Fertility , Health and Female Labour Force Participation in Urban Cameroon

Many studies report empirical relationship either between fertility and labour supply or, between health and labour market outcomes. In this paper, an extension of these ideas involves explicitly considering how fertility and health affect each other, and how they interrelate with labour force participation. A unifying framework is provided and a simultaneous three equations model developed to capture the interdependence between these variables as well as their respective determinants. The model is estimated using a cross-section data set obtained from a survey of the urban Cameroon population. The results indicate that: (i) fertility and health status are significantly interrelated, thus separate estimations of fertility (or health status) and participation will produce misleading results; (ii) working in either sector of the labour market significantly reduces fertility but, unlike many previous studies, fertility has a positive impact on the probability of labour force participation; (iii) there is strong evidence that health and disability status is a significant determinant of employment, but the reverse depend on the labour market sector and on the health indicator used.


Introduction
Since the pioneering work by Mincer (1962) and Cain (1966), there have been numerous studies on female labour force participation.These studies highlight that women's labour supply depends upon economic and demographic characteristics such as female earnings, male earnings, non-earnings income, schooling, age and the number of children (see Hill, 1983 for a review).Some of the studies have documented strong ties between women's work patterns and changes in their family status (Ellingsaeter and Rønsen, 1996;Rosenfeld, 1996;Rindfuss et al. 1999).These results are based on the evidence that women who work for pay have fewer children (on average) than women who do not, and that mothers spend less time in paid employment (on average) than childless women.
As the body of literature on labour supply grew, models were implemented to demonstrate that health was one of human capital characteristics (Scheffler and Iden, 1974;Bartel and Taubman, 1979;Parsons, 1980) and as such, health status variables influence labour force decisions; therefore, models could be refined in order to include these variables.There is extensive literature that demonstrates a positive relationship between health and economic prosperity (Marmot et al., 1991).Using self-reported measures, Bound et al. (1996) found that health has positive and significant effects on labour force participation; they demonstrated that the lower labour force participation rates of blacks in the United States (relative to whites counterparts) can be explained by differences in health status.
While there is evidence that fertility is endogenous to labour force decisions (Cramer, 1980;Mroz, 1987;Angrist and Evans, 1998), Stern (1989) and Leung and Wong (2002) documented the fact that health and labour force participation are interrelated (see also Haveman et al., 1989;Lavy et al., 1995).Yet, most of these studies treat these relationships separately (I mean fertility-labour supply, and health-labour supply) and ignore the well documented influence of fertility and health on each other (see Merchant and Martorell, 1988;Adair and Popkin, 1992;Miller et al. 1994).I argue that failing to account for this relationship may have led to biased estimates of the impact of fertility and health on labour force participation.
The objective of this study is thus to determine the impact of fertility and health on labour force participation in a simultaneous equations framework.The next section reviews the literature, followed by the methodology and the results.The final section discusses the results and concludes.

Literature review.
Over the life cycle, female labour force behaviour is governed by various factors.Very complex mechanisms determine the decision to enter, stay on, or leave the labour market (Lelièvre-Gauthier, 1994).These include economic (that is the labour market structure), individual (skills, marital status, labour force attachment, incentives and career expectations), and household characteristics (structure, domestic workload, presence and number of children).Many studies aimed at analysing trends (Chase, 1995;Bonin and Euwals, 2002), economic and social determinants of labour force decision (Benjamin, 1992;Hausman, 1980;Hill 1988Hill , 1994;;Saget, 1999;Fong and Lokshin, 2000) either at micro or macro level (See Griliches and Intriligator 1986 for a review).Here, we concentrate on the influence of individual and household characteristics on female labour force participation.
Economists and demographers have been interested on the impact of the number of children on how likely is a woman to go out to work, and if she does go out to work, for how many hours (Iacovou, 2001).One may argue that a woman with more children will be less inclined to go out to work, since the time she spends at work will be time foregone with her children, and the expense of childcare will reduce her effective wage.On the other hand, children are extremely expensive, and a mother may have to work more with every additional child to maintain the family income.As a consequence, there is no reason to believe a priori that the effect should go in either direction (or be positive or negative).Despite these arguments, Weller (1977) and Cramer (1980) gave four possible explanations of the association between fertility and female labour force participation: women's fertility influences their labour force behaviour; women's labour force behaviour influences their fertility; a reciprocal relationship exists between the two variables; and the association is spurious, reflecting other factors.If some evidence supports the first three hypotheses, the fourth one has proven more difficult to support.Furthermore, most estimates of this relationship have found a negative relationship between the number of children and a woman's labour supply (see Brewster and Rindfuss, 2000 for a review).The problem with these estimates is that they cannot say anything about causality.Following Grosman's arguments that "… health can be viewed as a durable capital stock that produces an output of health time" (Grossman, 1972, p.223), it became interesting to query the relationship between an individual's health status and his work behavior.These arguments led to an increased interest among researchers about the impact of health status on an individual's productivity and about the interaction between health and labour market decisions and outcomes.(Note 1).In a study of disability and labour force participation (LFP), Stern (1989) found that disability lowers the probability of LFP, but LFP increases the probability of disability.The finding on the effect of employment on health is not as unambiguous.Ekerdt et al. (1983) discussed the ambiguity concerning whether work improves or deteriorates health.Self-esteem, identity, and personal fulfilments from supplying labour efforts improve health.However, work pressure or poor working environment worsens health.Ross and Mirowsky (1995) found that health is protected by employment and improvements in health increase the probability of employment.Still, most of these studies have ignored the interrelationship between fertility and health status.Velkoff and Adlakha (1998) stressed the fact that female health problems in India are related to or exacerbated by high levels of fertility.Jejeebhoy and Rao (1995) showed that numerous pregnancies and closely spaced births increase health risks for mothers.Unwanted pregnancies terminated by unsafe abortions, diseases like malaria, HIV/AIDS, tuberculosis, also have negative consequences for women's health.Through the negative effect of poor health on birth outcomes, health status affects fertility.In the context of high fertility rates such as in Cameroon, (Note 2), women are exposed to much fertility related health problems and health matters are likely to determine their fertility choices.Thus, any attempt to determine the contributions of these to variables to female LFP must bear in mind their interactions.

Data source and sample characteristics.
This study seeks to evaluate the impact of fertility and health on female labour supply in urban Cameroon.Information was collected from 2,096 women aged 18 to 64 years, surveyed in Yaoundé and Douala.Many steps were followed in order to come out with this sample.The first step consisted of estimating the optimal size of the sample of women to survey in each city.
The choice of the sample size was guided by the rule of thumb suggested in sampling theory (see for example Mood et al., 1974) , where n represents the sample size, p the proportion of individual exposed to the phenomenon studied (in our case, women at the labour force age in urban Cameroon), ET the acceptable error, and 1.96 expressing the choice of a 95% confidence interval.In urban Cameroon, the proportion of women in the labour force age is 50.1% (INS/DSCN, 2001); assuming an acceptable error of 3%, the estimated sample was 1,086 women to survey in each city.We also had to take into account the possibility of incomplete information; due to budget constraints, we supposed we could end up with 5% incomplete questionnaires in each city.This led us to a total of 2,242 women to survey (that is 1,121 in each town).At the end of the survey, some observations with incomplete information were deleted and we ended up with 2,096 observations.
The remaining sample for the analysis comprised (see There are a number of concerns with such a measure (Bound, 1991;Tessier and Wolf, 2005).It may not be entirely comparable across respondents, it may not be independent of labour market outcomes, or respondents out of the labour market may mention health limitations to rationalize their behaviour.Such a health measure, therefore, is endogenous to the labour force status and each of the problems just listed may lead to a different kind of bias (Bound, 1991) (Note 4).Despite these and other concerns, this measure is still the most popular measure of health available.Tausman and Rosen (1982) even argued that this measure is close to the "objective" health.But, in seeking for appropriate ways to measure the relationship between women's health and labour force participation, I develop another measure of health status.
The dataset includes respondents' reports of specific health limitations; they were asked if they experienced difficulties with seeing, walking long distances, hearing, or standing long hours.Following the work by Dumont (1999), I define the second measure of health, the Composite Index of disability (CID) as: Where, in SRH we have the self-rated health and reports of health limitations (I mean answers to the question whether or not they had problems with seeing, walking long distances, hearing, or standing long hours).To construct our index of disability, I reversed the codes of all the measures in order to express poor health.(Note 5).
After the summation of these indicators, the obtained index ranges from 0 to 7.This index measures the potential of disability involved in each woman of the sample.It appears from Table 1 that 36.6% of working women have a potential of disability greater or equal to 4, while the same group is composed of 37.7% of non-working women.Then a woman is classified of as "Disabled" if her composite index of disability is greater or equal to four (4), that is: The most notable difference between developed and developing countries labour markets concerns the economic opportunities available to job seekers.Unlike developed countries where almost all the workers are employees, labour markets in developing countries were first characterised by dualism (see Ranis, 1988 for a survey on dualism) and the last two decades have been marked by the emergence of the informal sector.
In Cameroon, the urban labour market is characterised by two homogeneous sectors (public and formal private) and a non-homogeneous one, the informal sector (Abessolo, 2001).The informal sector comprises self-employed, unpaid family workers and casual workers with reduced job security, hazardous working conditions, and dangerous work environments.Factors determining labour market decisions and outcomes are different from one sector to another.Thus, instead of the usual dichotomy "to work or not to work" observed in industrialised countries and used in many studies, this study generalizes the standard labour force participation model by expanding the set of alternatives to four: working either in the public sector, in the formal private sector, in the informal sector, or not working.
Let us assume that preferences are described by a well-behaved utility function, the maximum utility attainable by individual i if she chooses the participation status j (j=p, fp, inf, np), (Note 6), and that this indirect utility function is composed of stochastic ( ) ji ε and non-stochastic ( ) ji S components, the indirect utility function is given by Equation 4and the probability ji P for individual i to choose alternative j is given by Equation 5:

Model estimation
The main objective of this study is to determine the influence of fertility and health status on female labour force participation.It explores the contextual factors across various sectors of the Cameroonian labour market using two indicators of health status.We consider the following simultaneous equations model: The estimation procedure of Equation 6 involves two steps.In the first step, we estimate reduced form equations (see Equation 7).

Description of variables
Female labour force participation is the main focus of this study, with special emphasis on the impact of fertility and health.The information in Table 3 displays a list of variables used in the estimations.Some descriptive statistics of the variables are presented in Table 4.The mean and the standard deviation of fertility are 2.59 and 2.38.A group mean comparison test between working women and those not working shows that working women have more children than those who do not work.The mean and standard deviation of self-reported health are 1.76 and 0.92 respectively.Hence, on average the respondents report a fair health condition.(Note 7).The average age is 33.038 years, hence our sample is much younger than (closer to) the ones studied in the literature on health status (fertility) and labour supply.For example, the mean age of the sample studied in Leung and Wong (2002) was 43.1 years, (Note 8), while in the study of Iacovou (2001) the mean ages of the samples were 33 and 35 years.We offer here a brief discussion of some possible relationships between the variables and our dependent variables.
Province: A well-known result in demography states that people from regions with high rates of fertility tend to reproduce the same habits no matter the place they live (urban or rural areas), even if they have migrated to other regions (Locoh, 1988).In Cameroon, people from the northern part and from west province have a high propensity to fertility related behaviour.
While Deaths is meant to capture the replacement hypothesis, Relatives and Child-non-charge capture respectively the idea that procreation is influenced by the household structure, and the extent to which own-child rearing fees are supported by somebody else (mainly relatives).This practice is common in Cameroon.
Education is expected to exert a positive effect on health at least in two ways.First, it improves health by increasing knowledge and efficiency in the production of health capital.Second, more education may imply higher willingness to invest in long-term capital including health capital.
A problem (14 days) controls for short-term shocks which may affect perception of health.Since health is naturally determined, then after controlling for long-term disease, other behavioural variables (such as age and education, to name few) may not determine health.

Results
The estimation results are presented in Tables 5 through 10.These results are obtained by using a 2SLS procedure.Two different specifications of the health equation are used and presented in all these tables.Specification (1) corresponds to estimates obtained using Sub-health, the self-reported health and (2) is obtained using the index of disability (Disabled).

The Fertility equation
Fertility is a count data variable and due to over-dispersion estimates is obtained using a negative binomial as suggested by Winkelmann (1997).We used the Consistent Akaike Information Criterion (CAIC) proposed by Gurmu and Trivedi (1996) to compare these specifications (Note 9).This test shows that both specifications do not differ.
It appears from the estimates that tertiary education has a negative impact on parity.This is consistent with the view that education increases the opportunity cost of female employment, and thus changes the objective conditions under which fertility decisions are taken (Ainsworth, 1988;Johnson-Hanks, 2002).Support from relatives in rearing children and the presence of other relatives in the household exert a positive influence on fertility.This result is a confirmation to the fact that, through reduced costs of rearing children, intra-family (and interfamily) solidarities hamper rational microeconomic behaviours (Rwenge, 1999), and leads to high levels of fertility.Health status exerts a positive and significant effect on fertility, but disability status does not.Participation in the labour market exerts a negative influence on fertility.

The Health equations
Health equations are estimated using probit (for disability index) and ordered probit (for the self-reported health).Our results show that controlling for short-term and long-term diseases reduces the impact of behavioural variables like age and education.Actually, age and health have a non-linear relationship, the disability status increases with age at a decreasing rate; education has no impact either on self-reported health or on the disability status, though from first stage results, it appears that education fosters health by reducing the probability of being disabled.These results are consistent with those obtained by Cai and Kalb (2004).
Fertility has a negative influence on health.It decreases the probability of reporting very-good health by 5.2% (while it increases the probability of reporting fair health by 5.8%).As expected, health limitations, long-term disease and problems within 14 days are found to be negatively related to health.Employment is found to be a significant determinant of health.Compared to non-employed women, public and informal workers are less likely to be disabled.Workers of the formal private sector, compared to non-workers, have a 40.4% higher propensity to declare fair health, a 14% lesser propensity to declare good health, and a 36.3%lesser propensity to declare very-good health.The positive effect of public and informal worker status on self-reported health indicates that justification bias is likely to be observed in these sectors.For instance in Cameroon, civil servants usually justify absenteeism by reporting that they have health problems.All these considered, the influence of employment status on health differs from one sector to another.The negative impact is likely to be caused by bad working conditions and stress.

The Participation equations
The decision to participate in the labour force is influenced by age, education, marital status, fertility, health and the province of origin.There is a concave relationship between age and participation, the highest contribution to this result being observed in the formal private sector.Single women (single, separated, divorced and widowed) have a greater tendency to participate in the labour market than those who live with man (married and cohabiting) do.Relative Risk Ratios (RRR) between these two groups of women are 1.476 for the public sector, 1.718 for the formal private sector and 1.756 for the informal sector; they express greater propensity for singles to supply their labour.
In general, education increases the probability of labour force participation.But, there are differential effects across the various sectors.While the higher the level of education the higher the probability of working in the formal sector (private and public), in the informal sector higher levels of education reduce this probability.Compared to those who did not attend school, having a tertiary education induces a RRR (between working and not working) equal to 7.890 for the public sector, 2.238 for the formal private sector, and 0.172 for the informal sectors.
The origin of the woman has a significant impact of her participation in the private (formal and informal) sector.To compare the influence of the ethnic group of the woman on her choice of the labour market sector as suggested by Lanot and Muller (1997), RRR risk ratios using public sector as the base outcome were computed.These results show that, women from the northern part of Cameroon (compared to those from the littoral) are 25.74% less likely to work in the formal private sector and 30.82% less likely to work in the informal sector.For those from the southern part of the country, the RRR recorded was equal to 41.86% and 64.76% respectively.Unlike these groups, women from the grassroots (western provinces of the country) have a greater propensity to choose the private sector.These propensities are 27.15% and 100.2% higher for the formal private and informal sectors.
As far as fertility and health are concerned, estimates show that, fertility increases the probability of working by 2.4% in the public sector, by 1.7% in the formal private sector, and by 7.5% in the informal sector.Health status is a significant positive determinant of employment, but there are differentials in this influence across the sectors of the labour market.While, self-reported health induces a 2.4% increase in the probability of working in the public sector and a 4.7% increase for the formal private sector, the disability status reduces these probabilities by 8.1% and 22.2% (respectively for the public and private sectors).Influences on the informal sector participation decision are not significant.

Discussion and conclusion.
This study aimed at determining the impact of fertility and health status on female labour force participation.Econometric analyses were based on a sample of urban workers aged 18 to 64 years old.In contrast to previous models in the literature, the study aimed to demonstrate the interrelationship between fertility and health status, and argued that failing to account for this interrelationship may lead to biased estimates of either the impact of health or fertility on female labour force participation.In this process, analyses show that it is not relevant to consider a single labour market as in the traditional neoclassical labour market; individual behaviours and labour force determinants differ across the various sectors.
The fact that the number of children exerts a positive influence on the participation contrasts with the results of other studies, which consider fertility as an exogenous variable.But this result is consistent with those of Cain and Dooley (1976), Hout (1978), andIacovou (2001).The non-significant impact of fertility in the informal sector is close to the conclusions made by Hill and Stafford (1985).
The differential impacts on the participation of the self-reported health and the disability status were questioned.Therefore relative risk ratios were computed for fertility, self-reported health and disability status, and estimates reported in table 11.As a means of comparison between the two indicators, results show that the use of self-reported health leads to an upper-bias of the impact of health status on labour market participation.
Another result of this paper is the evidence that fertility determines health status and vice versa.These results suggest that estimates of either the impact of fertility or the influence of health status on women labour force participation must take into consideration the interrelationship between health and fertility.
Throughout this paper, we considered only 2SLS estimates of our multiple simultaneous equations model.However, one could argue that these equations could have been estimated jointly.Attempts in this direction have not been successful, as handling a trivariate model is technically and computationally difficult.We hope this attempt contributes to defining the steps of this line of inquiry.

Notes
Note 1. Chirikos (1993) and Curie and Madrian (1999) review the literature on this issue.Note 2. In 2004, the fertility rate was 5.0 in Cameroon, the highest (6.1) being observed in rural areas (INS/DNSC, 2004).About 45% of women suffer from anaemia; those who have a child are almost 49%, the highest rates being observed in urban areas (54% in Yaoundé, 44% in Douala, and 42 % in rural areas).Female nutritional status (measured by the Body Mass Index), an important determinant of female mortality (WHO, 1995), is also a concern in Cameroon: 7% of women have a BMI less than 18.5 and 29% are over 25, the highest BMI being observed in Yaoundé and Douala (25.5).These results in high maternal mortality rates: between 1998 and 2004, the rate was evaluated at 669 female deaths for 100,000; this rate is far higher than that observed in developed countries.Note 3.According to the sampling theory, when computing the probability p, the acceptable error is minimum when the population variance is set equal to p(1-p)/n and the studied phenomenon approximated to a normal distribution.Then using the Moivre-Laplace theorem, the sample size is determined using the formula, where represents the p-value of the normal distribution N(0; 1).See also Adcock (1997) for further details on sample size determination.Note 4. Lack of comparability across individuals represents measurement error that is likely to lead to underestimates of the impact of health on labour force participation, while the endogeneity of self-reported health is likely to lead to overestimates.Biased estimates of health's impact on outcomes will also bias coefficients on any variable correlated with health.Finally, the dependence of self-reported health on economic characteristics will bias estimates of the impact of economic variables on participation, even if one correctly measures the impact of health itself.
Note 7. The sample mean 1.762882 is statistically equal to 2 as the t-ratio is equal to (0.237/0.925) and Pr(|T| > |t|) = 0.0000.Note 8.In the study by Sickles and Taubman (1986) of the relationship between health and labour force participation, the mean age is 63.3 years.Note 9. where, represents the value of likelihood function, is the number of parameters and refers to the number of observations.For both specifications, CAIC is equal to 6870.167 and 6869.48.($) Values represent percentages of women concerned with each level of disability.

Variables Description
Participation 1 = if employed in the public sector ; 2 = formal private sector ; 3 = informal sector ; 0 = not employed.

Fertility
Fertility refers to the parity, that is, the number of children born to a woman at the date of the survey.
1 if the respondent has any health-related problem in the last 14 days before the survey, 0 otherwise.
1 if the respondent has any disease which has lasted more than one month.

Mobility/standing
Watching/hearing Each of these variables is dummy coded 1 if the woman reports a health problem related to it and 0 otherwise.

Disabled
1 if the composite index of disability, CID≥ 4 and 0 otherwise.

Education
Highest level of education (for those who completed schooling).Education was classified in four levels: 0 = no education; 1 = primary; 2 = secondary; 3 = tertiary (university and other related categories of higher education).
Age squared divided by 100.
Single 1 if the respondent is single, separated, divorced or widowed; 0 if the respondent cohabitates or is married.

ε
6) Where, equation (a) represents the fertility equation, (b) is the health equation and (c) the participation equation.Fertility ( ) 1 Y is a function of the latent value of health status ( ) represents the error term.Equations (b) and (c) are defined along the same lines.

Y
are replaced in the structural equations as expressed in Equation8.

Child
whose charges are not supported by the woman nor her husband.Number of own child born alive who died.Number of relatives (other children and adults) living in the same house with the woman.
provinces out of which, we defined 5 groups according to social habits and customs.We have: 1 = North grouping the Far-North, North, and Adamaoua provinces; 2 = Central province; 3 = South is composed of South and East provinces; 4 = West comprised the West, North-West and South-West provinces; 5 = Littoral province.

Table 1 .
Characteristics of the final sample of women surveyed.

Table 2 .
Prevalence of potential disability in the sample

Table 5 .
Second stages maximum likelihood estimates of fertility equation Dependent variable: Fertility; (1) is estimated using Sub-Health as the health indicator; (2) is estimated using Disabled as the health indicator.Variables No-education, Other religions, Not-employed, are base outcomes for education, religion, participation status.Values within parentheses under estimators represent t-Student.

Table 7 .
Second stages maximum likelihood estimates of participation equations