10.7 in 1984, 39.8 in 1989 to a high of 60.8 in 1993. Between 1984 and 1993, the percentage tripled and the 1989 figures doubled by 1993. Among women aged 25-29 in 1984, the percentage wanting no more children rose from 23.4 to 58.2 in 1989 and to 74.7% in 1993. Similar increases were recorded among women between 30-34 and 35-39 years old in 1984. Among this cohort, the percentage increased from 45 in 1984 to 70.9 in 1989 and to 80.6 in 1993. Among the 35-39 age group, the percentage of women wanting no more children increased from 53.7 percent to 84.7 in 1989. The percentage declined by about 8 points between 1989 and 1993, from 84.7 to 76.7. Except for this cohort, the percentage wanting no more children increased monotonically during the periods studied. These increases were

consistent with the changes observed in desired family size, thus indicating a great deal of consistency in women's fertility attitude during the period.

It is plausible that a sequence of attitudial changes beginning with a transition from high DFS to low DFS is a precursor to fertility transition in Kenya. This seems to be borne out by data from different time periods presented in Figure 3. The observed declines in DFS over the past 15 years, and especially between 1989 and 1993, are closely correlated with declines in actual fertility for every synthetic cohort. This is a clear indication that changes in desired family size can and do affect changes

in fertility. Both the TFR and desired family size started from a very high level in 1978 and declined monotonically toward low levels in 1993, thus indicating that observed declines in fertility are driven by the changes in desired family size.

Differentials in desired family size

Tables 1 and 2 present differentials in desired family size in 1993 by selected background characteristics of the respondents. The dependent variable is divided into three categories: (1) those who gave non-numeric responses, (2) those who said they wanted three children or fewer, and (3) those who said they wanted four to six children or more. According to Table 1 (row 1), only 6.1% of the respondents gave non-numeric responses to the question or desired family size. This finding indicates that questions on desired family size are not only meaningful to women but also that the idea of numeracy about children has evolved, an attribute which Van de Walle (1992) said is a precursor to fertility transition. 36% of the sample opted for three or fewer children and 58%, for 4-6 or more children. Given the high percentage of women wanting 4-6 children or more, Kenya is still a pronatalist country. Compared to other countries in the region, however, they are far ahead in terms of becoming numerate about their desired family size.

Column 2 of Table 1 describes women who gave non-numeric answers by selected background characteristics. According to Table 1, women who own no durable goods, live in rural areas, or who are not literate are more likely to give non-numeric responses than women who own one or more durable goods, can read and write, or who are urban. The percentage of non-numeric responses is also higher among women in polygynous unions, who knew no method of family planning and never used any contraception than it is among women who knew a method of family planning, ever used one, or who are monogamously married. The percentage is also higher among women who did not know the husbands' desired family size than among those who knew it. Women who gave non-numeric responses were also older (their mean age is 34 years whereas the mean age in the sample is 31 and had larger families than those who gave numeric answers (Table 2). These findings indicate that women who gave non-numeric answers were more traditional and older than an average woman in the sample. Therefore, this percentage can be expected to decrease as the population in place ages out and women became more numerate as they become younger and better informed about population issues.

Table 1: Desired Family Size by Selected Background Characteristics, Currently Married Women* (Percentage), 1993

Source: Kenya DHS, 1993

* Figures in parentheses are Standard Deviations

Table 2: Desired Family Size by Selected Background Characteristics, Currently Married Women* (Means), 1993.

Columns 3 and 4 of Table 2 compare women who wanted three or fewer children with those who chose 4-6 or more children by selected socio-economic and demographic characteristics. Women who wanted 3 or fewer children were younger, married later and had lower actual fertility than those who opted for larger families. The percentage currently using contraception is higher among women who chose 0-3 (51%) children than among those who wanted larger numbers of children (46.3%), but the "ever use" rates are equal among both groups of women (Table 1). Although the difference in "current use" rates in these two samples is not large and it does indicate that the discontinuation rate is higher among women who demand large numbers of children and that women who wanted smaller families are likely to be more consistent contraceptive users than those opting for a large family size. Women who stood for three or fewer children are better educated or are married to husbands with higher education, wealthier and more urban than other categories of women. These results indicate that the modernizing effects of education and urban residence are correlated with the desire for smaller families.

We also examined the effect of husband-desired family size on wive's reproductive preferences. Table 1 reveals that, overall, the percentage opting for 4-6 or more children is larger than the one for three children or fewer among the women who did not know the husband's desired family size or who said their husbands wanted more children they expected. In other words, women tend to accept smaller families when their husbands want fewer or the same number of children as their wives; they are least likely to opt for smaller families when they do not know their husbands' fertility preference.

The results of the multivariate analysis are presented in Table 3. The model explains 18.6% of the variance in desired family size (p<.001). All the independent variables have significant effects on desired family size. The exceptions are age, type of union and the child mortality indicator (as described earlier). We observed no significant differences in desired family size among polygynous and monogamously married women. Thus, our hypothesis that polygynous women are more likely to display pronatalist orientations and, as such, are more likely to have higher desired family size than their counterparts in monogamous marriages is not supported by data. Polygynous women should be more likely to compete with co-wives for their husbands' affection and raising large numbers of children is perceived as one way of winning their husbands' love (see also Mason, 1993, 1987; Cain, 1993). Although the demographic literature indicates that polygynous women tend to have lower completed fertility than monogamous women (Ascadi and Ascadi, 1990), this is due to their relatively lower exposure to intercourse rather than a lower desire for children. However, our finding could be due to the fact that most polygynous marriages are contracted for economic reasons and, as such, polygynous women may not be significantly different from their counterparts in monogamous unions in the extent to which they possess pronatalist values. Additionally, the monogamous status is temporary in that currently monogamous man could take another wife as soon as he was economically capable of doing so (see also Speizer, 1995). Similarly, age has no significant effect on desired family size among the women, but the relationship is positive, thus indicating that the older the women, the larger the number of children desired. Also, the effect of child mortality experienced by women is positive but not significant.

However, education and household wealth (both of which are proxies for women's income) are negatively related to desired family size, as expected. An additional year of education reduces desired family size by .08 (p<.01) children. Similarly, the impact of household status is large and significant. Women who own one or two of the durable goods wanted .27 and .31 fewer children respectively, than those who have none of the items measured. Those who own 3 or all six goods measured wanted .34 fewer children than those who own none of the six durable goods in their households. These differentials are statistically significant at .01 alpha level (see Table 3). Therefore, the more goods individuals have in their households, the fewer the number of children desired. The classical economic hypothesis that an increase in income should lead to an increase in demand for children is not supported by these data. However, a variant of this hypothesis which states that an increase in income could cause individuals to increase investment in child quality rather than quantity is supported by the findings of this research. Indeed, although it has not been documented, it is plausible that the desire to educate children and increase their access to opportunities is a major force driving the engine of the fertility transition in Kenya. Parents want their children to have a better life than they themselves have. Land used to be a major endowment parents passed on to their children to guarantee a better life for them in the past (Mburugu, 1994, Mloyi, 1992). But, given the declines in land holdings and changes in the land tenure system in Kenya, parents are increasingly choosing to invest in their children's quality by giving them better education (see also Mburugu, 1994, 1996). Although not directly tested in this study, scholars have suggested a strong association between land shortage and parental educational aspiration for children in Kenya. For instance, Mloyi (1994) argued that in areas of considerable land shortages,"... education may have replaced land as an inter-generational status transfer" (pp. 8). Therefore the motivation to educate children could account for the preference for smaller family sizes.

Table 3 Parameter Estimates of the OLS Regression of Number of Children Desired on Indicator Variable for Currently Married Women.

Source: Kenya DHS, 1993, * Significant at .05 alpha level, **Significant at .01 alpha level + This excludes the 281 women who gave non-numeric answers.

Furthermore, the analysis revealed that Catholics and Protestants desire significantly fewer children than women affiliated with other traditional religions. Catholic and Protestant women wanted .29 and .33 (p<.01) fewer children, respectively, than other categories of women. The magnitude of the effect is greater among Protestants probably because Catholics tend to subscribe more to pronatalist values than Protestants. These differentials in desired family size by religion are consistent with our expectation that taste is significantly shaped by religion and that persons affiliated with other more traditional religions will be more likely to opt for larger numbers of children than those who are Catholics or Protestants.

Place of residence and women's work and income status are used as proxies for the opportunity cost of time spent producing children. As expected, these variables have significant inverse effects on DFS. Women who live in the urban areas or who work and earn income want 0.51 and .07 fewer children, respectively, than other categories of women.

Women's desired family size is significantly affected by their husband's reproductive preferences. When wives believed their husbands want more children than they do, they demanded 0.14 (p<.05) more children than their counterparts whose husbands wanted fewer or the same number of children as they did. The greater effect, however, is observed among women who do not know their husbands' desired family size. These women want .20 (p<.01) more children than wives whose husbands want fewer or the same number of children as envisaged by them. This result underscores the importance of spousal communication about children. Communication increases wives' understanding of husbands' preferences regarding the number of children to have. This analysis indicates that such an understanding could have significant effects on wives' reproductive choices.

Table 3 also shows that knowledge of modern contraception is significantly related to desired family size. Women who report that they know a modern method of contraception want .41 (p<.01) fewer children than their counterparts who do not have any such knowledge. The study revealed further that, apart from women's socio-economic status, perceptions of husband's preferences and other demographic factors, ethnicity has significant effects on DFS. Five ethnic dummies are included in the analysis as shown in Table 3. These groups are compared with the Luo (used as the reference group in this study). The study revealed that DFS is smaller among Kamba, Kikuyu, Luhya, and women from other ethnic groups than the Luo. The difference between the Kalenjin and the Luo is not significant. However, compared to other ethnic groups in the country, the Luo appear to have a higher demand for children. The strong preference for large family sizes probably accounts for the persistently high fertility and limited contraceptive practice observed among the Luo by other scholars (see Ndeti, 1988; Watkins, 1995; National Council for Population and Development et al., 1994). The ethnic differential in desired family size is probably attributable to fundamental differences in the prevalence of cultural forms supportive of high fertility in each group and in access to socio-economic opportunities. For instance, the early access to colonial education could have set the Kikuyu and the Kamba ahead of other ethnic groups and, as this study shows, of all the groups studied, the Kikuyu and the Kamba have the least preference for large families. The Kikuyu, for instance, live in the Central Province which has better infrastructure, education, and health facilities than other parts of the country.

Discussion and Conclusion

Studies of both trends and differentials in DFS are limited in Kenya. As such, there is neither an adequate understanding of the relationship between DFS and actual fertility in the country nor of the determinants of DFS. It is this gap that this study attempted to bridge.

The study began with the premise that changes in demand should lead to changes in the supply of children, all things being equal. Using the Easterlin framework, we hypothesized that attitudinal changes in desired family size should be related to changes in income, the opportunity cost of alternatives forgone and the taste for children. In order to understand these changes, we examined both the trend and differentials in DFS. The trend analysis was performed with cross-sectional data collected over a 15-year period for synthetic cohorts of women. The results show a significant and consistent decline in the desired family size by actual fertility across the years. In particular, over the 15-year period studied, a decline of about 3.3 children in desired family size was observed, from a high of 7.2 children in 1978 to a low of 3.9 children in 1993 for all women. The analysis of synthetic cohort differences in preferences for additional fertility also revealed significant increases in the percentage of women wanting no more children between 1984 and 1993 for cohorts 15-19 to 35-39. Further analysis revealed that the decline in DFS is highly correlated with declines in actual family size over this period.

The multivariate analysis revealed significant differentials in desired family size influenced by women's education, household wealth, women's work and income status, place of residence and religion. Better educated women who own one or more durable goods, work and earn income, live in urban areas and who are either Protestants or Catholics want fewer children than women with no income, limited or no education, who live in rural areas, and are affiliated with other religions. The study also found that women's perceptions of husband's ideal family size significantly affect their own DFS. Women who do not know their husbands' desired family size are significantly more likely to demand large numbers of children than women who know. This finding indicates that husbands' preferences do affect wives' reproductive preferences. As such, an understanding of the preference structure among men could provide useful insights in developing IEC programmes to change women's reproductive behaviour.

Knowledge of modern contraception also accounts for significant variations in DFS: women who know a modern method of contraception demand fewer children than those who know no modern method of contraception. These findings indicate that the on-going fertility transition in Kenya is due both to changes in reproductive preferences and increases in the implementation of those "new" preferences. Similar conclusions were drawn earlier by Bongaarts (1991) from an analysis of data on 18 developing countries.

The study also revealed significant ethnic differences in DFS: unlike the Luo women, the Kamba, the Kikuyu, the Luhya and the Kalenjin, to some extent, and women from other ethnic groups want smaller families. We argued that differences in fertility orientation, access to education and opportunities in each group could account for the observed ethnic variation in DFS.

Of all the variables examined, knowledge of modern contraception and place of residence explained most of the variance in DFS. This implies that, in addition to income, cost-benefit constraints, access to information about family planning and modern ideas, as provided in the urban context, lead individuals to re-evaluate their family size values and modify them accordingly. This fact probably accounts for why the TFR declined almost at the same rate as the DFS over the 15-year period studied. These findings indicate that policies oriented toward changing reproductive attitudes and behaviour among women should improve their socio-economic position, their access to education and information about family planning and small family size values. Such policies could significantly decrease the demand for children. Furthermore, the fertility transition that is currently under way in Kenya is likely to be sustained by policies which guarantee the continued modernization of values related to children as well as the provision of family planning services that enable women to implement their reproductive choices.

However, the hypothesis that changes in ideas precede changes in behaviour is not directly tested in this work. To do this, we need to model the change in DFS between Time 1 and Time 2 on independent variables measured at Time 1 or on changes in the independent variables between the two time periods. Such an analysis requires either a time series or longitudinal data sets. Neither of these data sets are available for Kenya. Future work on this topic should focus on generating such data sets for the country.

 Acknowledgements

The authors would like to acknowledge the support of the Population Council in providing the fellowship under which this paper was prepared. They are also grateful to Anrudh Jain, Susan Watkins, Cynthia Lloyd, Cheikh Mbacke, John Kekovole, Naomi Rutenberg and Evasius Bauni for helpful comments in improving this paper. They also thank Cheikh Mbacke for assisting with the Kenya DHS data.

References

• Ahmad, B. 1990, "Determinants of desired family size in rural Bangladesh: a two-stage analysis". Journal of Family Welfare 36 (1):22-31
• Ascadi T.F. and G.J. Ascadi, 1990. "Demand for children and for child-spacing". In Population and reproduction in sub-Saharan Africa, edited by TF Ascadi, GJ Ascadi & Rodolfo A. Bulatao, Pp 155-185, Washington, DC: The World Bank.
• Ascadi T.F., G.J. Ascadi and Rodolfo A. Bulatao, 1990. "The context and causes of high fertility". Pp 131-132 in Population growth and reproduction in sub-Saharan Africa: Technical analysis of fertility and its consequences, edited by T.F. Ascadi, G.J. Ascadi and R.A. Bulatao, The World Bank, Washington, D.C.
• Bankole, A. and C.F. Westoff, 1995. Child-bearing attitudes and intention. DHS Comparative Studies No 17. Calverton, Maryland: Macro International Inc.
• Becker, S. 1981. A Treatise on the Family. Cambridge, Massachussets/London, England: Harvard University Press.
• Benefo, K. 1990. The determinants of family size preferences and traditional child-spacing practices in West-Africa. Doctoral dissertation. Ann Arbor, University of Michigan.
• Bhuyan, S.B. and K.C. Bhuyan, 1985. "Fertility preference and family planning in rural Bangladesh". The Journal of Family Welfare; XXII (September):59-69.
• Blake J. 1986. "Are babies consumer durable? "Population Studies. 22:5-25.
• Bongaarts, J. 1991. "Do Reproductive Intention Matter?" Paperpresented at the DHS World Conference, Washington, D.C., Proceedings, Vol.1.
• ------1992. "The supply-demand framework of the determinants of fertility: An alternative implementation". Population Council, Research Division, Working Papers Series, No 44.
• ------1995. "The role of family planning programs in contemporary fertility transitions". The Population Council, Research Division Working Paper Series No. 71. New York, New York.
• Bushan, I. and K. Hill 1995. "The measurement and interpretation of desired fertility". Paper presented at the Annual Meetings of the Population Association of America, San Francisco.
• Caldwell, C. and P. Caldwell. 1987. "The cultural context of high fertility in sub-Saharan Africa". Population and Development Review 13(3):409-37.
• Campbell, E.K. 1991. "Sex preference of offspring among men in Western Sierra Leone". Journal of Biosocial Science. 23:337-342.
• Cochrane et al. 1990. "Education, income and fertility in Egypt: a revised perspective. "Economic Development and Cultural Change.
• Farooq G.M. et al. 1987. "Family preference and fertility in South-western Nigeria. "In Sex roles, population and development in West Africa. Edited by C. Oppong.
• Freedman, R. et al. 1974. "Do Statements about Desired Family Size Predict Fertility? The case of Taiwan, 1967-1970". Taiwan Population Studies, Working Paper No 27.
• Gray, and D.K. Morgan. 1976. "Desired family size and sex of children". The Journal of Heredity; 67:4(July/August):319-321.
• Halli, S.S. and D.V. Rao. 1992. Advanced techniques of population analysis. The Plenum series on Demographic Methods and Population Analysis. London, New York: Plenum Press.
• Hardee-Cleveland, K. 1988. Desired family size and sex preference in rural China. Evidence from Fujian Province. Doctoral Dissertation. Cornell University, Ithaca, New York.
• Kelly, A.C. and C. Nobbe (1990) "Kenya at the demographic turning point?: Hypotheses and a proposed research agenda". World Bank Discussion Paper 107. Washington D.C.: World Bank.
• Kent, M. and A. Larson. 1982. "Family size preferences: Evidence from the World Fertility Surveys". Population. Reports on the World Fertility Surveys No. 4, Washington, D.C.
• Knodel, J. and V. Prachnabmoh. 1977. "Desired Family Size in Thailand: Are the responses meaningful?" Family Planning Résumé 1 (1):32)43.
• Lightbourne; R.E. 1987. "Reproductive preference and behaviour". Pp 838-61, in The World Fertility Survey, an Assessment, edited by J. Cleland and C. Scott. Oxford: Clerendon Press.
• Mburugu, E. K., 1994. "The persistence of high fertility in Africa and prospects for fertility decline". in African Perspectives on Development. U. Himmelstrand, K. Kinyanjui and E. Mburugu; ed. London: James Currey, pp. 74-83.
• K. 1996. "A Research agenda for a proposed demographic research programme based in Kilifi District". Document prepared for the African Population Policy Research Centre, Population Council, Nairobi, Kenya.
• McCarthy, J and G.A. Oni. 1987. "Desired Family Size and its Determinants among Urban Nigerian Women: A Two-stage Analysis". Demography. 1987 May:24(2):279-90.
• Mhloyi M. 1992. "Changing factors affecting fertility decision in Africa". Third African Population Conference, Meeting of Experts, Dakar 7-10 December. UN and ECA, POP/APC. 3/92/Inf.6.
• -----1994. "Fertility transition in Zimbabwe", In The Onset of fertility transition in sub-Saharan Africa, edited by T. Locoh and V. Hertrich, p.p., Liège, Beligium: IUSSP.
• Mousa M. A. 1987. "Validity and consistency aspects in the quality of responses regarding desired family size in Sudan". Studies in African and Asian demography. Research monograph 16:259-81.
• Musyoki R.N. 1983. Education and desired family size: a study of Kenyan youth. Doctoral Dissertation. Tallahassee, Florida. Florida State University.
• National Council for Population and Development et al. 1994. Kenya Demographic and Health Survey 1993. Calverton, Maryland: NCPD, CBS, and MI.
• Njogu, W. (1991) "Trends and determinants of contraceptive use in Kenya". Demography 28, 1(February):83-99).
• Palmore J.A. and M.B. Conception. 1981. "Desired Family size and contraceptive use: An 11-country comparison". International Family Planning Perspective; Mar:7(1)37-40.
• Pullum T. 1981. "Fertility preference and contraceptive orientation the Philippines". Paper presented at the IUSSP Committee for the Analysis of Family Planning.
• Pust R.E. et al. 1985. "Factors affecting desired family size among preliterate New Guinea mothers". International Journal of Gynecology and Obstetrics, 23(5):413)20.
• Rao V.V. and N. Rao. 1981. "Family size and sex preference of children: a bi-racial comparison". Adolescence: XVI(62):385)401.
• Rasul, A. 1993. "Fertility preference: A study of some basic concepts and considerations". The journal of family welfare; 39(1).
• Riley, A.P. et al. 1993. "A new look at the determinants on non-numeric response to desired family size: the case of Costa Rica". Demography 30(2):159-74.
• Smith M.O. 1993. Desired family size and desired child education among recently married women in Egypt. Doctoral dissertation. Baltimore: Johns Hopkins University, Maryland.
• Speizer I. (1995) "New approaches to men and marriage in Francophone Africa". Paper presented at the Annual Meetings of the Population Association of America, San Francisco, May 6-8.
• Speizer I. and H. Mosley. 1995. "Men's desire for additional wives and children". Paper presented at the Annual Meetings of the Population Association of America, San Francisco.
• Stash, S. 1995. "Ideal Family Size and Sex Composition among Wives and Husbands in Chitwan, Nepal: The introduction of a multiple response fertility preference scale". Paper presented at the Annual Meetings of the Population Association of America, San Francisco.
• Van de Walle, E. and A. Foster. 1990. "Fertility decline in Africa: Assessments and prospects". Technical Paper No. 125. Africa Technical Department Series. The World Bank, Washington, D.C.
• ------1992. "Fertility transition, conscious choice and numeracy". Demography; 29(4): 487-502.
• Ware H. 1974. "Ideal Family size". Occasional Papers, No. 13. London: The World Fertility Surveys.
• Watkins, S.C. Naomi Rutenberg and S. Green (1995). "Diffusion and debate: controversy about reproductive change in Nyanza Province, Kenya". Paper presented at the Annual Meetings of the Population Association of America, San Francisco, April 6-8.
• Westoff, C.W. et al. 1961. "Desired Family Size". In Family Growth in Metropolitan America, edited by C.W. Westoff et al., Pp 136-148. New Jersey: Princeton University Press.
• Zehri K. et al. 1988. "Age at marriage and desired family size". In Médecine et Santé des Femmes, edited by L. Labidi, Tunis, Tunisie: Premières Journées Aziza Othmana.
• Zick C.D. and N. Xiang. 1994. "Assessing the correlates of desired family size in Shaanxi, China". Social Biology 41(3-4):150-167.

NOTES

• WF is based on assumptions regarding the wantedness of children. A child is considered wanted if a woman/couple reported wanting another birth at the time of the survey. On the basis of this information, "want more" age specific fertility rates (WMASFR) are calculated.
• The PDTFR proposed by Bushan et al. (1995) is constructed from responses to questions on the desire for additional fertility. The DPTFR is estimated from births which will occur in the 12 months following the survey assuming that women have the births they wanted. The numerator includes all births to non pregnant women who said they wanted a/another birth within 12 months after the survey, all wanted current pregnancies which could result in wanted births within the 12 months after the survey plus some adjustment for women not yet married who would marry in the next three months after the survey. The denominator covers all women of reproductive age (Bushan and Hill, 1995).
• Place of residence was used as an indicator of price by Ahmed (1984) in the application of the Easterlin model to data from Bangladesh.
• See page 2 for a detailed discussion of how the variable was derived.
• Women's work and income status is constructed from two sets of questions. In the first set of questions, women were asked whether they were engaged in any type of work: family business/farm, small retail trade (whether within or outside the home environment), formal employment, etc. Those who responded "yes" were then asked if they received cash (any amount) for this work. The indicator of women's work and income status is coded "1" if they said "yes" they work and earn cash, and "0" otherwise.
• A comparison of spousal responses to questions on reproductive preferences among 25% of the women whose husbands were interviewed indicate that there is considerable agreement between wife's and husband's responses. With regard to desire for additional fertility, for instance, among 30% of the spouses, both spouses said they wanted more children; among 28% both said they wanted no more. The proportion of couples in which the husband wanted more children than the wife is greater (12%)than the proportion in which the wife wanted more and the husband did not (7%). Among 22% of the respondents, one or both of the spouses appeared to be undecided about whether they wanted to have more children. This general high degree of agreement between spouses, according to the KDHS report, is consistent, regardless of parity. Similarly, the analysis of spousal perception of each other's attitude to FP also reveals a high degree of accuracy in the husbands' and wives' report of their spouses' attitude. For instance, in 94% of the cases in which wives reported that their husbands approved of FP, the husbands also said they approved (NCPD et al., 1993:162).
• This analysis is restricted to KCPS and DHS data sets. Data on desire for additional fertility were not collected in the Kenya WFS.
• Non-numeric responses are those who said that they did not know their desired family size or that it is up to God or chance.
• Number of surviving children was not included in the model because of high multicolinearity problems. Sex composition of children was included at first but was later removed for lack of significance and, more importantly, because its inclusion reduces the amount of variance explained.

Copyright 1997 - Union for African Population Studies.