Explaining Inter-Household Gender Inequality in Cameroon: An Oaxaca-Blinder Approach1

By

Francis Menjo Baye and Boniface Ngah Epo 2

Department of Economics and ManagementUniversity of Yaoundé II, Cameroon

P.O. Box 1365 YaoundéE-mail: bayemenjo@yahoo.com and epongahb@yahoo.fr

AbstractThis paper spatially and inter-temporally investigates inter-household gender inequality and poverty in Cameroon between 1996 and 2001 using the Oaxaca-Blinder approach and the ECAM I and ECAM II household surveys. Personal household characteristics are believed to be associated with inter-household inequality as suggested by gender-neutral and gender-bias indicators. We use a series of econometric analyses to identify variables such as: level of educational, household size and type, age and experience, sector of activity and milieu of residence as among those that are significantly associated with household real expenditures, gender inequality and poverty. Discrimination coefficients depict varying trends for the various factors that accounts for gender inequality in Cameroon with discrimination biased in favor of male headed households. Policy implications point to the need for bridging educational differences between genders, alleviating regional disparities, putting in place mechanism that rationalize household sizes, regulating the informal sector and encouraging development oriented investments in rural areas. Tacking these issues may reduce inter-household gender inequality and augment well-being of grass-root populations.

I. Introduction

Inequality remains a crucial issue in complementing welfare enhancement strategies (Atkinson, 1997; Atkinson and Bourguignon, 2000). Gender inequality analysis is currently a topic of intense debate by many researchers and policy makers. Most countries now integrate gender welfare issues in almost all their policy implications because most reports (UN Millennium development Goals Report, 2006; UN Millennium development Goals Indicator Database, 2007; UNESCO, 2007; Kabeer, 2003; etc.) reveal that there is a high degree of gender discrimination in favor of men when it comes to implementing or establishing policies that can help increase the standard of living of the grass root populations. Moreover, input

1This paper is extracted from a research work carried out with financial and scientific support from the Poverty and Economic Policy (PEP) Research Network, which is financed by the Australian Agency for International Development (AusAID) and by the Government of Canada through the Canadian International Development Agency (CIDA), and

the International Development Research Centre (IDRC) .2 Boniface Ngah Epo is a Research student in the Faculty of Economics and Management, University of Yaoundé II, and Francis Menjo Baye is an associate Professor in the same Institution.

1

analysis on public budget formulation through a gender lens is still lacking in Cameroon. Balmori (2003) asserts that budget initiatives worldwide (Cameroon inclusive) have inherited the gender-blindness of macroeconomic models which lack adequate data, transparency and participation, and inappropriate projections, etc. Likewise, Cagatay et al., (2000) takes this issue further by questioning how the design of macroeconomic frameworks and policies take into account the voices and interest of women and poor people with a provision for democratizing decision making processes.

The African Union Solemn Declaration of Gender Equality in Africa (SDGEA, 2004) aimed at addressing issues of major concern to women of Africa consolidated the importance of gender equality and women’s empowerment in policy debate and orientation. Currently, international policy agenda has been reshaped through policy platforms such as the 1994 International Conference on Population and Development (ICPD) in Cairo and the 1995 Fourth World Conference on Women in Beijing. In view of achieving national and global development goals, particularly those underscored in the Millennium Development Goals3

(MDG), with the third goal (MDG 3) specifically addressing the promotion of gender equality and the empowerment of women, most government now implement crucial policies that aim at bridging gender inequality and poverty.

In Africa in general, and Cameroon in particular, despite substantial strides towards promoting gender equality and empower women in some areas (e.g. in political participation), women still face enormous constraints (a disproportionate burden of the HIV/AIDS pandemic; they suffer more gender-based violence within their homes; limited access to productive resources especially land). Baye and Fambon (2009) further notes that with a high fertility level in Cameroon, women continue to bear huge health and reproductive health burdens. New challenges, including globalization and climate change, may also disproportionately affect women.

For instance, the UN Millennium Development Goal Report (2006) noted a failure in gender parity in the attainment of educational objectives as highlighted by the MDG3 objective which states the promotion of gender equality and empowerment of women (Department For International Development (DFID), 2007). Although women empowerment is one of the key MDGs enacted by the government of Cameroon, Cameroon has been very slow in moving this process4 forward. According to the UNDP Cameroon Office (MDG progress report, 2002; 2003) it seems unlikely, given the progress made for Cameroon to attain most of the MDG3 objectives before the dateline set for the full implementation of all the MDGs in view of halving poverty by 2015. Sen and Anand (1995), while evaluating theories and measurement of gender inequality in human development, acknowledge the key role of adequate indicators in illuminating differences between men and women.

3 An assessment of global progress on the MDGs noted a slight change in tides with “doors slowly opening for women in the labor market”. However, women still account for over 60% of unpaid family workers (UN DESA, 2007); only 17% of members of single or lower houses of parliament; and more girls than boys remain out of school (UNSD, 2007).4 Most advantages which constitute an important stock of resources, which can help boost and diversify the economy of a developing country like Cameroon are ill partitioned between men and women. It is thus urgent that this situation be overcome which if left unheeded, will weaken the foundations for sustainable development-undermining the country’s social fabric-and act as potential cause of the stagnation observed in the fight to curb poverty.

2

A better understanding of inequality patterns is vital in comprehending patterns of economic growth, development and enhancing local welfare. Among the 18 million inhabitants that live in Cameroon, about 52% is made up of women (INS, 2007; 2008). In addition, the overwhelming majority of the 70% of people living in rural areas, and mostly practicing farming activities are women (INS, 2004; 2005). In this context, the government of Cameroon has as one of its main objectives the fight against poverty and bridging inequality5 gaps between regions, population sub-groups and gender, with the aim of increasing development and grass-root well-being though efficient and effective policies (Government of Cameroon; 2000; 2003).

At the national level, Cameroon has ratified a number of international conventions and instruments related to gender issues, one of which is The Convention on the Elimination of all Forms of Discrimination against Women (CEDAW). Unfortunately, we still identify gender-bias and gender-neutral behaviours that discriminate and violate women rights. Since the crisis period of the mid 1980s, household income generating activities have been restructured, giving women a more important role in different activities to ensure the availability of goods and services for family consumption. Thus, at the root of gender dimension of inequality and poverty is unequal access and control of productive resources by men and women.

In Cameroon we observe a dichotomy when analyzing reproduction and productive works. While the overall burden of the former is shouldered by women (this is labor intensive and time-consuming), the latter considered as “real work” because it involves the production of goods and services are in favor of men. Moser and Levy (1980) observe that when most individuals are asked what they do, their response is often related to productive works in which they are either paid for, or which generates income. This is likely based on very constricted definition of the type of workload attributable to women (Kabeer, 2003).

Socio-cultural fabric in most traditions in developing countries and Cameroon have out rightly been in favor of men6, notably as concerns land inheritance, relegating the women to secondary roles that prevent them from acquiring and developing assets to participate fully in realizing growth potentials that can aid households break away from poverty and inequality traps (World Bank, 2005 and; Endeley and Sikod, 2006; Fonjong, 2001). These differences or disadvantages that are gender-oriented and skewed in favor of men affect supply responses and resource allocation both at micro (household) and macro levels of well-being (Sikod, 2006). Gender analysts remark that women and children are more vulnerable7. This dampens gender-neutral and gender-bias opportunity to undertake remunerative activities that will help them acquire their own assets, thus, limiting overall household and local community welfare.

5 Enough evidence shows that, despite a fall in the incidence of poverty between 1996 and 2001, after an increase within the period 1984-1996, inequality has at best marginally stagnated (Chameni, 2005; Baye, 2006, 2007; Epo, 2006; etc.). Statistically testing the change in incidence, depth and severity of poverty in Cameroon at a 5% significant level with the use of the DAD distributive software- developed by Jean-Yves Ducalos, Abdelkrim Araar and Carl Fortio of the MIMAP program, IDRC, University of Laval- Epo (2006) remark that while the null hypothesis was rejected for the incidence and depth of poverty, the story line was not the same as concerns the severity of poverty. This raises the question of shading more light on the inequality situation among the poorest of the poor captured by this index. 6 In Cameroon, for instance, fewer women compared to men own land because of certain socio-economic and cultural constraints, particularly, subordination of women within marriages and inadequate economic power to purchase land market prices.7 Vulnerability is considered as a type of asset. It entails defenselessness, insecurity and exposure to risk. Thus it is obvious that the more people are asset endowed, the less vulnerable they will be to protect themselves from poverty (Moser and Felton, 2006).

3

Overviewing gender issues in Cameroon, in 2001, the incidence of poverty for men was 39.9% compared to 40.5% for women, with women being more affected by poverty than men (INS, 2002a; 2002b; Government of Cameroon, 2003). In 2007, while out of 10 households headed by men at least 4 were poor, this ratio for women stood at 3 out of 10 households headed by women. To some extend this observation is accounted for by the smaller number of households headed by women, transfer payments received from third parties and low levels of expenditure by women outside their household setup (INS, 2008). Gender inequality can be perceived relative to the levels of human capital between women and men. When analyzing these characteristics it is observed that human capital among women remain inferior to men. For example, the rate of literacy between men and women stood at 63.7% for men against 40.7% for women in 1996; and in 2001 this stood at 66.5% among men and 46.6% among women. Despite a global increase in the rate of literacy, gender inequality that disfavours woman in terms of literacy rates is still rampant. This is due to the fact that women are usually in situations where they are incapable of generating adequate income (Government of Cameroon, 2003).

Gender disparity can be narrowly defined as the purely descriptive observation of different outcomes between males and females. However, moving beyond the descriptive level and asking what might cause gender disparities produces a complex interplay of the possible sources (Filmer et. al., 1998). The contribution of this paper anchors on the application of the Oaxaca (1973) and Blinder (1973) approach, which involves an econometric evaluation of determinants of household well-being, gender inequality, and discrimination index, as well as teasing-out the partial effects of each household or village endowment on gender inequality in Cameroon. These issues are important because gender considerations are still at an embryonic state in Cameroon in terms of limited empirical studies. Moreover, the Ministry of Women Affairs (leader in the fight for women emancipation) acknowledges that some strategies have been inappropriately implemented due to inadequate evidence-based policy information. Such knowledge is essential because the role played by women (in both urban and rural milieus) in fighting poverty8 and stirring development via gender enhancement and equalization is important (Boserup, 1970; Feldstein and Jiggins, 1994).

To tackle the household gender inequality question in Cameroon in order to participate in public policy debate, our main research question is what is the pattern of well-being and inter-household gender inequality in Cameroon? Specifically: (1) which gender-neutral and gender-bias factors explain household income in Cameroon; (2) what are the between gender contributions to total inequality in Cameroon; and (3) which variables account for the between-and-within gender inequality in Cameroon? Our main research objective is, therefore, to tease-out the nature of inter-household gender well-being and inequality in Cameroon. Specifically, the paper seeks: (1) to econometrically estimate a well-being generating function; (2) to identify gender-neutral and gender-bias factors that influence household well-being in Cameroon; (3) determine between group gender-inequality in Cameroon; and (3) suggest policy implications on the basis of the findings.

8 In most rural areas in Less Developed Countries Women are the bread winners of most families, and issues that help consolidate their position need to be clearly incorporated into the Poverty Reduction Strategies (MacCormack and Strathern, 1980; Zuidberb, 1994; Fonchingong, 1999; Mosse, 1994). Sikod (2006) reports that about 93.5% of people doing agriculture in rural areas are women.

4

Having outlined the research questions and objectives in section I, in Section II we review literature, section III methodology and data presentation, section IV empirical results and the conclusion.

II. Literature Review

Most studies using the Oaxaca-Blinded approach have tilted their attention in determining the role of labor market discrimination in assigning men and women to different occupations (Lissenburgh, (2000); González et. al., (2005); Pena-Boquete, Y. and F. Melchor (2006), etc.) Despite this popularity to the best of our knowledge there has been very little adaptation of this to study overall gender inequality. In this vein, Klassen (2005) in studying the impact of gender inequality of pro-poor growth recognizes that there is little or less information on the impact of gender gaps (disparity) on inequality. A review of literature reveals that despite a wide variety of gender disparities existing in various domains, analyzing inter-household gender inequality is quasi inexistent. Overall some gender material reviews include: Blackden and Bhanu (1996) who analyze human assets and finds that in sub-Saharan Africa (SSA) gender differentials in reproductive health disfavours women more than men; regarding social assets, precisely norms, DasGupta (1987) observes that cultural rights and obligations favor sons relative to daughters in rural India. Ahuja and Filmer (1996) identify disparities in male-female educational attainments and enrollment levels in developing countries. Wan and Cai (2005) applying a multinomial Logit regression decomposition note that unequal access to other factors rather than pay are causes of discrimination of gender wages in certain industries (sectors) in China.

Seeley et al., (2004) in analyzing the impact of gender and HIV/AIDS as well as constraints, identify gender-specific constraints, gender-intensified disadvantages and gender-imposed constraints that hamper the fight against the AIDS pandemic-concluding that unequal gender relations and the nature of development need to be altered. Poats et. al., (1988); Zuidberg (1994) and The Canadian International Development Agency (1998) evaluate the role of gender in farming extension activities and the development of rural areas in developing counties, identifying the eminent role women play in increasing rural welfare. Globally, sub-Saharan Africa could experience 6-20% increase in agricultural outputs if women’s access to agricultural inputs was equal to men’s. In the Middle East and North Africa, estimates reveal that if the female labour force participation rate had increased during the 1990s same like women education, the average household income could have been 25% higher (DFID, 2007).

Guijt and Shah (1989); Moser (1989; 1993); Kabeer (1994; 1995); and Levy (1996) underscore the key role of gender enhancement and institutional development, showing the positive correlation that exist between consolidating the rights of women and strengthening the macro and micro institutional frameworks in most developing countries. Jeff and Basu (1998) and Lock and Okali (1999) recognize the key political role of gender empowerment in consolidating democratic processes and fostering development. In this vein the DFID (2007) reports that despite limited progress, globally on average in the world, women represent only (17%) in parliaments, remaining severely under-represented in political and decision-making positions in many countries. This figure is considerably lower in countries such as Egypt, where just 2% of MPs are female. In Cameroon, the number is less that 17% of the total number of parliamentarians. Using cross country data, Filmer et al., (1998) established that the level of gender disparities in health and education outcomes for girls at the national level

5

in South Asia is the largest in the world; second, the differences in gender disparity among Indian states or among the provinces of Pakistan are typically larger than those across the nations of the world; and third, while the absolute level of health and education outcomes for girls are strongly related to economic conditions, the disparities between females and males are not.

There are very little empirically sound studies that target gender inequality in Cameroon. In this category, Sikod (2006) distinguishes between assets (private and public) that affect labour productivity and intra-household decision making processes. However, consistent with most works on gender issues this study is descriptive in nature. Endeley and Sikod (2005) evaluate the impact of the Chad-Cameroon pipeline and how this affects gender relations, land resources and community livelihood. The study finds a bias in favour of men in terms of recruitment, and benefits obtained from the construction of the pipeline and the communities that benefited from this process.

Ngome (2003) elicits the role of gender division of labor and women’s decision making in rural households. The study finds that in rural areas, socio-economic and cultural constraints cause women to get only secondary roles that impairs on the capability to generate resources and thus they cannot bargain adequately in decision making processes in their households. Fonchingong (1999) evaluates the impact of the structural adjustment reforms on women and how this affects agricultural output in Cameroon. This study reveals that enhanced agricultural productivity could be observed if adequate governmental policies that empower women are enacted. Fonjong (2001) questions the role of NGOs in augmenting the participation of women in fostering development aimed at increasing welfare. These studies aim at understanding gender issues in Cameroon, but fail to use methodologies that go beyond descriptive analysis. The value added of this paper is to resolving this issue.

III. Methodology and Data

III.1 The Oaxaca-Blinder traditional decomposition Approach

Applying the Oaxaca-Blinder traditional decomposition methodology applied to gender inequality in Cameroon, we +elicit the contribution of factors that explain gender-neutrality (gender-expenditure discrimination) and gender-bias (personal-specific characteristics) in Cameroon.

In tackling gender inequality and factors that contribute to household income/expenditure inequality within male-headed and female-headed household, we follow Oaxaca (1973) and Blinder (1973). This technique hinges on the basic premise that in the absence of societal discrimination (considered as a source of inequality), the household income expenditure structure faced by men also applies to women. Measuring sources of the difference in gender expenditure-discrimination9 and personal-specific characteristics we estimate separate semi-log income expenditure functions for male and female, and decompose the differences in the mean of the log of household income expenditure. In this section, we: (1) study the partial gender inequality of a particular characteristic; and (2) determine a discrimination index that captures discrimination between male-and-female headed households.

9 Discrimination here is considered as the difference not explained by differences in household income expenditure-determining personal characteristics.

6

Let the male and female geometric mean of household expenditure be denoted by MtY−

and FtY−

, we decompose the log-differential of the geometric mean ∆:

0 0log log log log logM

M F F Ftt t t t

Ft

YY Y Y Y

Y

−− − − −

−

∆ ≡ = − + − (1)

Let 0FtY−

be a hypothetical distortion-free or discrimination-free mean household expenditure

for woman, the subscript t represents the year or period considered. This could be equal to t for the initial period and t+1 the final period. Furthermore, summarizing the variation in the male-female income expenditure cross section sample by using the following common place statistical models:

, , ,i M

i t M t t i tIn y X β ε= + (2)

and

, , ,j F

j t F t t j tIn y X β ε= + (3)

where , ,i t My is the log income expenditure of man i at the considered period and , ,j t Fy is the

log income expenditure of woman j at the considered period; Mtβ and F

tβ are the coefficients

that determine the effects of characteristics on household spending and itX and j

tX are the vectors of the mean personal endowments related characteristics of man i and woman j. Since the regression function passes through the sample mean of X and y, taking the arithmetic average of equation (2) and (3), the stochastic ε -term drops out.

Designating arithmetic mean by an over lined variabletyα−

, where ForM=α for men and

women, and applying a semi-log regression we have:

M M Mt t tIn y X β− − ∧

= (4)

and

F F Ft t tIn y X β− − ∧

= (5),

simply stating that mean expenditures are predicted by using mean characteristics and, Ftβ∧

and Mtβ∧

the vectors of the estimated coefficients of the female and male groups. Since ty

− is

the mean of the log at the considered period, and is the log of the geometric means tY

−we then

plug equation (4) and (5) into (1), and averaging we obtain:

2 2

M F M FM F M Ft t t tt t t t

X XX X

β β β β∧ ∧ − −

− − ∧ ∧ + + ∆ = − + − (6)

Equation (6) is a decomposition of the effects of difference in average characteristics (the first term) and the effects of difference in treatment of characteristics (second term). Distinguishing the contribution of different characteristics on the one hand and of the unexplained differential effect on the other hand, the structural form of equation (6) resolves the critical issue10 of having to define a priori a reference structure as per the base and/or final years used in this analysis. This structure (equation 6) emerges by avoiding the arbitrariness in

10 A review of literature raises two issues: (1) how do we account for the unexplained part in explaining the considered variable and; (2) how do we choose the reference structure. In Oaxaca and Ransom, (1994) the

7

selecting the gender structure reference by using the Shapley value approach. The first term simply shows the part of the log- expenditure differential between men and women that can be explained by different personal characteristics. The second term is the difference not explained by the difference in expenditure-determining personal characteristics and thus could be considered as the rough measure of expenditure discrimination for female household heads.

Similar to Takahashi (2007) we can estimate the partial effect of a particular individual endowment or characteristic on gender inequality as follows:

, , , ,, , , , ,2 2

M F M Fi t i t i t i tM F M F

i t i t i t i t i t

b b x xPE x x b b

∧ ∧ − −− − ∧ ∧ + + = − + −

(7)

where , ,M Fi t i tb and b∧ ∧

are the estimated coefficients of the thi − characteristics, ,i tx

− denotes the

mean value of the thi − characteristic or endowment and ,i tPE is the partial gender inequality attributable to the characteristics of the individual.

Inter-temporally computing gender neutral and gender biased characteristics between 1996 and 2001, hinging on Shorrocks (1999) decomposition framework, we extend equation (6) to:

, 1 , , 1 , 1 1

2001 1996 1 1 , 1 , ,2 2

M M F F M M F Fi t i t i t i t t t t t

M M F F M Mt t t t i t i t i t

X X X X

Inter X X X X

β β β ββ β β

∧ ∧ ∧ ∧ − − − −

+ + + +− − − − ∧ ∧

− + + + +

− + − − + − = + − + + + +

1 ,F F

i tβ∧ ∧

+

(8)

Having formulated gender income/expenditure inequality, we are going to evaluate the rate of discrimination between the men and women based on the variables considered; by constructing discrimination index (Lissenburgh, 2000; Chzhen, 2006; Gonzalez et. al., 2005). This analysis will inform us on gender income inequality using a discrimination or segregation index. Let us denote the Discrimination coefficient or index by:

, exp 1 100M F Ff t t t tDISC Xβ β

∧ ∧ − = − − ×

(9)11

Here ,f tDISC equals to the percentage change in the income/expenditure women could undertake given that they have the same attributes as men or if they had the same income/expenditure as men. This, therefore, represents the increase in well-being of women if the discriminations were eliminated12.

reference structure is chosen as a matrix weighted average of the male and female β -vectors. However, Oaxaca and Ransom (1997) cautions us on the use of categorical variables because its presence implies that the unexplained part of each of these variables cannot be uniquely determined. Therefore, as in Vartiainen, (2002) and Papapetrou, (2005), we choose simply to present the computations using both male and female coefficient as references. The value-added of this procedure is that it permits a clear interpretation: using male (female) structure as reference and computing the effects of different characteristics yields an answer to the question “how much would be left of the gross differential if women (men) were treated as men (women) with similar characteristics?”11 This index may be subjected to modifications to incorporate other methods as in Chzhen (2006) or Gonzalez et al, (2005).12 The discrimination coefficient can be carried out using the male human capital attributes as standardize factors.

8

III.2 Econometric Approaches

Beyond the standard OLS approach, going through comparative analysis of econometric methods, most studies carrying out the Oaxaca-Blinder approach have out rightly identified the inherent econometric limitations appended to the standard OLS method. Among these we identify potential endogeneity, multicorrelation and heterosckedasticity. These limitations if not appropriately targeted may tend to produce less robust results as per indicators obtained. In view of the above, in this study we compare three regression procedures (OLS, IV (2SLS) and the Control Function approaches), attempting to adopt the approach that purges potential endogeneity, unobservables and non-linear correlation between the regressands and regressors, or the independent variables, and verifying the veracity of each estimation procedure. The educational level of household heads is envisaged as being complementary to relevant unobserved inputs that cause household economic well-being. The causal relationship between household welfare and educational can be depicted by controlling for potential endogenity and unobserved effects as:

1, , 1,log t t y t t t tY x Lδ η ε= + + (10)

2, , 2,t t l t tL x δ ε= + (11)

where Y and L are household income expenditure and the educational of household head-the endogenous explanatory variable in the household income expenditure function-surrogated by

the educational level of the household head. 1,tx is a vector of exogenous variables that

capture individual, household and spatial characteristics, 2,tx is a vector of exogenous

variables, comprising 1,tx correlates that belong to the income expenditure function and a

vector of instrumental variables that affect educational levels, L, but have no effect on

household income expenditure and t tandδ η are parameters to be estimated, and ,i tε , the

respective error terms.

The null hypothesis of no complementary effects of education on household income

expenditure is 0 : 0H η = . The alternative hypothesis (assuming complementarities) 0η > is

presumed to act via increase possibility of generating extra-income because household head is educated, thus employed in better jobs. The vital argument in the complementary hypothesis here is not that education directly increases household income expenditure, but rather that education elicits extra-employment possibilities and knowledge, which is strongly correlated with augmenting the income earning capabilities of the household head and thus revenue for expenditure.

Equation (10) being the structural equation of interest may be affected by the reverse causality if education is expected to complement with some inputs that are omitted from the household income expenditure. Thus OLS will produce biased results. Purging potential endogeneity we adopt instrumental variables that do not belong to the outcome equation but are correlated with the endogenous explanatory variable (education), conditional on the other covariates. Additionally, heterogeneity of income expenditure due to non-linear interaction of educational with unobserved or omitted variables could bias the estimated structural coefficients. Equation (11) is the linear projection of the potential endogenous variable, L, on all the exogenous variables, that is, a reduced form linear probability model of education of the

9

household head. Augmenting equation (10) into a control function to address this potential issue, we have:

1, 1, 2, 2,log ( . )t t t t t t t t t t t tY x L Lα δ η α ε γ ε µ∧ ∧

= + + + + + (11)

where 2,tε∧

is the fitted residual of education, derived from the reduced form linear

probability model of educational level, that controls for unobserved variables correlated with

L; 2,( . )t tLε∧

is the interaction of the fitted educational level residual with the actual value of

educational levels13 and tµ is a composite error term comprising 1,tε and the unpredicted

part of 2,tε , under the assumption that ( )1, 0tE ε = ; 1,, ,t t t tandδ η α γ are parameters to be

estimated.

As noted in Wooldridge (1997), the IV estimates are unbiased and consistent only when (a)

the expected value of the interaction between education level and its residual 2( . )Lε

∧is zero or

the correlation is linear and (b) there is no sample selection problem. However, if the correlation is non-linear, then the control function approach is required and the inclusion of

the interaction term2( . )Lε

∧ in equation (11) purifies the estimated coefficients of unobservable

variables (Card, 2001).

13 Wooldridge (1997) noted that, the IV is unbiased and consistent only when: (a) the expected value of the

interaction between literacy and its residual 2( . )Lε

∧ is zero or the correlation is linear and; (b) there is no sample

selection problem. However, if the correlation observed is non-linear, then the control function approach is required and the inclusion of the interaction term necessary.

10

III.3 Data

The data used in this study is the ECAM I (1996) and ECAM II (2001). The ECAM I survey was conducted by the DSCN-MINEFI in 1996 (over three months). This survey is comprised of a random sample of approximately 1800 households (both urban and rural) in the ten provinces of the national territory (DSCN, 1997). This survey was aimed at the following objectives:(1) Measure the effects of the economic crisis and adjustment measures on the level and standards of living of the households; (2) Establish the relationship that exist between the different dimensions of well-being of these households and; (3) Analyse the tendencies and evolutions of household well-being relative to other sources of data. The ECAM II survey was undertaken from September to December 2001. This household survey was carried out to remedy mistakes made in the first household survey and ameliorate information concerning the poverty profiles. This survey was comprised of 11,553 households of whom 10,992 were actually visited. In additions, (1) it was conducted to propose an adequate methodology for calculating poverty lines and profiles acceptable to major development partners, and which serves as a reference for further analysis. This acts as a follow up of the poverty reduction program; (2) To analyse monetary poverty, poverty in terms of living conditions of most households and potential poverty, while establishing the correlation between them; (3) Consolidate past analysis at a national and regional level, while isolating the two large towns (Douala and Yaoundé) and also distinguishing area of residence (urban or rural) and; (4) To produce an adequate data base to ameliorate different statistics (of the population), notably in establishing household consumption in national accounts and updating data used in calculating price indexes (INS, 2002a; 2002b).

These surveys contain information on expenditures of households and demographic data. They also contain individual characteristics, household characteristics and community characteristics. Variables selected for our empirical work alongside their descriptive statistics are hosted in Tables 1-3.

11

Table 1: SUMMARY DESCRIPTIVE STATISTICS OF VARIABLESGeneral SampleVariable Year: 1996 Year: 2001

Mean SD Min Max Mean SD Min Max

VariablesLog Total Expenditure Per Adult 12.8188 0.8150 10.181 16.084 13.0176 0.7587 10.403 17.7654Educational Level 1.7661 0.6619 1 3 2.2293 0.9364 1 4Age 42.5087 13.3895 14 99 42.9270 15.0635 13 99Age Squared 1986.1630 1269.6510 196 9801 2069.6190 1455.3450 169 9801Household Size 5.8925 3.9008 1 32 5.1349 3.5188 1 38Fraction of Active Household Members 0.3953 0.3007 0 1 0.4926 0.2994 0.0370 3Primary Sector(PSU Percentage) 1.1290 1.5237 0 4.2857 0.2570 0.2954 0 0.8841Secondary Sector (PSU Percentage) 0.6436 0.7218 0 4.3860 0.1581 0.1841 0 1.8057Sickness_past_2_Weeks (PSU Percentage) 0.7454 0.4177 0 2.0339 0.1928 0.1086 0 0.4942Access Medical Doctor(PSU Percentage) 0.6424 0.6194 0 2.5157 0.1745 0.3464 0 3.5484Water (PSU Percentage) 0.6715 0.4667 0 2.0095 0.1582 0.1150 0 0.5359Electricity(PSU Percentage) 0.6244 0.3696 0 1.6472 0.1526 0.1125 0 0.5304RegionsDouala 0.2207 0.4148 0 1 0.1017 0.3023 0 1Rural Forest 0.1213 0.3266 0 1 0.1497 0.3568 0 1Rural Haut Plateau 0.1207 0.3259 0 1 0.2112 0.4081 0 1Savannah 0.1207 0.3259 0 1 0.1865 0.3895 0 1Control for unobserved variablesPredicted Educational Level residual 3.24E-02*108 0.5757 -1.5226 1.6239 3.75E-02*108 0.7399 -2.3054 4.2091Interaction between Educational Level and its residual 0.3311 1.1848 -1.5226 4.8717 0.5474 1.9250 -3.3514 16.8362Instruments for LiteracyLand Surface Area 4.2467 2.0796 0 12.085 9.2209 26.1166 0 751.200Own Television (PSU Percentage) 0.6124 0.4693 0 2.0258 0.1411 0.1202 0 0.5564Own Radio (PSU Percentage) 0.7239 0.2738 0.1684 1.3468 0.1754 0.0704 0 0.4775Public Schools (Sum PSU squared) 0.3761 1.4494 0 16 2.2377 16.0335 0 225

Source: Computed by Authors using the ECAM1 and ECAM11 surveys

12

Table 2: SUMMARY DESCRIPTIVE STATISTICS OF VARIABLESYEAR: 1996

Variable Male Sub Group Female Sub GroupMean SD Min Max Mean SD Min Max

VariablesLog Total Expenditure Per Adult 12.7803 0.8123

10.1810 16.0847 12.9712 0.8075 10.5602 15.0631

Educational Level 1.7868 0.6731 1 3 1.676 0.6037 1 3Age 42.7864 13.4505 14 99 41.3830 13.0986 19 80Age Squared 2011.463

01289.2110 196 9801 1883.6290 1183.3480 361 6400

Household Size 6.2685 3.9849 1 32 4.3594 3.0932 1 17Fraction of Active Household Members 0.4001 0.2896 0 1 0.3746 0.3411 0 1Primary Sector(PSU Percentage) 1.2014 1.5635 0 4.2857 0.8376 1.3136 0 4.2857Secondary Sector(PSU Percentage) 0.6625 0.7478 0 4.3860 0.5670 0.5989 0 2.6316Sickness_past_2_Weeks (PSU Percentage) 0.7443 0.4258 0 2.0339 0.7487 0.3827 0 2.0339Access Medical Doctor (PSU Percentage) 0.6153 0.6192 0 2.5157 0.7511 0.6085 0 2.5157Water (PSU Percentage) 0.6588 0.4809 0 2.0096 0.7245 0.4011 0 2.0096Electricity(PSU Percentage) 0.6054 0.3740 0 1.6473 0.7005 0.3420 0 1.6473RegionsDouala 0.2239 0.4170 0 1 0.2087 0.4070 0 1Rural Forest 0.1238 0.3295 0 1 0.1101 0.3135 0 1Rural Haut Plateau 0.1246 0.3303 0 1 0.1072 0.3099 0 1Savannah 0.1317 0.3383 0 1 0.0754 0.2644 0 1Control for unobserved variablesPredicted Educational Level residual 0.0413 0.5808 -1.522 1.6239 -0.1797 0.5167 -1.2823 1.1780Interaction between Educational Level and its residual 0.4147 1.2317 -1.522 4.8717 -0.0325 0.8672 -1.2823 3.5341Instruments for LiteracyLand Surface Area 4.3065 2.1803 0 12.0855 3.9917 1.5849 2.4171 9.6684Own Television (PSU Percentage) 0.6051 0.4719 0 2.0258 0.6411 0.4572 0 1.8416Own Radio (PSU Percentage) 0.7289 0.2816 0.1684 1.3468 0.7019 0.2392 0.1684 1.3468Public Schools (Sum PSU squared) 0.3110 1.1606 0 16 0.6377 2.2463 0 16

Source: Computed by Authors using the ECAM1 and ECAM11 surveys

13

Table 3: SUMMARY DESCRIPTIVE STATISTICS OF VARIABLESYEAR: 2001

Variable Male Sub Group Female Sub GroupMean SD Min Max Mean SD Min Max

VariablesLog Total Expenditure Per Adult 13.0028 0.7336 10.40 17.7654 13.0633 0.8304 10.55 16.857Educational Level 2.3059 0.9328 1 4 1.9918 0.9076 1 4Age 42.1269 14.7852 13 99 45.4073 15.6401 14 98Age Squared 1993.255

01419.2780 169 9801 2306.3450 1538.2330 196 9604

Household Size 5.4530 3.6586 1 38 4.1488 2.8264 1 28Fraction of Active Household Members 0.4904 0.2990 0.0370 3 0.5002 0.3006 0.0714 2Primary Sector(PSU Percentage) 0.2656 0.3003 0 0.8841 0.2305 0.2782 0 0.8841Secondary Sector(PSU Percentage) 0.1609 0.1866 0 1.8057 0.1494 0.1758 0 1.8057Sickness_past_2_Weeks (PSU Percentage) 0.1926 0.1088 0 0.4942 0.1933 0.1081 0 0.4942Access Medical Doctor(PSU Percentage) 0.1712 0.3429 0 3.5484 0.1847 0.3570 0 3.5484Water (PSU Percentage) 0.1564 0.1152 0 0.5359 0.1635 0.1142 0 0.5359Electricity(PSU Percentage) 0.1487 0.1130 0 0.5304 0.1649 0.1100 0 0.5304RegionsDouala 0.1031 0.3041 0 1 0.0974 0.2965 0 1Rural Forest 0.1492 0.3563 0 1 0.1514 0.3585 0 1Rural Haut Plateau 0.1973 0.3980 0 1 0.2540 0.4354 0 1Savannah 0.2058 0.4043 0 1 0.1268 0.3328 0 1Control for unobserved variablesPredicted Educational Level residual 0.0692 0.7426 -2.305 4.2091 -0.2389 0.6785 -2.031 3.1320Interaction between Educational Level and its residual 0.7067 1.9936 -3.351 16.8362 -0.0023 1.5459 -2.777 12.5281Instruments for LiteracyLand Surface Area 9.3522 26.5666 0 751.2004 8.8133 24.6673 0 501.62Own Television (PSU Percentage) 0.1384 0.1204 0 0.5564 0.1495 0.1195 0 0.5564Own Radio (PSU Percentage) 0.1761 0.0708 0 0.4775 0.1735 0.0693 0 0.4774694Public Schools (Sum PSU squared) 2.0377 15.0872 0 225 2.8579 18.6539 0 225Constant

Source: Computed by Authors using the ECAM1 and ECAM11 surveys

Results of Descriptive Statistics

Tables 1 note that average household age for 1996 and 2001 was 42 years. Household size for both periods revolved around 5 persons. Log of total expenditure per capita was marginally higher for 2001 as compared to 1996, with the female sub sample having a slightly higher value. Average fraction of active adult’s household members was 0.49 for 2001 and 0.39 for 1996. Tables 2 remarks that the analytic sample population for 1996 is made up of 80.2% of men and 19.8% of women. In 2001 (table, 3) men constituted 75.6% of the total population as

14

compare to 24.4 for women. While average age for me was superior to women in 1996, in 2001 the opposite is observed.

IV. Empirical Results

A) Econometric Analysis

Prior to reporting results of the household income expenditure function, it is important to recall that if education is not an individual decision, but is determined entirely by public policy and is orthogonal to individual characteristics and income generating activities of household head, the OLS estimates of the structural parameters of our income equation will

yield unbiased estimates of any spill-over effects on tη . We posit that being educated

enables household heads to adopt practices that will improve income earned by households, thus their expenditure. This implies there exist a positive empirical relationship between household income expenditure and education, depicting positive externalities, as stated in the complementary hypothesis. The inverse will translate negative externalities. The results obtained fro the three regression procedures ties with the complementary hypothesis (table 4, 5 and 6). All empirical specifications generate a positive and significant effect of literacy on household income expenditure, comforting the complementary hypothesis. Thus, there is no reason attaching directly the influence of household income expenditure and literacy, rather there are spill-over effects of education on household heads capability of generating further income.

Sometimes, the educational levels of household heads could indeed correlate with elements of the error term apart from the excluded inputs, responding to changes not made explicit in the structural equation. The direction of bias is an empirical issue, because it could be that household heads with lower education that are susceptible to diminishing income expenditure due to poor endowments may be likely to demand for more education geared at increasing

income expenditure, thus biasing tη towards zero and under estimating the spill-over effect.

These considerations entails that it is important to properly estimate the structural parameters to correctly attribute causality for policy guidance.

For instance, comparing the OLS estimates (table 4) and results from Control Function Approach (CFA) (table 6) , we remark that the variable education increase by three folds for both the general and sub groups. Including the interaction of the educational status with the errors term reveals an increase of two folds, with the variable being significant.

Scrutinizing the different empirical specifications of the structural equations, Table 4, reveal that for OLS estimates of the 1996 and 2001 general and sub samples, educational level of household head was significant for all groups except the 1996 female sub groups. This apparently translates the bias women face in accessing educational opportunities and consequently on female household income generating abilities, depicted by the OLS method. This result is identical with the 2SLS and the Control Function approaches (table 5 and 6). All other variables were significant in explaining household income expenditure for 1996. In 2001, a part from fraction of adult active members, working in the secondary sector, access to water and electricity, all other variables explained general income expenditures. The variable household size significantly reduced household income expenditure for all samples. This translates the key role of family size. Households were parents have a lot of dependents generally tend to have little income attributed to individuals under their care. This is because

15

many children and other dependent members will entail more food and non-food expenditures, thus constituting a difficulty for parents to ensure adequate household welfare.

Table 4: ORDINARY LEAST SQUARE ESTIMATIONS

VariableGeneral Sample Male Subgroup Female Subgroup

1996 2001 1996 2001 1996 2001Educational Level 0.2885*** 0.2200*** 0.3578*** 0.2149*** 0.0298 0.2372***

(9.32) (26.56) (10.72) (23.51) (0.36) (11.42)Age 0.0752*** 0.0512*** 0.0820*** 0.0513*** 0.0686** 0.0463***

(7.42) (20.62) (7.48) (18.65) (2.51) (8.13)Age Squared*102 -0.0692*** -0.0507*** -0.0729*** -0.0488*** -0.0750** -0.0536***

(-6.07) (-19.91) (-6.00) (-17.27) (-2.27) (-9.39)Household Size -0.0365*** -0.0271*** -0.0257*** -0.0235*** -0.0585*** -0.0509***

(-5.71) (-11.73) (-3.66) (-9.35) (-3.51) (-7.84)Fraction of Active Household Members 0.1437** -0.0026 0.3287*** 0.0017 -0.1122 0.0446

(1.9) (-0.09) (3.71) (0.06) (-0.76) (0.73)Primary Sector(PSU Percentage) 0.1078** -0.4429*** 0.1161** -0.4368*** 0.0924 -0.4231***

(2.39) (-11.33) (2.39) (-10.07) (0.77) (-4.78)Secondary Sector(PSU Percentage) 0.0503* 0.0045 0.0804*** 0.0151 -0.0590 0.0063

(1.78) (0.12) (2.69) (0.37) (-0.70) (0.07)Sickness_past_2_Weeks(PSU Percentage) -0.1774*** -0.4315*** -0.1368* -0.4121*** -0.3417** -0.5708***

(-2.62) (-5.29) (-1.87) (-4.61) (-2.07) (-3.03)Access to Medical Doctor(PSU Percentage) 0.1715*** -0.0377* 0.1301*** -0.0404* 0.3217* -0.0146

(5.12) (-1.9) (3.62) (-1.82) (3.74) (-0.34)Water (PSU Percentage) 0.2004*** 0.0652 0.1692*** 0.1214* 0.2115 -0.1836

(3.42) (1.06) (2.75) (1.8) (1.19) (-1.29)Electricity(PSU Percentage) 0.2998*** 0.2757*** 0.2880*** 0.2535*** 0.3410 0.3706**

(3.31) (3.8) (2.9) (3.14) (1.58) (2.34)Douala 0.2092*** 0.1666*** 0.1771*** 0.1685*** 0.2667** 0.1815***

(4.02) (7.13) (3.09) (6.6) (2.24) (3.35)Rural Forest -0.5154*** -0.1016*** -0.5997*** -0.0813*** -0.3610 -0.1562**

(-4.27) (-4.22) (-4.54) (-3.04) (-1.29) (-2.94)Rural Haut Plateau -0.4789*** -0.2328*** -0.5222*** -0.2305*** -0.3564 -0.2535***

(-3.92) (-11.47) (-4.00) (-10.1) (-1.06) (-5.85)Savannah -0.4812*** -0.1235*** -0.4858*** -0.1049*** -0.4783 -0.2033***

(-3.79) (-5.25) (-3.62) (-4.1) (-1.29) (-3.58)Constant 10.41*** 11.7422*** 9.920*** 11.6555*** 11.51 12.1900***

(43.71) (174.26) (37.69) (158.4) (19.63) (74.49)R-Squared 0.308 0.3041 0.346 0.3069 0.243 0.3400Adjusted R-Squared 0.301 0.3030 0.337 0.3056 0.195 0.3356Number of Observation 1338 10176 1088 7888 250 2288Source: Computed by Authors using STATA 10. ***, ** and * are 1, 5 and 10 percent significance levels.

Likewise the age of household head, access to water and electricity all contribute to enhancing household income expenditure when looking at the OLS. Economies of scales obtained by household heads from accessing portable water cheaper and spending less time to fetch this is shifted to other sources. This trend is identical to the IV 2SLS results depicted in table 5, with the 2001 general and sub samples, except the female sub group with an opposite sign

16

identified for, though not significant. Considering the male and female sub groups, while for the 1996 male subgroup, all the variables were significant in explaining the income function, in 2001the same variables that explain that account for general income expenditure also explain male-households determinants of income expenditures, except for access to water that now explains 2.75 standard scores of household income expenditure. The female sub group note for 1996 female education do not significantly affect income expenditure though respecting the complementary hypothesis. Other variables that explain female income expenditure are age, age squared, household size, sickness, consulting a medical doctor and residing in urban areas. The 2001 female sub group recognizes that all variables significantly explain expenditure behaviors, except fraction of adult active members, working in the secondary sector, access to medical consultation by a doctor and water.

Spatially, all OLS estimates for both periods identify living in urban areas as significant in enhancing household income expenditures, with the sole exceptions being that while in the female sub sample residing in urban areas will increase income expenditures of female households, rural areas captured by the rural forest, rural haut plateau and the rural savanna regions don not significantly reduce household income expenditures. Resolving heterogeneity among clustered variables, we estimate robust estimate and clustering for both the OLS and the control function estimates (See appendix).

Results from the IV 2SLS estimates in table 5 similarly to table 4 note a percentage significance level for the education of household heads for alls samples except the female 1996 sub sample. While all the variables were significant for the global 1996 sample the only non significant variables in the 2001 general samples were access to water and living in Douala. The non significance of this regional variable is also remarked in the 1996 and 2001 male sub samples. However, for both female sub samples for the two respective years, residing in Douala have a standard 2.44 and 2.55 student scores in enhancing income expenditure by female household heads. Access to water and electricity increase increases household income expenditures via extra savings generated from positive externalities as per these facilities. The frequency of sickness of the household head is negatively related to income expenditures for all samples. This implies, constantly sick household heads have little time they spend on either educationally empowering themselves (see appendix: reduced form estimates of education) or his ability to earn revenue. This result is made very apparent in female headed households where they are generally alone and fend solely for the family, unlike male headed households where the man being the head is sometimes assisted by the wife in periods of hardship to generate income for household welfare. The variable access to medical doctors, while positive for 1996 and its sub samples was negative for 2001 except the female sub group.

Sector of activity particularly, the primary sector was positive for all 1996 and negative for 2001 samples. However, working in the secondary sector is positive for both years except for the 1996 female sub group. In addition, all samples are significant, except the female sub samples for the respective years. This logically ties down with the difficulty faced by women in accessing jobs appended to this sector of activity or job discrimination experienced by this group of individuals. Spatially, rural areas significantly aid in decreasing 1996 general and sub samples income expenditure. This observation is recognized for 2001, except rural savanna where residing in this locality significantly increases the overall income expenditure by 4.25 standard student scores, rural areas decrease income expenditures for both sexes because of low rural cost of living.

17

Table 5: INSTRUMENTAL VARIABLE TWO STAGE APPROACH

VariableGeneral Sample Male Subgroup Female Subgroup

1996 2001 1996 2001 1996 2001Educational Level 0.8353*** 0.7488*** 0.9739*** 0.7717*** 0.2780 0.6717***

(6.28) (15.75) (6.37) (13.87) (1.12) (7.21)Age 0.0674*** 0.0556*** 0.0708*** 0.0469*** 0.0635** 0.0662***

(5.94) (18.74) (5.54) (13.93) (2.31) (8.86)Age Squared*102 -0.0538*** -0.0434*** -0.0531*** -0.0338*** -0.0651** -0.0617***

(-4.11) (-14.09) (-3.62) (-9.06) (-1.91) (-9.57)Household Size -0.0339*** -0.0291*** -0.0194*** -0.0194*** -0.0537*** -0.0452***

(-4.78) (-10.59) (-2.38) (-6.32) (-3.15) (-6.26)Fraction of Active Household Members 0.2107** 0.1039*** 0.4289*** 0.1300*** -0.0684 0.1566**

(2.48) (3.12) (4.14) (3.38) (-0.45) (2.20)Primary Sector(PSU Percentage) 0.1108** -0.1812*** 0.1120** -0.1002 0.1191 -0.2980***

(2.22) (-3.51) (2.02) (-1.61) (0.99) (-2.99)Secondary Sector(PSU Percentage) 0.0758** 0.0716* 0.1152*** 0.1150** -0.0300 0.0649

(2.38) (1.63) (3.29) (2.30) (-0.34) (0.70)Sickness_past_2_Weeks(PSU Percentage) -0.1242 -0.4643*** -0.0678 -0.5065*** -0.3383** -0.5257**

(-1.64) (-4.81) (-0.80) (-4.66) (-2.08) (-2.55)Access to Medical Doctor(PSU Percentage) 0.1203*** -0.0482** 0.0801* -0.0716*** 0.2853*** 0.0121

(3.09) (-2.05) (1.87) (-2.64) (3.12) (0.26)Water (PSU Percentage) 0.1671** 0.0259 0.1067 0.0835 0.2429 -0.2206

(2.56) (0.36) (1.49) (1.02) (1.37) (-1.42)Electricity(PSU Percentage) 0.1877* -0.2865*** 0.1284 -0.3685*** 0.3061 -0.0133

(1.82) (-2.89) (1.07) (-3.19) (1.42) (-0.07)Douala 0.1308** 0.0345 0.0636 0.0106 0.2884** 0.1549**

(2.16) (1.14) (0.9) (0.30) (2.42) (2.55)Rural Forest -0.3921*** -0.1502*** -0.4756*** -0.1686*** -0.3180 -0.1346**

(-2.87) (-5.21) (-3.1) (-5.02) (-1.14) (-2.32)Rural Haut Plateau -0.3326** -0.2072*** -0.3604** -0.2423*** -0.2959 -0.1721***

(-2.39) (-8.59) (-2.35) (-8.75) (-0.88) (-3.44)Savannah -0.2541* 0.1585*** -0.2254 0.1848*** -0.4172 0.0343

(-1.69) (4.25) (-1.36) (4.4) (-1.13) (0.43)Constant 9.501*** 10.1689*** 8.9356*** 10.167*** 11.0847*** 10.4920***

(28.09) (63.66) (23.39) (59.52) (15.73) (26.48)R-Squared 0.1453 0.0226 0.1358 0.2144 0.2093Wald Chi2 451.31 2883.65 389.60 2164.27 78.75 918.24Number of Observation 1338 10134 1088 7858 250 2276Source: Computed by Authors using STATA 10. ***, ** and * are 1, 5 and 10 percent significance levels

The reduced form educational status determinants (see appendix) makes explicit that land, owning a television and accessing public schools, all enhance the flow of information in view of empowering household heads towards generating extra revenue to further accentuate household income expenditure.

A robust Control Function approach (table 6) notes education as being statistically significant for all samples unlike the two above estimation methods earlier discussed which recognized education in the female sub sample as not being significant in explaining female household

18

income expenditures. This approach identify a 1.67 standard score significance of educational level of female household heads in explaining their income function.

Table 6: ROBUST CONTROL FUNCTION APPROACH (with Interaction between residual and instrumental variable)

VariableGeneral Sample Male Subgroup Female Subgroup

1996 2001 1996 2001 1996 2001Educational Level 0.7997*** 0.7532*** 0.9191*** 0.7547*** 0.4012* 0.7303***

(5.18) (10.61) (5.51) (9.68) (1.67) (6.67)Age 0.0637*** 0.0527*** 0.0701*** 0.0528*** 0.0634** 0.0484***

(7.12) (18.6) (7.1) (16.2) (2.29) (8.32)Age Squared*102 -0.0503*** -0.0402*** -0.0528*** -0.0378*** -0.0656* -0.0451***

(-4.78) (-13.75) (-4.62) (-11.39) (-1.85) (-7.34)Household Size -0.0336*** -0.0283*** -0.0232*** -0.0251*** -0.0552*** -0.0520***

(-5.2) (-10.05) (-3.24) (-8.53) (-3.83) (-7.7)Fraction of Active Household Members 0.1900** 0.1117*** 0.3713*** 0.1197*** -0.0567 0.1493**

(2.21) (3.24) (3.72) (3.19) (-0.40) (2.27)Primary Sector(PSU Percentage) 0.1103** -0.1544** 0.1128** -0.1410* 0.1089 -0.1496

(2.32) (-2.15) (2.19) (-1.82) (0.87) (-1.24)Secondary Sector(PSU Percentage) 0.0733* 0.0824 0.1013*** 0.0992* -0.0241 0.0699

(1.82) (1.58) (2.6) (1.77) (-0.26) (0.78)Sickness_past_2_Weeks(PSU Percentage) -0.1213 -0.4562*** -0.0799 -0.4186*** -0.2930* -0.6394***

(-1.34) (-3.32) (-0.86) (-2.86) (-1.79) (-2.73)Access to Medical Doctor(PSU Percentage) 0.1205*** -0.0475 0.0764* -0.0472 0.2820*** -0.0237

(2.8) (-1.25) (1.72) (-1.30) (3.24) (-0.38)Water (PSU Percentage) 0.1680** 0.0217 0.1362* 0.0786 0.1783 -0.2341

(2.09) (0.17) (1.65) (0.63) (0.86) (-1.19)Electricity(PSU Percentage) 0.1885* -0.2815*** 0.1560 -0.3047** 0.2726 -0.1345

(1.77) (-2.00) (1.33) (-2.20) (1.12) (-0.52)Douala 0.1458** 0.0510 0.1050 0.0562 0.2189* 0.0761

(2.17) (1.26) (1.39) (1.28) (1.73) (1.18)Rural Forest -0.3785** -0.1446*** -0.4376*** -0.1229*** -0.3124 -0.1940***

(-2.3) (-3.81) (-2.71) (-3.05) (-1.13) (-3.18)Rural Haut Plateau -0.3354** -0.1957*** -0.3610** -0.1866*** -0.2574 -0.2205***

(-2.18) (-5.66) (-2.31) (-5.12) (-0.78) (-4.26)Savannah -0.2476* 0.1710*** -0.2261 0.1938*** -0.3214 0.0784

(-1.62) (3.27) (-1.53) (3.38) (-0.99) (0.99)PEduc_residual -0.8019*** -0.7463*** -0.8283*** -0.8180*** -0.4992 -0.6042***

(-4.46) (-9.46) (-4.39) (-9.35) (-1.45) (-4.64)Edu*Educ_residual 0.1308** 0.0788*** 0.1148** 0.1025*** 0.0562 0.0383

(2.45) (7.01) (2.13) (8.30) (0.31) (1.48)Constant 9.6054*** 10.1482*** 9.0557*** 10.0177*** 10.8593*** 10.7371***

(27.78) (45.08) (23.85) (39.73) (16.69) (31.63)R-Squared 0.3243 0.3236 0.3622 0.3290 0.2524 0.3483Fisher 22.77 85.90 21.42 67.81 5.74 58.80Number of Observation 1338 10134 1088 7858 250 2276Source: Computed by Authors using STATA 10. ***, ** and * are 1, 5 and 10 percent significance levels

The secondary sector is significantly biased in favor of men in explaining income expenditures, revealing the key role of this sector in enhancing disparities between

19

households. Sickness reduces household income expenditure for all samples, with a skewed significance toward the 2001 samples. Spatially, rural milieus, except the savanna region for the 2001 samples, reduce household income expenditure.

The absolute value of the coefficient on the fitted residual of education without controls for non-linear interaction significantly reduces income expenditure for all the samples and for the two years under review (see appendix). This entails linear endogenous relations that negatively affect income expenditure habits of household heads. In addition, controlling for non-linear interactions between literacy and unobservables (table 6), the value of the fitted residuals of education significantly increases for all samples except for the female sub-samples for 1996 and 2001. The statistical significance of this variable for the overall and male sub samples indicates evidence of heterogeneity of response of education on income expenditure. In this regards, the Control Function modeling has the advantage over other modeling approaches because it is capable of purging out most structural parameter of potential econometric problems, notably linear and non-linear endogeneity and heterogeneity of unobservable with the endogenous variable. Thus in the next section, undertaking the Oaxaca-Blinder approach we use results computed from this approach, because of their robustness.

B) Oaxaca-Blinder Decomposition Analysis

In this section we use unbiased results obtained from the econometric specifications above to compute gender neutral and gender biased characteristics that account for gender household disparities. Using estimates from the CFA, we firstly, analyze the pooled data comprising of the global neutral and personal characteristics to evaluate overall disparities; secondly compute gender neutral and gender-biased characteristics for both years as well as their inter-temporal values; we also compute percentage shares of our decomposed Oaxaca-Blinder formulation as well as discrimination coefficient and their absolute contributions.

Pooling both samples for the 1996 and 2001 (Table 7), neutral characteristics, considering the OLS, 2SLS and Control Function (CFA) approaches reveal that, educational level, age, household size, fraction of active adult household members, the primary sector, being sick (health variable), and access to electricity (except for the OLS method) increases inter-household gender disparity. Neutral endowments that decrease household inequality are the square of age (experience), working in the secondary sector, access to a medical doctor and infrastructural setting such as portable water. Water reduces neutral disparities because accessing this facility, enables households headed by both sexes to prevent extra expenditure on issues such as falling sick, extra time spent to fetch water from distant localities which can be used for other income generating activities, etc which augment disparities, particularly disfavoring women. Consulting a medical doctor entails being in good health and consequently extra rewards, such as no work absenteeism, will be channeled towards income earning activities. Spatially, residing in urban (Douala) and rural areas (Rural Forest, Rural Haut Plateau and the Savanna) regions will reduce inequality between male and female headed households.

Over viewing personal specific characteristics, age, educational status, fraction of adult active household members, working in both the primary and secondary sector, being sick and consulting a medical doctor, and availabilities of infrastructural facilities like accessing portable water and electricity all reduce inter household inequality. Characteristics that increase household disparity are the square of age and household size. Spatially, while

20

residing in urban areas reduces household inequality, rural areas rather increases household disparities. Urban area constitute an adequate pool of skilled workers of both sexes, adequate infrastructure and potential industries, generating employment opportunities for educated household heads, consequent to income generating possibilities for male and female residents. Controlling for unobserved variables, the interaction between the predicted residual with the educational level of the household head increases household inequality as per neutral characteristics, exception being for the latter term which is negative for personal characteristics.

Table 7: POLLED GENDER-NEUTRAL AND GENDER-BIASED CHARACTERISTICS Neutral characteristics Personal specific characteristics

OLS IV 2SLS CFA OLS IV 2SLS CFA

VariablesEducational Level 0.0803 0.2501 0.2452 -0.0925 -0.1170 -0.0629Age 0.2542 0.2474 0.2341 -0.6895 -0.3390 -0.3150Age Squared -0.254 -0.2055 -0.1913 0.3194 0.1796 0.1728Household Size 0.0067 0.0066 0.0065 0.0442 0.0227 0.0249Fraction of Active Household Members 0.0089 0.0198 0.0189 -0.0508 -0.0371 -0.0272Primary Sector(PSU Percentage) 0.1017 0.0214 0.0134 -0.3922 -0.2079 -0.1885Secondary Sector (PSU Percentage) -0.0114 -0.0308 -0.0325 -0.0189 -0.0017 0.0038Sickness_past_2_Weeks (PSU Percentage) 0.1691 0.1634 0.1604 -0.1085 -0.1452 -0.1430Access Medical Doctor(PSU Percentage) -0.0379 -0.0204 -0.0207 -0.1034 -0.0833 -0.0830Water (PSU Percentage) -0.0745 -0.0542 -0.0532 -0.0532 -0.0555 -0.0575Electricity(PSU Percentage) -0.1541 0.0264 0.0249 -0.0089 -0.1747 -0.1732RegionsDouala -0.0209 -0.0092 -0.0110 -0.0111 -0.0252 -0.0248Rural Forest -0.0127 -0.0112 -0.0108 0.0989 0.0578 0.0559Rural Haut Plateau -0.0522 -0.0396 -0.0390 0.0651 0.0332 0.0369Savannah -0.0156 -0.0025 -0.0020 0.0912 0.1053 0.1068Control for unobserved variablesPEduc_residual 0.0459 0.0215Edu*Educ_residual 0.0032 -0.0586ConstantSource: computed by authors.

B-1) Computing Gender-Neutral and Gender-Biased Characteristics

Undertaking a gender neutral and gender biased via the CFA analysis for 1996 (Table 8), we observe that in 1996 gender-neutral characteristics that exacerbated inter-household gender disparities are educational level, age, the fraction of Active adults, working in the primary and secondary sector, being sick and residing in Douala. Gender-biased inter-household enhancing endowments are education, age, the square of age, household size, the fraction of Active adults, working in the primary and secondary sector, being sick, consulting a medical doctor, access to portable water and electricity and residing in the savanna regions. Spatially,

21

while gender-neutral urban localities contributed in increasing gender disparities, rural areas help decrease inter-household gender differences. Gender-biased localities that increase between household gender inequalities were Douala, Rural Forest and Rural Haut Plateau regions.

Table 8: GENDER-NEUTRAL AND GENDER-BIASED CHARACTERISTICS via the CFA FOR 1996 and 2001Gender-neutral characteristics Gender-biased characteristics

CFA96 CFA01 CFA Inter Temporal

CFA96 CFA01 CFAInter Temporal

VariablesEducational Level 0.0731 0.2332 0.0350 0.8967 0.0523 0.2264Age 0.0936 -0.1659 0.0303 0.2815 0.1946 0.0187Age Squared -0.0756 0.1297 -0.0328 0.2490 0.1567 0.0406Household Size -0.0748 -0.0503 0.0021 0.1697 0.1295 -0.0302Fraction of Active Household Members 0.0040 -0.0013 -0.0004 0.1658 -0.0146 0.0430Primary Sector(PSU Percentage) 0.0403 -0.0051 -0.1022 0.0040 0.0021 -0.0096Secondary Sector (PSU Percentage) 0.0037 0.0010 0.0049 0.0771 0.0045 -0.0711Sickness_past_2_Weeks (PSU Percentage) 0.0008 0.0004 0.0017 0.1591 0.0426 -0.2402Access Medical Doctor(PSU Percentage) -0.0243 0.0005 0.0320 -0.1404 -0.0042 0.1157Water (PSU Percentage) -0.0103 0.0005 0.0171 -0.0291 0.0500 -0.1439Electricity(PSU Percentage) -0.0204 0.0036 0.0483 -0.0761 -0.0267 0.1423RegionsDouala 0.0025 0.0004 -0.0020 -0.0246 -0.0020 0.0155Rural Forest -0.0051 0.0004 0.0025 -0.0147 0.0107 -0.0018Rural Haut Plateau -0.0053 0.0115 -0.0042 -0.0120 0.0076 -0.0077Savannah -0.0154 0.0107 0.0555 0.0099 0.0192 0.0132Control for unobserved variablesPEduc_residual -0.1466 -0.2191 -0.0251 0.0228 0.0181 0.0085Edu*Educ_residual 0.0382 0.0499 -0.0174 0.0112 0.0226 0.0198ConstantSource: computed by authors. In 2001, age, household size, fraction of active adult members, working in the primary sector and the predicted education level of household heads were gender neutral characteristic the decrease inter-household gender inequality. Personal specific characteristics that fill inter-household gender gaps were the fraction of active adult members, consulting a doctor (which reflects health), access to basic facilities such as electricity and residing in Douala. Along residing localities, while both urban and rural areas enhance differences that exist between male and female household heads by gender-neutral criterion, basing oneself on personal specific characteristic, urban areas help bridge income expenditure gaps between men and women, unlike rural localities that dampen women ability to fill the gap that exist between them and their male counterparts.

Inter-temporally determining gender-neutral and gender-biased characteristics, individual and household gender neutral endowments that fills income expenditure gaps between men and women are education, age, household size, working in the primary and secondary sectors, consulting a medical doctor, being ill, access to water and electricity, as well as the predicted education residual and the interaction of this term with education. Gender-biased determinants than rather worsens inter-household gender inequality are education, age, the square of age, fraction of active adult members, access to electricity and being able to consult a medical

22

doctor. Douala and the Rural Forest are areas that decrease gender disparities along gender expenditure discrimination characteristics. In the Rural Forest and Rural Haut Plateau regions person characteristics as per each household head do not increase household gender disparities.

B-2) Computing the Oaxaca-Blinder Decomposition and the Discrimination Coefficient

In this section we decompose inter-household gender inequality and compute the rate of discrimination between men and women that reveal the percentage change in income/expenditure women could effectuate given they have the same attributes of income expenditure as men (Table 9). In 1996 the decomposing the Oaxaca-Blinder function reveals that the educational status of the household head overwhelmingly accounts (60%) for inter-household gender disparities. This was followed by age, the square of age (experience) and the fraction of active household members. The least contributions were access to medical consultation by a doctor and the accounted unobserved linear endogeneity component. Regionally, Douala had the least contributions and the Savanna region the most. Urban/rural wise, rural areas attenuate household gender inequality more than urban areas. Policy implicating that government should skew their policy towards these localities.

In 2001, education, age squared, household size and the unobserved non-linear variables globally explained income expenditure differences that exist between male and female headed households. Spatially, while Douala contributed solely contributed in bridging this gap, all rural areas marginally account for increase this gap with the Savanna region being the most rural contributive milieu in increasing household gender inequality.

Inter-temporal decompositions highlight that education more than overwhelmingly explain the income gap disparity between 2001 and 1996. Other variables are age, fraction of active household members, electricity and access to a medical doctor. The least contribution are being sick and access to portable water. Regionally, while the rural Savanna and Rural forest worserned inter-household gender inequality within the period under review, Rural Haut Plateau rather improved this disparity across time. Living in urban areas has tended to bridge disparities over time.

23

Table 9: Decomposition of the Gender Well-Being & Discrimination CoefficientsDecomposition of the Oaxaca-Blinder coeffecient Discrimination coefficient (%)

Total 96 Total 01 Total Inter Temporal

Disc96 Disc 01 Disc Inter Temporal

VariablesEducational Level 0.9698 0.2855 0.2613 145.1 5.36 25.4

(59.5%) (43.0%) (141.6%) (0.595) (0.076) (1.108)Age 0.3752 0.0287 0.0490 32.5 21.4 1.88

(23.0%) (4.3%) (26.5%) (0.133) (0.303) (0.082)Age Squared 0.1735 0.2864 0.0078 28.27 16.9 4.14

(10.6%) (43.2%) (4.2%) (0.116) (0.239) (0.181)Household Size 0.0948 0.0792 -0.0282 18.4 13.8 -2.97

(5.8%) (11.9%) (-15.3%) (0.075) (0.195) (-0.130)Fraction of Active Household Members 0.1698 -0.0160 0.0427 18.0 -1.44 4.31

(10.4%) (-2.4%) (23.1%) (0.074) (-0.020) (0.188)Primary Sector(PSU Percentage) 0.0443 -0.0030 -0.1118 0.40 0.21 -0.95

(2.7%) (-0.4%) (-60.6%) (0.002) (0.003) (-0.041)Secondary Sector (PSU Percentage) 0.0807 0.0055 -0.0661 8.0 0.45 -6.86

(5.0%) (0.8%) (-35.8%) (0.033) (0.006) (-0.299)Sickness_past_2_Weeks (PSU Percentage) 0.1599 0.0429 -0.2384 17.2 4.35 -21.3

(9.8%) (6.5%) (-129.2%) (0.071) (0.062) (-0.929)Access Medical Doctor(PSU Percentage) -0.1647 -0.0037 0.1478 -13.09 -0.42 12.26

(-10.1%) (-0.6%) (80.0%) (-0.054) (-0.006) (0.535)Water (PSU Percentage) -0.0395 0.0506 -0.1268 -2.86 5.10 -13.4

(-2.4% (7.6%) (-68.7%) (-0.012) (0.072) (-0.584)Electricity(PSU Percentage) -0.0965 -0.0231 0.1906 -7.32 -2.63 15.29

(-5.9%) (-3.5%) (103.2%) (-0.030) (-0.037) (0.667)RegionsDouala -0.0222 -0.0016 0.0135 -2.49 -0.21 1.56

(-1.4%) (-0.2%) (7.3%) (-0.010) (-0.003) (0.068)Rural Forest -0.0198 0.0110 0.0007 -1.46 1.07 0.17

(-1.2%) (1.7%) (0.4%) (-0.006) (0.015) (0.007)Rural Haut Plateau -0.0174 0.0192 -0.0118 -1.19 0.76 -0.76

(-1.1%) (2.9%) (-0.064%) (-0.005) (0.011) (-0.033)Savannah -0.0056 0.0299 0.0687 0.99 1.93 1.32

(-0.3%) (4.5%) (37.2%) (0.004) (0.027) (0.058)Control for unobserved variablesPEduc_residual -0.1239 -0.2010 -0.0166 2.32 1.82 0.85

(-7.6%) (-30.3%) (-9.0%) (0.010) (0.026) (0.037)Edu*Educ_residual 0.0494 0.0725 0.0023 1.12 2.28 1.99

(3%) (10.9%) (1.3%) (0.005) (0.032) (0.087)Log Total Expenditure Per Adult gap/ Discimination Index 1.631 0.6632 0.1846 243.89 70.73 22.93

(100%) (100%) (100%) 1.00 1.00 1.00Computed: By Authors. While the values in brackets below the Oaxaca-Blinder decomposition translate the percentage contributions to total gender income expenditure gap, the brackets below the discrimination coefficients show the absolute contributions of this value.

Discrimination coefficients for 1996 show that in absolute terms equating male and female educational attainment via policies that bridge inequality in education and schooling schemes will increase income earned by women by more than 60%, considerably reducing disparities in household income expenditures between men and women. In this vein, age, experience,

24

household size, increasing the number of active working adults in female headed households, working in the primary and secondary sector will decrease income expenditure gaps between men and women. Accessing water and electricity will very marginally, increase household gender disparities, revealing that this is not a significant discrimination factor in household gender disparity in Cameroon.

In 2001, education, age, age square, household size, primary and secondary sectors, access to water all helped fill the gender income expenditure gap. Fraction of active adult household members, ability to consult a medical doctor and electricity marginally augments inequality between male and female headed households. Spatially while urban areas marginally enhanced gender disparity, rural areas marginally decreased gender income expenditure gap disparity.

Between 1996 and 2001, education stands out as the main income expenditure equalizing endowment between male and female headed households. In addition, age, experience, fraction of active adult household members, access to medical doctors and electricity all reduce inter-household gender inequality. Within the period under review, household size has tended to exacerbate disparities in income expenditures between men and women. Also, the primary and secondary sector, as well as access to water increase household gender inequality. Spatially, residing in Douala, the Rural Forest and the Savanna regions overtime have tended decrease income expenditure between male and female headed households. Both unobserved control variables for both years, as well as over time have all help bridge gender disparities, revealing the possibility of unaccounted for variable that can also retreat income expenditure gaps.

Conclusion

In this study using appropriate econometric techniques that purges out potential bias of estimated results such as linear and non-linear endogeneity and heterogeneity among clusters, we attempted to develop an approach that can be used to investigate inter-temporal and spatial inter-household gender inequality in Cameroon, identifying (1) gender-neutral and gender-bias factors that influence household income/expenditure in Cameroon and (2) determine inter-temporal and spatial discrimination coefficients. Overall, both personal specific and gender discrimination expenditure characteristics of household heads linked to their gender orientation shapes the patterns of inequality observed between these two groups, as well as the spill-over effects of the gender type on household income expenditure.

Regressed variables were consistent with economic literature and significant in determining the global and sub gender samples determinants of household income expenditure for the Ordinary Least Square, Two Stage Least Square and Control Function approach. Controlling for potential endogeneity that could produce biased estimates; we instrumented the household welfare function by the educational status of household head. Complementary hypothesis enunciated in our methodology was verified with no reason attaching directly the influence of household income expenditure and educational status of household head, rather there are spill-over effects of education on household heads capability of generating further income.

25

Basing ourselves of each group, variables differ as per estimation procedures in explaining household income expenditure. However, educational status is crucial when exacerbating inter household gender inequality. In addition variables such as the age, the age squared (experience), household size, fraction of active adult household members, sectors of activity, access to drinking water and electricity as well as residential milieu influences both household types.

Gender neutral and gender biased results for the 1996 and 2001 respective sub samples reveal similar results when compared with the pooled or global decomposition of the sample sets of the respective years. Urban milieus tend to generally increase gender disparities as per both gender neutral and gender-biased characteristic, with rural areas global revealing the opposite trends. Inter-temporally computing gender expenditure and personal specific characteristics, overall, gender-biased characteristics of household heads tend to exacerbate inter-household gender inequality.

Decomposing gender income expenditure disparity, education largely accounts for yearly and inter-temporal household gender inequality. Other variables reveal varying trends as per the year considered or within the period under review. The discrimination index corroborates with the decomposition results with education being highly equalizing as concerns income expenditure gender gaps.

Although a series of policy conclusion could be drawn from the estimation results, the following policy implications stand out; the government should put in place program-and-project that curb gender inequality, strategizing on mechanisms that bridge educational differences between men and women, equate residential milieu disparities, put in place mechanism that control household sizes, regulate employment in the primary and secondary sectors towards encouraging women to get employed, etc., all in view of bridging inter-household gender differences that could fuel moderate and persistent poverty status.

26

Bibliography

Atkinson, A. B. (1997). ‘Bringing Income Distribution in from the Cold’. Economic Journal, 107: 297–321.

Atkinson, A. B., and F. Bourguignon. (2000). ‘Introduction: Income Distribution and Economics’, in A. B. Atkinson and F. Bourguignon (eds.), Handbook of Income Distribution Volume 1. Amsterdam: North-Holland, 1–58.

Balmori, H.H. (2003). Gender and Budget: Overview report. Institute of Development Studies. February 2003.

Baye, M. F. (2006). “Growth, Redistribution and Poverty changes in Cameroon: A Shapley Decomposition Analysis”. Journal of African Economies, 15, No.4, pp. 543-570.

Baye, M. F. and Fambon, S. (2009). “Linking Parental Education, Child Health and Economic Well-being in Cameroon.” Paper presented at the Centre for the Study of African Economies (CSAE) Conference, 2009 at St Catherine’s College, Oxford, 22-24 March, 2009.

Baye, M.F. (2007). “Exact Configuration of Poverty, Inequality and Polarization trends in the Distribution of Well-being in Cameroon.” Final report, AERC Nairobi, Kenya.

Blackden, M. and C., Bhanu. (1998). “Gender, Growth and Poverty Reduction in Sub- Sahara Africa.” SPA Report, Executive Summary. Washington, The World Bank.

Blinder, A.S. (1973). ‘Wage Discrimination: Reduced Form and Structural Estimates.’ Journal of Human Resources, 8:436-55.

Boserup, E. (1970). “Women’s Role In Economic Development.” London: George Allan and Unwin.

Cagatary, N., M. Keklik, R.. Lal and J., Lang. (2000). Budget as if people mattered: democratizing Macroeconomic Policies. United Nations development Program/ SEPED Conference Paper series.

Canadian International Development Agency (CIDA). (1989). “Women in Development: A Sectoral Perspective.” Hull: Public Affairs Branch, Department of External Relations and International Development.

Card, David. (2001). ‘Estimating the return to Schooling: progress on some Persistent econometrics Problem’. Econometrica, 69(5): 1127-1160.

Chameni, C. N. (2005). “A Three component Subgroup Decomposition of the Hirschman-Herfinduhl Index and Households Income Inequality in Cameroon|” Applied Economic Letter, 12, 941-947.

Chzhen, Y. (2006). “Occupational Gender segregation and Discrimination in Western Europe.” Centre for Research and Social Policy (CRSP). Loughbourgh University, UK.

Department For International Development, (DFID). (2007) Gender Equality. November, 2007.

Duclos, J.Y., Araar, A., and Fortin, C. (2003). “DAD3.4-A: Software for Distributive Analysis/ Analyse Distributive” MIMAP program, International Development research Center, Government of Canada and CREFA, University of Laval.

Endeley, J. B., and F., Sikod. (2005). “The Impact of the Chad-Cameroon Pipeline operations on Gender Relations, Land Resources and Community Livelihood.” University of Buea, Cameroon. Unpublished Research Report.

Epo. B. N. (2006). ‘Implications of Economic Growth and Redistribution to Poverty Alleviation in Cameroon: A Shapley Value Decomposition Analysis’. A dissertation presented in partial fulfillment of a DEA in Mathematical Economics and Econometrics. University of Yaoundé II, SOA, Cameroon.

27

Feldstein, H., and Jiggins, J. (eds). (1994). Tools for the Field: Methodologies Handbook for Gender Analysis in Agriculture. Connecticut: Kumarian Press.

Filmer, D., E. M., King and L., Pritchett. (1998). “Gender Disparity in South Asia: Comparison Between and Within Countries.” World Bank Research Committee GrantNo.RPO581-26. www.worldbank.org/research/projects/gender/gensa.htm.

Fonchingong, C. (1999). “Structural Adjustment, Women and Agriculture in Cameroon.” In Oxfam, Gender and Development, 7(3), pp. 73 – 79. Oxford: Oxfam, GB.

Fonjong, L. (2001). “Fostering Women’s Participation in Development through Non-Governmental Efforts in Cameroon.” The Geographical Journal of the Royal Geographical Society, 167 (3), pp. 223-234. UK, Blackwell Publishers Ltd.

González M.P, Santos M.C., and S.L., Delfin. (2005) “Gender wage gap in Portuguese: recent Evolution and Decomposition”. Research Center on Industrial, Labour and Managerial Economics. DP 2005 – 05, October 2005.

Government of Cameroon, (1996). Data Base of ECAM I, Ministry of Economy and Finance.Government of Cameroon, (2003). The Poverty Reduction Strategy Paper, Ministry of

Economic Affairs, Programming and Regional Development.Guijt, I., and Shah, M. (eds). (1998). The Myth of Community: Gender Issues in Participatory

Development. London: Intermediate Technology Publications.Institue Nationale des Statistiques. (2002b). Evolution de la Pauvreté au Cameroun entre

1996 et 2001. DecembreDécembre 2002.Institue Nationale des Statistiques. (2008). Tendance, Profil et Déterminant de la Pauvreté au

Cameroun en 2007. Disponible sur www.statistic-Cameroun.org.Institue Nationale des Statistiques.(2002a). « ECAM II : Document de Méthodologie »,

Yaoundé. Jeffery, P., and A., Basu (eds.). (1998). Appropriating Gender: Women’s Activism and

Politicized Religion in South Asia. London: Routledge.Kabeer, N. (1994). “Reversed Realities: Gender Hierarchies in Development Thought.”

London: Verso.Kabeer, N. (1995). “Targeting Women or Transforming Institutions? Policy Lessons from

NGO antipoverty efforts.” Development in Practice,5 (2): 108-116.Kabeer, N. (2003). Gender Mainstreaming in Poverty Eradication and the Millennium

Development Goals. A Handbook for Policy-makers and the Stakeholders. London. Commonwealth Secretariat.

Klassen, S. (2005). “Pro-poor Growth and Gender: What can we Learn from the Literature and the OPPG Case Study?”. Discussion Paper by Stephan Klassen, University of Göttingen to the Operationalizing Pro-poor Growth (OPPG) Working Group of AFD, DFID, BMZ (GTZ/KfW) and the World Bank.

Lambert, P. J. and J. R. Aronson (1993) “Inequality Decomposition Analysis and the GiniLevy, C. (1996). “The Process of Institutionalizing Gender in Policy and Planning: The Web

of Institutionalization”. Working Paper № 74. London: University College, Development Planning Unit.

Lissenburgh,S. (2000). “Gender Discrimination in the Labour Market: Evidence from BHPS and EIB Surveys.” PSI Research Discussion Paper, 3. Policy Research Institute, 2000.

Locke, C., and C., Okali,. (1999). “Analyzing Changing Gender Relations: Methodological Challenges for Gender Planning.” Development in Practice, 9(3): 274-286.

MacCormack, C., and Strathern, M. (eds). (1980). Nature, Culture, and Gender. Cambridge: Cambridge University Press.

Moser, C. (1989). “Gender Planning in the Third World: Meeting Practical and Strategic Gender Needs.” World Development, 17 (11): 1799-1825.

28

Moser, C. (1993). Gender Planning and Development: Theory, Practice and Training. London: Routledge.

Moser, C., and A., Felton. (2006). “The Construction of an Asset Index: Measuring Asset Accumulation in Ecuador.” Brooking Institution/ University of Maryland. Mimeo.

Moser, C., and Levy, C. (1986). “A Theory and Methodology of Gender Planning: Meeting Women’s Practical and Strategic Needs.” London: Development Planning Unit, University College.

Mosse, D. (1994). “Authority, Gender and Knowledge: Theoretical Reflections on the Practice of Participatory Rural Appraisal.” Development and Change, 25 (3): 497-526.

National Institute of Statistics. (2007) Enquête Camerounaise Auprès des Ménages (ECAM III). National Institute of Statistics, 2007.

National Institute of Statistics. (August, 2005). Les Comptes Nationaux Du Cameroun 1993-2004. Cameroun.

National Institute of Statistics. (October, 2004). Annuaire Statistique du Cameroun. Edition 2004. Tome III.

Ngome, A., N. (2003). “Gender Division of Labour and Women’s Decision-Making Power in Rural Households: The Case of Mbalangi, Ediki and Mabonji Villages of Meme Division.” An Unpublished Master’s Thesis, University of Buea, Department of Women and Gender Studies.

Oaxaca, R. (1973). ‘Male-Female Wage Differences in Urban Labour Markets.’ International Economic Review 14(3): 693-709.

Oaxaca, R. and Ransom, M. (1994). On Discrimination and the Decomposition of Wage Differentials. Journal of Econometrics, 61:5–21.

Oaxaca, R. and Ransom, M. (1999). “Identification in detailed wage decompositions.” Review of Economics and Statistics, 81(1):154–157.

Papapetou, E. (2005). “Gender Wage Differential in Greece”. University of Athens, Athens, Greece.

Paternostro, S. and D. E., Shan. (1998). “Wage Determination and Gender Discrimination in a Transition Economy: The Case of Romania.” University of Cornell, USA.

Pena-Boquete, Y. and F. Melchor. (2006). “A Comparative Analysis of the Evolution of Gender Wage Discrimination: Spain vs. Galicia.” European regional Science association, conference paper ersa06p340, www.ersa.org.

Poats, S., Schmink, M., and A., Spring,. (eds). (1988). Gender issues in farming systems research and extension. Boulder: Westview Press.

Poverty Reduction Strategy Paper (PRSP), (2000). Interim Poverty Reduction Strategy Document, Republic of Cameroon, August 2000.

Poverty Reduction Strategy Paper (PRSP), (2003) Poverty Reduction Strategy Document, Republic of Cameroon, August 2003.

Seeley, J., R., Grellier and T., Barnett. (2004). “Gender and HIV/AIDS Impact Mitigation in Sub-Saharan Africa-Recognizing the Constraints.” Journal of Social Aspects of HIV/AIDS, 1, No. 2, 87-98. August 2004.

Sen, A., and S., Anand. (1995). “Gender Inequality in Human Development: Theories and Measurement.” Background Paper for the Human Development Report 1995, New York: Human Development Report Office.

Sikod, F. (2007). “Gender division of Labor and Women’s Decision-Making Power in Rural Households in Cameroon.” Africa Development, 32, No. 3.

Sokoloff, K.L. and S.L. Engerman. (2000). “Institutions, Factor Endowments, and Paths of Development in the New World.” Journal of Economic Perspective, 14, No. 3, 217-32.

29

Takahshi, K. (2007). “Sources of Regional Income Disparity in Rural Vietnam: Oaxaca-Blinder decomposition”. Discussion Paper No. 95. Institute of Developing Economies, JETRO, Japan

UN Millennium Development Goals Indicator Database (2007). United Nations.UNDP Cameroon Office. (2002). The Republic of Cameroon. Country Report. Millennium

Development Goals Progress. YaoundéUNDP Cameroon Office. (2003). The Republic of Cameroon. MDG Progress Report at

Provincial Level. Yaoundé.UNESCO. (2007). Education For All. Global Monitoring Report, 2007.United Nations, The Millennium Development Goals Report (2006).Vartiainen, J. (2002). “Gender wage differentials in the Finnish Labour Market” Ministry of

Social Affairs and Health Finland.Wang, M. and F., Cai. (2008). “Gender Earnings Differential in Urban China.” Review of

Development Economics, 12(2); 442-454, 2008.Wooldridge, J. M. (1997). ‘On Two Stage least Squares estimation of average treatment effect

in a Random coefficient Model’. Economics letters, 56, 129-133.World Bank. (2001). “Country Assistance Evaluation: Cameroon”. (January 24, 2001).

Operation Evaluation Department. Zuidberg, L. (1994). “Integrated Rural Development: for Whom and with Whom?” In

Assessing the Gender Impact of Development Projects. Edited by Gianotten,V., Groverman, V., Van Walsum, E., and Zuidberg, L. Amsterdam: Royal Tropical Institute. 49-70.

30

APPENDIX: Table A: REDUCED FORM ESTIMATION OF EDUCATIONAL LEVELS

VariableGeneral Sample Male Subgroup Female Subgroup1996 2001 1996 2001 1996 2001

Age 0.0112 -0.0103*** 0.0145 0.0054 0.0270 -0.0460***(1.28) (-3.54) (1.48) (1.61) (1.29) (-8.34)

Age Squared*102 -0.0244** -0.0114*** -0.0278*** -0.0241*** -0.0481* 0.0191***(-2.49) (-3.82) (-2.57 (-7.07) (-1.9) (3.41)

Household Size -0.0052 0.0027 -0.0120* -0.0001*** -0.0134 -0.0128**(-0.94) (1.01) (-1.9) (-2.76) (-1.06) (-1.99)

Fraction of Active Household Members -0.0930 -0.1771*** -0.1454* -0.2114*** -0.0907 -0.2212***

(-1.43) (-5.59) (-1.85) (-5.82) (-0.8) (-3.67)Primary Sector(PSU Percentage) -0.0084 -0.2037*** -0.0109 -0.3212*** -0.0595 0.0530

(-0.21) (-4.07) (-0.25) (-5.66) (-0.64) (0.54)Secondary Sector(PSU Percentage) -0.0778*** -0.1477*** -0.0990*** -0.1914*** -0.0885 -0.1559*

(-3.1) (-3.33) (-3.6) (-3.81) (-1.38) (-1.82)Sickness_past_2_Weeks(PSU Percentage) -0.1285** 0.0964 -0.1438** 0.2395** -0.0136 -0.1531

(-2.18) (0.99) (-2.19) (2.16) (-0.11) (-0.82)Access to Medical Doctor(PSU Percentage) 0.0796*** 0.0116 0.0704** 0.0442* 0.1297** -0.0576

(2.76) (0.50) (2.2) (1.64) (2.00) (-1.38)Water (PSU Percentage) 0.0412 -0.1704** 0.0848 -0.1534* -0.1531 -0.2240

(0.81) (-2.30) (1.55) (-1.84) (-1.13) (-1.55)Electricity(PSU Percentage) -0.1362 0.2208** -0.1146 0.3209*** -0.0953 0.0655

(-1.52) (2.23) (-1.11) (2.81) (-0.51) (0.37)Douala 0.1243*** 0.2656*** 0.1569*** 0.2901*** -0.1005 0.1097**

(2.73) (9.52) (3.03) (9.26) (-1.09) (2.00)Rural Forest -0.2940*** 0.0960*** -0.2477** 0.1591*** -0.2634 -0.0344

(-2.69) (3.38) (-1.98) (4.89) (-1.2) (-0.65)Rural Haut Plateau -0.1874* 0.0049 -0.1778 0.0682** -0.1304 -0.1135***

(-1.71) (0.20) (-1.46) (2.43) (-0.51) (-2.61)Savannah -0.3700*** -0.4668*** -0.3594*** -0.4603*** -0.2505 -0.4520***

(-3.37) (-17.06) (-3.00) (-15.01) (-0.88) (-8.17)Land Surface Area 0.0233** 0.0017*** 0.0230** 0.0018*** 0.0153 0.0015***

(2.54) (5.85) (2.31) (5.47) (0.62) (2.61)Own Television (PSU Percentage) 0.4022*** 1.9556*** 0.4172*** 1.8991*** 0.3471*** 2.0381***

(8.49) (19.59) (7.7) (16.73) (3.62) (10.8)Own Radio (PSU Percentage) 0.0111 0.0352 0.0305 -0.0933 -0.2402 0.0841

(0.12) (0.25) (0.29) (-0.59) (-0.99) (0.31)Public Schools (Sum PSU squared) 0.0418*** 0.0029*** 0.0307** 0.0024*** 0.0606*** 0.0040***

(3.94) (4.71) (2.06) (3.29) (4.02) (4.00)Constant 1.604*** 2.7678*** 1.5884*** 2.4998*** 1.5198*** 3.6258***

(7.72) (37.07) (6.64) (29.38) (3.44) (25.44)R-Squared 0.2436 0.3684 0.2582 0.3655 0.3122 0.4562Adjusted R-Squared 0.2333 0.3673 0.2457 0.3640 0.258 0.4519Number of Observation 1338 10134 1088 7858 250 2276Source: Computed by Authors using STATA 10. ***, ** and * are 1, 5 and 10 percent significance

31

Table B: ROBUST ORDINARY LEAST SQUARE ESTIMATIONS

VariableGeneral Sample Male Subgroup Female Subgroup1996 2001 1996 2001 1996 2001

Educational Level 0.2885*** 0.2200*** 0.3578*** 0.2149*** 0.0298 0.2372***(7.65) (17.05) (9.25) (15.38) (0.34) (9.76)

Age 0.0752*** 0.0512*** 0.0820*** 0.0513*** 0.0686** 0.0463***(8.6) (17.65) (8.48) (15.52) (2.52) (7.77)

Age Squared*102 -0.0692*** -0.0507*** -0.0730*** -0.0488*** -0.0751** -0.0536***(-6.96) (-17.58) (-6.79) (-14.69) (-2.18) (-9.27)

Household Size -0.0365*** -0.0271*** -0.0257*** -0.0235*** -0.0585*** -0.0509***(-5.66) (-9.54) (-3.65) (-7.84) (-3.94) (-7.64)

Fraction of Active Household Members 0.1437* -0.0026 0.3287*** 0.0017 -0.1122 0.0446

(1.76) (-0.08) (3.47) (0.05) (-0.83) (0.72)Primary Sector(PSU Percentage) 0.1078* -0.4429*** 0.1161** -0.4368*** 0.0924 -0.4231***

(1.96) (-6.64) (2.01) (-6.28) (0.71) (-3.71)Secondary Sector(PSU Percentage) 0.0503 0.0045 0.0804 0.0151 -0.0590 0.0063

(1.05) (0.09) (1.63) (0.26) (-0.69) (0.07)Sickness_past_2_Weeks(PSU Percentage) -0.1774* -0.4315*** -0.1368 -0.4121*** -0.3417** -0.5708**

(-1.93) (-3.03) (-1.43) (-2.66) (-2.15) (-2.47)Access to Medical Doctor(PSU Percentage) 0.1715*** -0.0377 0.1301*** -0.0404 0.3217*** -0.0146

(4.46) (-0.98) (3.21) (-1.05) (3.92) (-0.24)Water (PSU Percentage) 0.2004** 0.0652 0.1692** 0.1214 0.2115 -0.1836

(2.53) (0.51) (2.04) (0.98) (1.13) (-0.91)Electricity(PSU Percentage) 0.2998** 0.2757** 0.2880** 0.2535** 0.3410 0.3706*

(2.39) (2.22) (2.17) (2.08) (1.32) (1.63)Douala 0.2092*** 0.1666*** 0.1771** 0.1685*** 0.2667** 0.1815***

(2.84) (3.95) (2.06) (3.58) (2.24) (3.05)Rural Forest -0.5154*** -0.1016*** -0.5997*** -0.0813** -0.3610 -0.1562***

(-3.54) (-2.66) (-4.21) (-1.99) (-1.29) (-2.56)Rural Haut Plateau -0.4789*** -0.2328*** -0.5222*** -0.2305*** -0.3564 -0.2535***

(-3.44) (-6.58) (-3.63) (-6.16) (-1.13) (-4.86)Savannah -0.4812*** -0.1235*** -0.4858*** -0.1049** -0.4783* -0.2033***

(-3.58) (-3.03) (-3.63) (-2.45) (-1.65) (-3.21)Constant 10.4053*** 11.7422*** 9.9196*** 11.6555*** 11.5091*** 12.1900***

(46.48) (128.79) (37.81) (114.87) (20.26) (68.2)R-Squared 0.3085 0.3041 0.3464 0.3069 0.2439 0.3400Fisher 23.65 86.31 22.76 65.24 5.86 63.80Number of Observation 1338 10176 1088 7888 250 2288Source: Computed by Authors using STATA 10. ***, ** and * are 1, 5 and 10 percent significance levels

32

Table C: ROBUST CONTROL FUNCTION APPROACH (without Interaction between residual and instrumental variable)

VariableGeneral Sample Male Subgroup Female Subgroup1996 2001 1996 2001 1996 2001

Educational Level 0.8353*** 0.7488*** 0.9509*** 0.7559*** 0.4168* 0.7232***(5.44) (10.28) (5.76) (9.35) (1.72) (6.63)

Age 0.0674*** 0.0556*** 0.0735*** 0.0559*** 0.0645** 0.0500***(7.47) (18.82) (7.33) (16.37) (2.31) (8.6)

Age Squared*102 -0.0538*** -0.0434*** -0.0561*** -0.0413*** -0.0666* -0.0468***(-5.12 (-14.56) -4.87 (-12.15) -1.86 (-7.7)

Household Size -0.0339*** -0.0291*** -0.0236*** -0.0258*** -0.0553*** -0.0526***(-5.23) (-10.23) (-3.29) (-8.61) (-3.84) (-7.83)

Fraction of Active Household Members 0.2107** 0.1039*** 0.3926*** 0.1086*** -0.0526 0.1453**

(2.48) (3.01) (3.93) (2.85) (-0.37) (2.23)Primary Sector(PSU Percentage) 0.1108** -0.1812** 0.1127** -0.1756** 0.1083 -0.1586

(2.32) (-2.50) (2.18) (-2.23) (0.87) (-1.31)Secondary Sector(PSU Percentage) 0.0758* 0.0716 0.1026*** 0.0823 -0.0211 0.0686

(1.89) (1.36) (2.61) (1.44) (-0.23) (0.77)Sickness_past_2_Weeks(PSU Percentage) -0.1242 -0.4643*** -0.0818 -0.4334*** -0.2920* -0.6418***

(-1.39) (-3.31) (-0.89) (-2.86) (-1.79) (-2.75)Access to Medical Doctor(PSU Percentage) 0.1203*** -0.0482 0.0768* -0.0531 0.2827*** -0.0200

(2.83) (-1.27) (1.74) (-1.42) (3.27) (-0.32)Water (PSU Percentage) 0.1671** 0.0259 0.1364* 0.0848 0.1689 -0.2300

(2.08) (0.20) (1.65) (0.67) (0.85) (-1.16)Electricity(PSU Percentage) 0.1877* -0.2865** 0.1537 -0.3297** 0.2725 -0.1204

(1.75) (-2.01) (1.31) (-2.29) (1.12) (-0.47)Douala 0.1308* 0.0345 0.0900 0.0305 0.2100* 0.0703

(1.94) (0.83) (1.17) (0.67) (1.74) (1.08)Rural Forest -0.3921** -0.1502*** -0.4522*** -0.1326*** -0.3151 -0.1965***

(-2.39) (-3.88) (-2.80) (-3.19) (-1.14) (-3.21)Rural Haut Plateau -0.3326** -0.2072*** -0.3595** -0.2076*** -0.2535 -0.2215***

(-2.18) (-5.96) (-2.31) (-5.62) (-0.78) (-4.28)Savannah -0.2541* 0.1585*** -0.2330 0.1801*** -0.3187 0.0696

(-1.66) (3.00) (-1.56) (3.09) (-0.99) (0.88)PEduc_Levresidual -0.5856*** -0.5525*** -0.6316*** -0.5637*** -0.4285* -0.5172***

(-3.81) (-7.76) (-3.79) (-7.17) (-1.64) (-4.59)Constant 9.5009*** 10.1689*** 8.9598*** 10.0504*** 10.8239*** 10.7420***

(27.46) (44.17) (23.47) (38.67) (16.22) (31.62)R-Squared 0.3201 0.3150 0.3589 0.3190 0.2519 0.3474Fisher 23.61 87.49 22.12 66.43 5.94 61.42Number of Observation 1338 10134 1088 7858 250 2276Source: Computed by Authors using STATA 10. ***, ** and * are 1, 5 and 10 percent significance

33