The Nexus between Poverty, Inequality and Growth: A Case Study of Cameroon and Kenya

Abstract

1. Introduction

While economic growth is generally accepted as a condition for poverty reduction, a good understanding of the nature, sources and quality of growth required to reduce poverty is lacking. Thus, exploring the determinants of household well-being, as well as the impacts of regression-based well-being sources¹ on social inequalities, and growth inclusiveness, would be a useful input into the ongoing policy debates on pathways to poverty reduction in African countries. These issues are often dealt with in isolation, but our present endeavor is arguably a new attempt at addressing poverty and inequality comprehensively.

We contextualize and concretize these issues with a case study of Cameroon and Kenya, the leading economies in the Central African Economic and Monetary Union² and the East African Community (African Development Bank, 2020a, 2020b), respectively. In SSA countries, the downsizing of social spending that started in the early 1990s and continued into the 2000s to attenuate macroeconomic imbalances might have led to the current social disparities in Cameroon and Kenya. Economic and social policies in these countries are now directed at reducing disparities in opportunities, while focusing on inclusive growth. Evidence on impacts of household endowments on well-being, inequalities and inclusive growth is needed to sustain debate and implementation of these policies.

In terms of welfare indicators, the poverty headcount ratio at national poverty lines in Kenya fell by 10.7 percentage points from 46.8% in 2005 to 36.1% in 2015. The Gini inequality fell by 5.7 percentage points from 46.5% to 40.8% between 2005 and 2015 (Kenya National Bureau of Statistics, 2005, 2015). In the case of Cameroon, the poverty rate fell by 2.4 percentage points from 39.9% in 2007 to 37.5% in 2014, meanwhile, the Gini inequality increased by 5 percentage points from 39% in 2007 to 44% in 2014 (National Institute of Statistics, 2014). These differential trends make the assessment of the poverty–inequality–growth nexus in these two countries an interesting and informative case study.

Without tackling inequality, the economic growth rates required to eradicate poverty among African countries may not be achievable. However, inequality, per se, may not be a ‘bad thing’ because some effort-based inequality is essential for freedom of choice, which itself is a source of well-being (Sen, 1993, 1999). Therefore, it is useful to understand the nature, sources and consequences of inequality in order to design and implement effective and socially acceptable growth and anti-poverty policies. The phenomenon of jobless growth reported in African countries has deepened income inequalities, a situation that calls for growth inclusiveness. Identification of a household economic well-being generating function for each country and accounting for inequality of this well-being that is driven by circumstance-based and effort-based variables, is the first contribution made in this paper.

The other contribution is the conceptualisation of inequality of household well-being, as being generated by inequality of exogenous circumstances for which an individual household may not entirely be held responsible, in contrast to that stemming from inequality of efforts. While the conceptualisation itself is not new as such, we make it concrete using African evidence. We analyse the levels and inequalities of well-being outcomes among Cameroonian and Kenyan households, generated by differences in circumstances and efforts. We examine circumstances and efforts, as an integral pair in a well-being function, in an endeavor to provide a better understanding of the sources of a household's economic prosperity and its disparity over time and space. We argue that inequality of opportunity emanating arguably from inheritance and other exogenous conditions (Arneson, 1989 and Cohen, 1989), leads to inequality of outcomes. Equalisation of circumstance-based opportunities, as a response to inequality in the well-being outcome, has a long and veritable literature (Roemer, 1998, 2002; Bourguignon and Morrison, 2002 and Lefranc et al., 2008). However, to our knowledge, no published works exist in Africa that link inequality of household well-being to inequality of opportunity and inequality of effort. This linkage is rigorously demonstrated in this paper.

We argue that evidence on the nature and quality of growth is key to the design of desirable redistributive policies. We estimate the relative measure of pro-poor growth (Kakwani et al., 2006)³ using comparable survey data sets from Cameroon and Kenya and show that pro-poor growth strategies reduce extreme poverty in study countries. We conclude the analysis of inclusive growth by comparing income growth among the 40% least well-off households, with that of the overall population, and find sizeable shared prosperity gains for both countries, driven by human capital accumulation.

This study differs from past research on poverty and inequality in several important respects. First, while previous studies on distributive analysis largely ignore potential econometric problems associated with the estimation of the contributions of effort-based and circumstance-based variables to the well-being of a household, we fully address these issues using the control function modelling strategy explained in Wooldridge (2015). In particular, we use the approach to specify an income generating function that corrects for potential sample selection bias due to non-random participation of households in labor markets, and controls for the endogeneity of human capital regressors in the household well-being function. We also net out any unobserved heterogeneity from the parameter estimates of regressors of policy interest, particularly those associated with health and education inputs into the well-being function. Second, while previous works using SSA data are generally saddled on the growth incidence curve or on the poverty growth curve, we adopt a non-parametric relative measure of growth inclusiveness (Kakwani et al., 2006, 2010) to gauge the extent of income improvements over time and to assess success of any redistributive policies in reducing inequality. Since inclusive growth implies both participation in the growth process and fair redistribution of the fruits of growth, we extend our analysis to measure the extent of shared prosperity.

In addition to applying control functions to measure household welfare levels, we employ regression-based decompositions and simulated factual and counterfactual distributions to elicit impacts of human capital on social inequalities as measured by the Gini index. We further measure pro-poorness of growth and isolate the extent of shared prosperity by locality. In each case, we use cross-sectional household surveys for 2007 and 2014 (Cameroon) and 2005 and 2015 (Kenya).

The counterfactual distribution of total expenditures per adult equivalent (our proxy for well-being) is simulated in two steps. First, we equalize well-being/income sources of interest (e.g., effort) in the predicted factual distribution, which corresponds to the distribution of income as measured in the survey. Second, we estimate the well-being–generating function when income sources of interest are equalised for everyone, and then similarly predict a new income distribution. The simulated new income is the counterfactual distribution, in the sense that it does not exist in survey data. It is a distribution of economic well-being generated by a notional policy scenario that removes disparities from well-being sources of interest, e.g., differences due to exogenous circumstances. We compute well-being impacts of the hypothetical policy measure by comparing the Gini indices, relative pro-poor growth indices, shared prosperity indices and the Palma ratios, in the factual and counterfactual distributions.

2. Literature and knowledge gaps

An overview of the literature indicates that considerable research has been done using parametric and non-parametric methods to characterize and measure inequality (Heshmati, 2004). However, until around 2000s, economists had not seriously attempted to develop a regression-based approach to decomposing measured inequality (Fields and Yoo, 2000; Morduch and Sicular, 2002; Wan, 2004 and Bourguignon et al., 2008), as the classical papers (Blinder, 1973; Oaxaca, 1973; Juhn et al., 1993) served the purpose of the day. Despite the rapidly growing research on inequality, regression-based decomposition analysis of income inequality in Africa is sparse (Alayande, 2003; Epo and Baye, 2013; Gutema, 2019).

In Roemer's (1998) theory of equality of opportunity, the type of income source is an important construct in characterisation of the well-being distribution. The novelty of the equal-opportunity approach is the art of partitioning income differentials into two categories, the first is due to circumstances beyond the control of individuals, and the second arises from voluntary effort. Inequality studies over the past decades have been less challenging to conduct than what is accomplished here because the older investigations concentrated on observed drivers of inequalities of outcomes, neglecting influences of unmeasured or unmeasurable variables, such as innate ability or motivation of workers.

As indicated in Lefranc et al. (2008), previous studies did not give sufficient attention to identification of the kind of inequalities that are attributable to circumstances and to efforts. Most empirical works measuring inequality of opportunity have relied on the ordinary least squares (OLS) estimates to simulate benchmark distributions that equalize observed circumstance-related variables, to generate well-being distributions in which the influence of exogenous factors have been eliminated. The simulated inequalities are then compared with inequalities of observed outcomes to isolate the effect of opportunity equalisation on inequality of outcomes (Bourguignon et al., 2007; Nunez and Tartakowsky, 2007). However, the earlier studies fail to use methods that adequately address commonly encountered estimation problems, such as endogeneity, sample selectivity and unobserved heterogeneity (Wooldridge, 2002; Schultz and Strauss, 2008; Mwabu, 2009; Baye and Epo, 2015; Wooldridge, 2015).

Pro-poor growth has broadly been defined by a number of international organisations as growth that leads to significant reduction in poverty. In the literature, absolute (Ravallion and Chen, 2003) and relative (Kakwani and Son, 2003; Son, 2004) pro-poor growth have been the main strands of distributive analysis. The construct of pro-poor growth is embedded in the concept of income elasticity of poverty (Essama-Nssah and Lambert, 2006). It is evident from a review of the literature on pro-poor growth that there has been a vigorous search for measures that can indicate whether growth has been pro-poor or not (Son, 2007). Other studies in this vein include Mohamd Shoukry et al. (2018); Grosse et al. (2008); Kraay (2006); Duclos and Wodon (2004); and Kakwani et al., (2004).

In Africa, few studies on pro-poor growth are available (Harmáček et al., 2017; Epo and Baye, 2012, 2016; Mpenya et al., 2017; Essama-Nssah and Bassolé, 2010). However, aggregate analysis of pro-poor growth, as undertaken in most studies, masks the heterogeneity in growth patterns with sources of pro-poor growth (Ravallion, 2001), complicating rather than simplifying the policy issue as to what can be done to reduce poverty.

3. Methodology

3.1. Overall framework

We assume that the household economic well-being is generated by two sets of inputs: a vector of exogenous variables that an individual cannot influence directly and a vector of endogenous variables, essentially under the control of an individual/household. Following Roemer (1998, 2002), these inputs into the well-being generating function can loosely be labelled circumstance-related and effort-related factors, respectively.

Since C₁ is a subset of C, and C minus C₁ = C₂, it follows that C₂ is a vector of the excluded variables, that is, the instrumental variables for E that help identify the effect of E on the welfare function. As suggested earlier, effort-related variables are largely attributable to individual responsibility and those related to circumstances are generally exogenous to an individual/household; ε₁ is the error term of the structural equation and ε₂ is the error term of the reduced-form equation (the first stage regression) that projects each endogenous effort-based variable onto exogenous instruments and other circumstance-based variables. Thus, effort-based variables (E) depend on circumstances, but circumstances do not depend on idiosyncratic efforts. The effort-based variables are largely endogenous due to unobserved innate ability, measurement error or simultaneous occurrence of income and effort; ε₁ is a proxy for unobserved variables in ε₂ that co-vary with effort-based variables.

The key effort-based variables of interest are the human capital related, i.e., education, health and wage employment. Meanwhile, circumstance-based variables are captured at the individual or societal levels. At the individual level, the circumstances include age, gender, location, disability, ethnicity, family socioeconomic background and inherited genetic endowments. At the societal level, they encompass public provision of healthcare and educational services, customs and conventions, climatic conditions and government regulations, as long as the unit of analysis is the individual. Most of these factors belong in vector C. Since many of these factors are not captured in our survey data, the reduced-form error term (ε₂) also absorbs the unmeasured circumstance-based variables. In this context, ε₂ becomes a critical unobservable variable to measure and predict because it fundamentally affects the well-being generating function. It is the capturing of the predicted reduced-form error term of the effort-based variables that distinguishes the present endeavor from most of the previous studies in this literature. It is worth stressing that without using the predicted reduced-form error term of the effort function (the predicted value of unobserved effort) to specify the well-being generating function, the impacts of unobserved effort on both the level and distribution of household well-being cannot be identified.

In the case of Kenya, it is the household total land holding that is in C₂. To identify the effect of sickness reporting in the survey on household well-being, we instrumented sickness with total household landholding. The underlying idea is that the size of total landholding was fixed during the reporting (survey) period. Thus, total land is exogenous to the income generating decisions of households. Since land is a good proxy for wealth, especially in agriculture, it is assumed to correlate strongly with sickness reporting, in accordance with related literatures in Asia and Africa (Akin et al., 1986; Mwabu, 1986).

In the case of Cameroon, since ε₃ and ε₄ are likely to be dependent, we use the bivariate probit model to jointly estimate equations (1c) and (1d). Subject to the usual test, we generated the inverse Mills ratio (IMR) upon estimating the joint decision of participating in the labor market and finding wage employment. In the case of Kenya, we used the simple probit estimates to generate the IMR for labor market employment. In both countries, the computed IMRs are used as additional regressors in the well-being equations.

3.2. The household well-being generating function

Concerns about unobservable variables (the potential control function variables) in the well-being generating function in terms of the circumstance versus the effort dichotomy, is an important one in some instances. For simplicity, Table 1 shows that the control function variables are entirely attributable to the effort-based component of the welfare generating function as reflected in Equations 7 and 8.

3.3. Regression-based decomposition of inequality of outcomes

3.4. Factual and counterfactual analyses to capture inequality impacts of welfare sources

Following Roemer (1998) and Bourguignon et al. (2007), we associate exogenous-opportunity with circumstance-related variables: determinants of well-being over which individual households are thought to have little or no control. We also follow Roemer (1998) and Bourguignon et al. (2007) in classifying the other determinants of well-being that can be influenced by households' decisions as endogenous effort-related variables.

In this setup, measured inequality is attributable to observed circumstance-related variables and unobserved effort-based variables since the observed effort-based variables are equalized.

If Θ_I > 0, the equalised source is inequality reducing in the factual distribution. If Θ_I = 0, the equalised source is inequality-neutral in the factual distribution. If Θ_I < 0, the equalised source is inequality augmenting in the factual distribution. |$I\left({Y}_{\overline{A}}\right)$| is an inequality index with implemented policy. The notation |${\varTheta}_I$| indicates that the explained share of inequality is conditional on the chosen inequality index. By the same token, the unexplained share of inequality is given by the complement of |${\varTheta}_I$|⁠. To address these issues empirically, we appeal to the Gini index.

3.5. Analysing growth inclusiveness

3.5.1. Pro-poor growth (Kakwani et al., 2006)

3.6. Shared prosperity analysis

In a partial sense, inclusive growth relates to pro-poor growth, which concerns the extent to which the fruits of growth benefit the poor in absolute or in relative terms. In a complete sense, it refers to the extent to which the least well-off participate in the growth process, as well as benefit from the fruits of that growth. Although to the best of our knowledge, no index has been constructed in the literature that captures the dual aspects of inclusive growth, it has been argued by the World Bank, and perhaps others—that the shared prosperity index mimics the dual aspects. The World Bank (2013) considers the least well-off to be households at the bottom 40% of the distribution of a well-being indicator. In this context, shared prosperity is the change in average well-being at the bottom 40% of the distribution of well-being between two periods. This is typically operationalised as the annualised growth rate in average well-being at the bottom 40% of the well-being distribution. In this case, there is shared prosperity premium when the annualised growth rate in average well-being in the bottom 40% of the well-being distribution is greater than the annualised overall average growth rate in well-being.

3.6.1. Measuring shared prosperity

3.6.2. Measuring the Palma ratio

Notwithstanding the attractiveness of the Gini coefficient as a measure of inequality, it is typically insensitive to the tails of a distribution. For this reason, the Palma ratio, which measures inequality between those at the top and bottom of the well-being distribution, turns out to be preferred to the Gini coefficient in the context of prosperity sharing. The World Bank measures the Palma ratio as the well-being share of the richest 10% divided by that of the poorest 40% of the distribution. The downside of the Palma ratio is that it may satisfy the transfer sensitivity principle only weakly⁵, if at all.

3.7. The data

We perform the case study analyses with survey data from Cameroon and Kenya. The third (CHCS III) and fourth (CHCS IV) Cameroonian Household Consumption Surveys data were collected by the Cameroon National Institute of Statistics in 2007 and 2014 using a similar sampling frame (National Institute of Statistics, 2007, 2008, 2014). We recognize that prices can vary across regions and over time. There is, therefore, need to harmonize the third and fourth data sets both spatially and inter-temporally. This harmonisation is carried out by using a spatial index to reflate household expenditures of the different regions to render them comparable with a referenced locality, Yaoundé. After spatially harmonising expenditure, the poverty lines for the initial and final periods are used to rendered comparable expenditures in the two surveys.

The Kenya Integrated Household Budget Surveys for 2005 and 2015 were produced by the Kenya National Bureau of Statistics [KNBS] (Kenya National Bureau of Statistics, 2005, 2015). The Kenyan household surveys were conducted using the same approaches as those used in Cameroon, as both sets of data follow the principles used in the Living Standards Measured Surveys (LSMS), pioneered by the World Bank in the 1980s (Deaton, 1989). The Kenya surveys were subjected to a similar spatial and intertemporal harmonisation as Cameroon's, mutatis mutandis. Tables 1A and 1B provide descriptive statistics for Cameroon and Kenya, respectively.

4. Findings and discussions

We present empirical results for Cameroon in Tables 2 to 13 and Kenya (Tables 14 to 25), respectively.

4.1. Results for Cameroon

Table 2 hosts estimates of reduced-form OLS and of the control function approach for the structural equation of the well-being generating function. The sample selection estimates resulting from biprobit estimates of the labor market participation and employment models are hosted in Table 2A in the Appendix. The sample statistics are in Appendix Table 1A.

4.1.1. Determinants of the well-being generating function

Reduced form parameter estimates for the education equation

Table 2, columns 1 and 2 show reduced-form estimates of years of schooling and years of schooling times year dummy. The instrumental variable for education is the non–self-cluster mean years of schooling, whereas a unit increase in neighborhood average years of schooling increases the excluded household head's educational attainment by 0.54 years, a unit increase in the proportion of neighborhood labor market participation is associated negatively with the educational attainment of the excluded household head. Being a female head or residing in rural areas correlates negatively whereas marriage and community ownership of land are associated positively with years of schooling. Household heads drawn from the 2014 survey are more schooled compared to those from the 2007 survey.

The effects of effort-based variables on household well-being

Table 2, column 3 shows OLS estimates for the well-being function that are likely to suffer from endogeneity, sample selection and unobserved heterogeneity biases. Column 4 of Table 2 presents unbiased and consistent control function estimates. The results in columns 3 and 4 of Table 2 show that education is positively and significantly correlated with household economic well-being. Since the control function variables—the residual of years of schooling, the IMR and the interaction of education with its residual are statistically different from zero; we prefer the control function estimates.

The control function estimates in column 4 of Table 2 indicate that an additional year of schooling produced a 3.4% increase in household economic well-being in 2007. Between 2007 and 2014, an additional year of schooling increased household economic well-being by 2.1%. The implication is that gains in household economic well-being emanating from a year of schooling is about 5.5% over the study period. The indirect effect of education as indicated by the interaction of education with its reduced-form residual has a positive coefficient. This shows that the reduced-form unobservable variables are complementary to education in explaining household economic well-being. Improving education would increase household well-being via better employment opportunities and rational spending strategies as articulated by Becker (1964, 1975).

Other effort-based variables included in Table 2 are formal sector employment and marriage. Column 4 of Table 2 also show that correcting for sample selectivity bias, securing formal sector employment is associated with a 15.9% increase in household economic well-being relative to that of informal sector operators. Meanwhile, household heads who are married were less endowed in terms of well-being compared with their unmarried counterparts.

Effects of circumstance-based variables

The circumstance-based variables in the income generating function are age, age squared, gender, location and regions. Column 4 of Table 2 shows that below 43 years of age any additional year associates negatively with household economic well-being, while above 43 years any additional year improves household welfare. Lack of education, precarious jobs, underemployment or outright unemployment characterize the early life course of most household heads. Contrary to the general expectation as to gender bias, the income endowment of female household heads is 9% higher than households headed by men.

In addition, rural household heads are less endowed compared with their urban counterparts. Similarly, land ownership at the community level associates negatively with household economic well-being. This finding is attributable to the observation that financiers seldom accept community land as individual collateral security to attract a credit facility due to farmer-grazer conflict risks. Relative to dwellers in the two metropolitan regions of Yaoundé—the political capital city and Douala—the economic capital city, households in the other 10 regions register household economic well-being deficits ranging from 47% in the Extreme North and Northwest regions to 16% in the Southwest region. In the case of the year dummy variable, households in the 2014 sample incurred well-being deficits in the order of 10% compared to those in the 2007 sample.

Regression-based inequality decomposition analysis

In this section, we compute the weighted contribution of the income sources for the effort component and the circumstance component in explaining total inequality using the analytical and Shapley value decomposition approaches⁶ (Table 3). We comment on the results using the Shapley value decomposition. The Shapley decomposition of the Gini shows that the effort component (0.2637 [relatively by 63%]) contributed more than the circumstance component (0.0457 [relatively by 11%]) in accounting for observed total inequality in both absolute and relative terms (Table 3, column 3). These suggest that policy orientations that target effort-based sources would have a significant impact in reducing total inequality. Decomposing the effort component into the different estimated sources indicates that education overwhelmingly contributes to explaining total inequality (0.2337 [relatively by 56%]). Labor force participation and marriage are other important determinants of inequality (Table 3, column 3). The contribution of the predicted residual term to income inequality for Cameroon was 25%. Informing political leaders on how regressed-sources can explain overall measured inequality, included variables accounted for over 75% of total inequality suggesting some level of confidence in terms of policy value of the results presented.

Table 4 shows marginal contributions of the income sources to measured overall inequality when we consider two scenarios, i.e., when we include and when we exclude a given source. For education, the 0.1980 of the overall contribution of 0.2337 to total inequality (0.4113) is realised at level 1, that is, when we exclude all the regressed-income sources. As we gradually consider level 2 to level 5, the sum of the outstanding marginal contributions of education is 0.0357 with the entries of levels 4 and 5 registering negative values. The implication here is that promoting education together with policies that curb inequality associated with circumstances and other unobserved income sources as revealed by the predicted residual would enhance the effectiveness of education in mitigating income inequality.

For the income source—formal employment, its marginal contribution in explaining inequality at level one was 66% of its share in explaining observed inequality. The remaining levels accounted for 33%. Considering marriage, the overwhelming share of level 1 in worsening total inequality withers off from levels three to five when we consider policies associated with other estimated sources.

Looking at the circumstance component, the first level of entry posts a marginal contribution of 0.1577. As we incorporate other regressed sources, it is observed that combining circumstances related with effort related sources and the predicted residuals considered in this study reduces inequality from the third to the fifth entries.

Analysis of Inequality in well-being

From the estimated parameters of the household well-being generating function, factual and counterfactual distributions are simulated to inform analysis of inequality in well-being across households. Three factual distributions are simulated—the overall distribution, circumstance-based distribution and the effort-based distribution.

Similarly, three counterfactual bench-marking scenarios are simulated under the ceteris paribus assumption.

• (1) The first counterfactual distribution of well-being is achieved by equalising circumstance-based variables, while allowing in the function, the effort-based variables as observed. Inequality due to circumstances is thereby eliminated in the counterfactual distribution. Inequality in this counterfactual distribution of well-being is entirely attributable to variations in effort-based variables. The variability in the factual distribution of well-being depends on the circumstance- and the observed and unobserved effort-related variables, whereas variation in this counterfactual distribution of well-being is attributable entirely to the observed and unobserved effort-related variables. This scenario is constructed to ensure that circumstances beyond the household control do not determine differences in life outcomes.

• (2) The second counterfactual distribution is obtained by equalising years of schooling in the overall distribution. Inequality in the resulting counterfactual distribution is free of any variations attributable to education.

• (3) The third counterfactual is achieved by equalising years of schooling in the effort-related distribution. The resulting effort-based inequality is void of education-related components.

Gini indices for the overall factual distribution and the circumstance and effort related distributions, by year

As Table 5 shows that the measured inequality as captured by the Gini coefficient was about 0.41 for the factual distribution; 0.125 (=0.159 × 0.786) for the factual circumstance distribution; and 0.286 (=0.364 × 0.786) for the overall factual effort distribution over the period 2007 to 2014. This is an indication that over the period of analysis, effort-based sources of the well-being accounted for over 70% of measured inequality. The 30% of measured inequality attributable to circumstances beyond the control of the household is non-negligible even though measured inequality is largely effort-based and perhaps justifiable. Between 2007 and 2014 measures inequality increased by 0.03 points, whereas inequality attributable to circumstances declined marginally by 0.01 points, inequality due to effort increased by 0.05 points.

Impact of circumstances on overall inequality, by location

When circumstance-based variables are equalised across households, Table 6 shows that the counterfactual Gini of 36.4% is about 4.7% significantly lower than the overall measured inequality of 41.1%. This is an indication that levelling the playing field in terms of equalisation of circumstances is inequality reducing. Although the counterfactual distribution registered lower inequalities by location than the factual distribution, the decline was only statistically significant in rural areas. These results show that over the period of study circumstance-based variables were inequality augmenting in the actual distribution. Furthermore, inequality decreases from the urban to the rural areas in both the factual and counterfactual distributions.

Impact of education on overall inequality by location, 2007–2014

In Table 7, equalisation of educational achievements (counterfactual distribution), other things being equal, reduces inequality overall, across urban, semi-urban and rural areas. Thus, improving broad-based educational opportunities does have a well-being inequality mitigating effect.

Impact of education on effort-based inequality, by location

Table 8 shows that equalising educational endowments in the effort-based distribution engenders a significant decline in effort-based inequality overall and by location. This suggests that observed educational endowments tend to reinforce effort-based inequality. Thus, leveling schooling opportunities through increases in the density of schools would be inequality reduction.

Pro-poor growth and inequality

Panels A and B of Table 9 show no relative pro-poor growth for total expenditures per adult equivalent over the period 2007 and 2014, whereas observed growth in household economic well-being expanded over the study period, poor households appear to have witnessed a well-being contraction attributable to dynamics of inequality. The contraction in expenditure per adult equivalent for the poorer households between the surveys was driven by an increase in inequality among the households at the bottom of the distribution. The decrease in poverty between 2007 and 2014 of 2.4 percentage points was not accompanied by a decline in inequality among the poor.

Eliminating the inequality associated with circumstance-based sources; increases household expenditure over the period under review. In terms of the differences in outcomes (between the factual and counterfactual growth and the pro-poor growth rates), it is observed that when we eliminate inequality due to circumstances, total expenditures for 2007 and 2014 witness a decline of 0.11 and 0.10 percentage points, respectively. Overtime, while average growth decreased by 0.01, pro-poor growth increased by 0.01 percentage points—the net effect being a gain in growth of 0.02% attributable to a decline in inequality (panels C and D of Table 9). This finding reveals the importance of improving access for poorer households to circumstanced-based opportunities and of promoting pro-poor policies.

Elimination of education-based inequality between 2007 and 2014 would have occasioned an average total expenditure declined of 0.19% and zero pro-poor growth. This indicates gains in pro-poorness of 0.19 percentage points attributable to a decline in inequality at the bottom of the distribution of well-being (see, panels E and F of Table 9).

The last panel of Table 9 shows the pro-poor analysis of the effort-based distribution. The distribution of effort witnessed pro-poor growth between 2007 and 2014, with gains in growth accompanied by decreases in inequality for households situated at the bottom of the effort distribution. The effort distribution increased slightly over the period under review when the effect of education on inequality is eliminated. This indicates that poor households are likely to experience an expansion in their effort endowments if interventions are put in place that would eliminate inequality in terms of education for households situated at the bottom of the distribution. Over the period under review, both average growth and pro-poor growth decreased by 0.58 and 0.18 percentage points, respectively—generating a net loss of 0.40 percentage points, which is attributable to an increase in inequality in the effort distribution among the least endowed. This finding reveals the importance of improving access for poorer households to educational opportunities and encouraging pro-poor policies in terms of education.

Shared prosperity analysis for Cameroon

In this section, we report results from shared prosperity analysis (Tables 10-12) and the Palma ratio (Table 13) for Cameroon and by location. Table 10 suggests that in 2007 and 2014, average household expenditure of the bottom 40% of the population was lower overall, and in rural areas in the factual distribution compared with the counterfactual distribution. In this context, circumstances are well-being reducing among the least well-off. However, urban areas posted lower values for the counterfactual distribution compared with the factual distribution. This is an indication that in urban settings circumstances were well-being augmenting in the factual distribution over the period under analysis. These results suggest that interventions that level the playing field in circumstance-related variables would increase well-being overall and raise it higher in rural than in urban areas.

Table 11 shows the shared prosperity indices nationally and by location. In particular, circumstances were shared-prosperity improving in the overall factual distribution. Looking at the urban–rural dynamics, circumstances were shared-prosperity increasing in rural areas, but shared-prosperity dampening in urban areas in the factual distributions of the least well-off households. This finding suggests that designing interventions that equalize circumstances would enhance shared-prosperity more in urban than rural settings.

Table 12 presents results of the shared prosperity premium for Cameroon. The shared prosperity premium is negative in both the factual and circumstance equalising distributions. The shared prosperity deficits are seen in both distributions (with and without the policy scenario). However, although the share of benefits from growth accruing to the population increased over the period under review, the 40% least well-off households benefited less from the country's rising prosperity. While circumstances were mitigating deficits in the overall shared prosperity, they were exacerbating the deficits in the urban and rural settings.

Gauging other dimensions of inequality between the tails of the distributions, Table 13 reports the Palma ratios associated with household expenditure for the factual and counterfactual distributions over the period 2007 to 2014. Results indicate that nationally and by location (urban versus rural) circumstances are Palma ratio increasing in the factual distribution. This is an indication that the circumstances are increasing disparities between the top and bottom of the distributions of household well-being—implying that they are inequality increasing in the factual distribution. This finding suggests that eliminating differences in circumstances would decrease the Palma ratio over time. Thus, levelling the playing field in terms of equalising circumstances would have an inequality reducing tendency between the tails of the distribution of well-being.

4.2. Results for Kenya

Tables 14 to 25 present estimation and simulation results for Kenya whilst Table 3A in the Appendix contains sample selection estimates of the wage employment status. For Kenya, the main effort variables driving the level and distribution of household economic well-being are levels of education and sickness, both of which are dichotomous. Descriptive statistics are in appendix Table 2B.

4.2.1. Determinants of the well-being generating function for Kenya

Reduced form parameter estimates for the health status model

Columns 1 and 2 of Table 14 indicate that a percentage increase in the total acreage is associated with a 3.5% reduction in the probability of reporting an illness. However, this finding is context-specific because individuals from poor households can be accustomed to common ailments and therefore report sickness less frequently than those from rich households (Akin et al., 1986).