1,465
Views
5
CrossRef citations to date
0
Altmetric
ARTICLES

Decomposing inequality and poverty in post-war Rwanda: The roles of gender, education, wealth and location

Abstract

This paper provides an overview of poverty and inequality in post-war Rwanda. Rwanda is one of the poorest countries in the world, and has recently become one of the most unequal. High levels of poverty and inequality have important implications not only in terms of evaluations of social welfare, but also for management of social tensions and the propensity for violent conflict in the future. This paper uses the first two available and nationally representative rounds of household surveys –EICV1 2000 and EICV2 2005 – to decompose and identify the major ‘sources' of poverty and inequality in the country. I find stark differences in vulnerability to poverty by region, gender and widow status of the head of household. I additionally find important changes in the ‘income generating functions' of Rwandan households, and that distribution of land and financial assets are increasingly important in determining the inter-household distribution of income.

JEL codes:

1. Introduction

Rwanda is one of the poorest countries in the world. Despite substantial economic growth since the genocide of 1994, in 2011 it ranked 115 out of 123 countries in terms of gross national income per capita (World Bank, Citation2013) and 167 out of 187 countries for the Human Development Index (UNDP, Citation2013). In 2005 over one-half of the population (57%) were considered poor and 37% were chronically poor. Unlike many other countries with substantial poverty, Rwanda is also one of the most unequal countries in the world. Since the early 1980s Rwanda has gone from levels of consumption inequality on a par with many former Soviet countries (Gini coefficient of 0.28 in 1983) to now having one of the highest levels of inequality (Gini coefficient of 0.51 in 2005), closer to levels in the most unequal countries, such as South Africa.

These high levels of poverty and inequality have important implications not only in terms of human well-being, but also for management of social tensions and the propensity for violent conflict in the future. Despite the considerable work done on the links between poverty and inequality, on the one hand, and conflict, on the other, there is a notable absence of empirical work attempting to understand the structural causes of poverty and inequality in Rwanda. Given the paucity of such analysis, this paper uses the first two available and nationally representative rounds of post-war household surveys – the Enquête Intégrale sur les Conditions de Vie des ménages, EICV1 2000 and EICV2 2005 (Rwanda Household Living Conditions Survey 2000 and 2005) – to decompose and identify the major ‘sources' of poverty and inequality in the country. To do so, the analysis draws upon the work of Foster et al. (Citation1984) for the decomposition of poverty and of Fields (Citation2003) for the decomposition of inequality, in addition to standard correlation analysis.

Given the macroscopic nature of this exercise, this analysis is necessarily simply a diagnostic. The primary focus of the paper is to look at four important correlates of consumption income and to examine their importance in determining the poverty of a household and its position in the income distribution. These correlates are human capital (proxied by education levels), asset ownership (captured by land and livestock ownership), geographical location (captured by provincial and urban residence) and, finally, gender (female and widow-headed households).Footnote2

To state briefly the more important results: national poverty rates declined slightly between 2000 and 2005, although this hides important regional differences (poverty rates increased in five out of 12 provinces). I find stark differences in vulnerability to poverty by region and gender of head of household. Additionally, for poor households human capital (such as education), while important, provides much less of a buffer against poverty than physical capital (such as cattle ownership).

I find that consumption inequality has risen substantially compared with previous comparable survey rounds done in the 1980s. Moreover, the sources of inequality between 2000 and 2005 changed markedly. The most significant change is the rising importance of regional differences. Other factors also now matter more for inequality – for example, whether any household member has savings, the dependency ratio and land holdings. These findings suggest there are important changes in the ‘income generating functions' of Rwandan households, and that distribution of land and financial assets are increasingly important in determining the inter-household distribution of income.

The rest of this paper is structured as follows. Section 2 examines changes in poverty between 2000 and 2005. I explain and use the Foster–Greer–Thorbecke (FGT) (Foster et al., Citation1984) poverty decomposition method to examine vulnerabilities by type of household, region and ownership of agricultural assets. Section 3 gives an overview of consumption inequality in 2000 and 2005. I explain and use a regression-based decomposition method developed by Fields (Citation2003) to examine the main contributing factors to changes in consumption inequality. Section 4 concludes with a summary of the findings.

2. Poverty in Rwanda, 2000 and 2005

For some time, poverty has been a pressing development issue facing Rwanda. Poverty rose sharply during the civil war and genocide and the accompanying displacement of the majority of the population (Justino & Verwimp, Citation2006). During this time a number of populations were identified as being particularly vulnerable to poverty – notably female-headed households (especially widows) and child-headed households (MINECOFIN, Citation2000, Citation2002a, Citation2002b, Citation2007). A recent paper by Koster (Citation2008) has questioned this assumption, finding female-headed households no more likely to be poor. This has brought to light a number of important gaps in our understanding of vulnerabilities to poverty after the genocide.

2.1 Data background

This study uses data from the 2000 Rwanda Household Living Conditions Survey (EICV1) and the 2005 Rwanda Household Living Conditions Survey (EICV2). Both surveys were nationally representative integrated household surveys based on the same methodology, conducted by the Direction de la Statistique in 2000/01 and the National Institute of Statistics of Rwanda in 2005/06 with donor funding and technical support. The surveys asked detailed information on living standards including household and individual socio-economic characteristics, consumption expenditures, asset and production information. EICV1 sampled 6400 households and EICV2 sampled 6900 households. In all results reported in this paper, the sample data are weighted by household level weights in order to be nationally representative.

Poverty in Rwanda is determined by the absolute poverty line of FRw 64 000 (US$144.5) that was set in 2001 based on the widely used ‘Cost of Basic Needs' method and an equivalence scale. Using data from the EICV1 it was calculated based on a basket of food and non-food commodities, reflecting consumption patterns of the three poorest quintiles, sufficient to provide 2500 kcal per adult. To calculate poverty incidence, for each household, total expenditure per annum was calculated, which was deflated by a regional price index for the relevant period to give real expenditure, and subsequently divided by an index of household size to give real expenditure per equivalent adult (McKay & Greenwell, Citation2006). An extreme poverty line of FRw 45 000 (US$101.6) in 2001 was also set, representing the level of expenditure needed if all consumption went to provide the 2500 kcal per adult. Converting the poverty lines into January 2006 prices gives an absolute poverty line of FRw 90 000 and an extreme poverty line of FRw 63 500.

2.2 Measuring poverty

Amartya Sen's seminal article on poverty pointed out the importance in understanding three dimensions to poverty – incidence, intensity, and distribution of poverty (Sen, Citation1976). Various measures to examine these dimensions have been created (see Deaton, Citation1997; Ravallion, Citation1999). An extremely useful class of indices is derived axiomatically by Foster et al. (Citation1984). FGT measures are often preferred and widely used in the literature on poverty because of their intuitive appeal and their flexibility in terms of decomposability.

Using the absolute poverty lines set by the government of Rwanda, presents the three FGT measures of poverty. There has been a decline in incidence, intensity and distribution of poverty between 2000 and 2005. A similar trend is shown with the extreme poverty line. While these national trends are encouraging, I am interested in looking at group-based dimensions of poverty, and in particular at whether there are important regional and household characteristics associated with higher levels of poverty.Footnote3

Table 1: Trends in Foster–Greer–Thorbecke measures of poverty, 2000 and 2005

With this in mind, I first examine the correlates of poverty using a discrete choice model, and then decompose the FGT index based on subgroups whose relevance is suggested by the results of the estimation.

The logistic model estimates the probability that a household's consumption income will fall below the national poverty line (FRw 64 000) based on a number of regional, household and individual characteristics:(1)

where Yi = 1 if the household's consumption income is less than FRw 64000, otherwise Yi = 0; and where:and β0 is the intercept, βi is the slope coefficients for regressors and X i ={regional, household and individual characteristics}.

The regional variables include urban/rural and province-based dummies. The household variables include structure of the household, cattle ownership, and whether the household owns and cultivates land. The individual (household head) variables include education, age, employment as farm labourer, and employment as waged worker.

presents the mean and standard deviation of the variables used in this analysis. Mean per-capita expenditures for poor households increased from FRw 86 809 (US$223) in 2000 to FRw 100 423 (US$258) in 2005 in constant 2000 Rwandan francs. The largest group of female-headed households was widows. The percentage of female widow-headed households has declined from 25% to 22% of all households between surveys. Female non-widow-headed households remained constant at 6%.

Table 2: Descriptive statistics for poor households

Poor households were more likely to have heads that had no education. Households where the head was employed as a farm labourer make up a large proportion of the poor. In 2000, 84% of households where the head was employed as a farm labourer were poor, declining only slightly to 81% in 2005. The proportion of poor households owning cattle increased from 18% in 2000 to 23% in 2005. The proportion of poor households owning land between 0.9 and 1.65 ha declined from 16% in 2000 to 14% in 2005. Correspondingly, there has been an increase in the proportion of poor households with less than 0.5 ha of land, from 57% in 2000 to 59% in 2005.

2.3 Discussion of results: household poverty estimation

Household characteristics indicating composition, size and structure of the household often show important patterns in poverty analysis. Female-headed households, younger households, and those in larger households are often more likely to be poor (World Bank, Citation2005). Households headed by women are often poorer due to labour market discrimination, barriers of access to productive assets, limited access to finance, and inequities in access to education and healthcare.

Female-headed households are more vulnerable to poverty (Sorensen, Citation1998; UNIFEM, 1998; Date-Bah, Citation2001; Lindsey, Citation2001; UNDP BCPR, 2002). In Rwanda, in addition to the barriers and discrimination just mentioned, they entered the paid labour market during a period of social upheaval and severe contraction of formal employment opportunities. After conflict, female-headed households in rural settings often face higher poverty due to widespread destruction or contestation of ownership over physical assets such as land or livestock and limited employment opportunities outside of the agricultural sector.

The majority of the Rwandan population reports widows and their families as much more likely to be destitute. The Ubedehe survey in 2006 conducted a nationwide participatory poverty analysis that recorded subjective perceptions of vulnerability to poverty. The category in the participatory poverty analysis referring to those in abject poverty was Umutindi nyakujya. The characteristics of this group were listed as follows: they had to beg to survive; they have no land or livestock, they lack shelter, clothing and food; they fall sick often and have no access to medical care; their children are malnourished and they cannot afford to send them to school.

In the Ubedehe survey, widows were overwhelmingly ranked as one of the most destitute groups in Rwanda (MINECOFIN, Citation2007). This type of exercise, while subjective, does encompass a much broader understanding of poverty beyond expenditures. It is closer to Sen's capability approach to human development as it included access to assets, health outcomes and education.

In the poverty estimation focusing solely on expenditures (see ), I find female-headed households are indeed more likely to be poor. However, in 2000 widow-headed households were 23% more likely to be poor, while in 2005 this result was no longer statistically significant. In 2000, non-widow female-headed households were more likely to be poor, although this was only statistically significant in rural areas. By 2005, non-widow female-headed households were 34% more likely to be poor. For rural non-widow female heads of household there has been a slight worsening of their likelihood of poverty from 35% to 37% between the two surveys. These results provide tentative confirmation of stagnation or even worsening of gendered structural constraints that limit income-generating opportunities.

Table 3: Household poverty estimation

The most striking change has been a decrease in likelihood of being poor for widow-headed households in rural areas. This finding may reflect a number of changes occurring in the rural sector. Firstly, there has been government commitment and strong international support for targeting of social safety net programmes (including cash transfers and land) for widows of the genocide. Secondly, a number of non-governmental organisations, such as the Association of Genocide Widows (Avega Agahozo), have arisen providing a collective voice for the needs and issues facing widows and their families. One of the most vocal issues of these groups has been women's access to their deceased husband's land.

Moving to other factors, in both 2000 and 2005 education was a significant correlate of income. Education acted as a buffer to poverty, with the effect being somewhat larger for urban households. In 2005 education appeared to have a slightly larger impact on determining poverty than in 2000.

Ownership of physical agricultural assets such as land and livestock should be important buffers to poverty in a primarily agrarian economy. I find a negative relationship between ownership and cultivation of land and poverty (increasing with the more land a household owns). This relationship is not as strong in urban areas where opportunities for income outside subsistence farming are greater. Household ownership of cattle acted as a strong buffer against poverty. In 2000 households with cattle were 70% less likely to be poor, but by 2005 this had reduced to 60% less likely to be poor. The significance of cattle as a buffer against poverty was not small – and of the same magnitude as owning land in the largest landholding category. Quantifying cattle ownership is a relatively easy task compared with gathering data about land (where quality, quantity and inputs matter). The ownership of cattle provided almost double the protection against poverty provided by primary or secondary education. This finding is troubling given the reconstruction imperative facing Rwanda. It also indicates that increasing education levels of the rural poor will be less effective without taking into account wider changes occurring in the agrarian sector, such as the concentration of physical assets.

At the regional level there may be a number of characteristics associated with poverty. Poverty is often found in remote rural areas, where there are adverse resources, inhospitable climate conditions and terrain or low levels of physical infrastructure. There are often country-specific regional dimensions to poverty. I find an overwhelming increase in the significance of regional level effects on poverty.

In 2000, households in five out of 11 provinces registered a statistically significant likelihood of being poor compared with households in the capital city, Kigali (the omitted variable). By 2005, households in every single province showed a large positive and statistically significant likelihood of being poor compared with Kigali (ranging from 100% more likely in Umutara to over 900% in Gikongoro). Much of these increases were driven by rising urban poverty by province.

Given Rwanda's history of uneven regional development, finding regional disparities is not surprising. What is interesting is the dramatic change between 2000 and 2005, the period during which Rwanda moved away from post-conflict humanitarian development activities to long-term development planning.

Overseas development assistance may have been another important contributor to regional disparities. From 2000 a large proportion of overseas development assistance was channelled into rural extension activities in Umutara, Kibungo and Kigali-Nagali provinces. In all three of these provinces, poverty rates declined between 2000 and 2005. The most substantial decline was seen in Kigali-Ngali, where poverty rates fell from 69.4% to 46.5%. The other nine provinces not receiving as much overseas development assistance had mixed results in terms of poverty reduction (discussed later).

2.4 Poverty decomposition

The household poverty estimation is useful in identifying a number of important trends in the correlates of poverty. To understand what is driving changes in poverty by particular household characteristics, I decompose the FGT poverty measures and examine the contribution of various subgroups to the poverty in Rwanda.

The first decomposition of interest is by the gender and age structure of the household. I divide the sample into four reference categories for this purpose: male-headed households, female-headed widow households, female-headed non-widow households, and child-headed households (age less than 21 for household head). The results are presented in .

Table 4: Poverty decomposition by gender and age of household head

The two most striking results are that poverty rates of female-headed widows have fallen the most rapidly, while female-headed non-widows have seen a rise in poverty between 2000 and 2005. The decline in poverty for widow household heads is indeed encouraging and points to the effectiveness of government and non-governmental organisations targeting, including the genocide survivors fund. It may also indicate greater access to agricultural income-generating assets, with important changes in widow's rights to land with the passage of the Land Law during this time. There has also been a decline in the population share of widow heads of household (from 22% to 18.7%), which may explain a portion of the decline in poverty rates if more destitute widows remarried.

The second decomposition of interest is by the urban or rural residence of the household. As the regression analysis of poverty indicated, the correlates of poverty vary by urban or rural residence. presents the poverty decomposition by residence. Between 2000 and 2005 there was almost a 60% rise in the percentage of the population residing in urban areas, although the level relative to other sub-Saharan African countries remains low (Uchida & Nelson, Citation2008). Rising poverty accompanied the growth in population residing in urban areas. This rise was not small; urban poverty more than doubled between 2000 and 2005. Urban areas also saw an increase in the depth of poverty between years, as shown by the increase in the poverty gap ratio from 18 118.1 in 2000 to 22 856.3 in 2005. Nevertheless, the urban population rate remained considerably below the rural poverty rates.

Table 5: Poverty decomposition by urban/rural residence

The last decomposition exercise is by province. The dramatic increase in poverty of households based primarily on whether they resided within a particular province is a particularly troubling trend. presents the poverty decomposition by province. Here we see poverty increased in five out of 12 provinces. Of these, three provinces (Kigali urban, Gisenyi and Gikongoro) also saw the poverty gap index rise, indicating rising intensity of poverty.

Table 6: Poverty decomposition by province

The highest rates of poverty were in Gikongoro, where poverty rose from 76% to 79%. The mean poverty gap was also highest in Gikongoro. It is not surprising to find that extreme poverty rates in Gikongoro also increased, from 57% in 2000 to 63% in 2005. Gikongoro is one of the most food-insecure regions with high population density and low soil fertility (soils are acidic and steep topography has led to erosion) (MINECOFIN, Citation2002a).

examines the correlates of poverty growth by province more closely. Three variables are strongly correlated with headcount poverty growth. First, there is a strong negative correlation between initial poverty in 2000 and poverty growth between 2000 and 2005, suggesting a process of mean reversion. Secondly, areas with larger urban populations in 2000 saw greater poverty growth (lower poverty reduction). Finally, areas in which there were higher proportions of household heads in the agricultural sector in 2000 experienced higher growth rates of poverty (lower poverty reduction).

Figure 1: Correlates of poverty growth by province, 2000 to 2005

Figure 1: Correlates of poverty growth by province, 2000 to 2005

Understanding the sources of poverty is important in and of itself. However, in the context of the fragile peace and social tensions in Rwandan society, it is imperative that we move beyond understanding poverty and examine the dynamics of changing inequality.

3. Inequality in Rwanda, 2000 and 2005

Between 2000 and 2005 Rwanda experienced strong gross domestic product growth of 6.7%. Per-capita gross domestic product growth was less impressive at 3.4%. This growth was accompanied by a rather small reduction in national poverty rates. By contrast, virtually no attention has been given to what has happened to inequality, despite the risk that this could be among the sources of conflict in the future. The Rwandan genocide and reconstitution of society afterwards caused enormous upheaval and disruption in survival strategies for many households. An important and continued source of tension has been around land – for old and new returnees,Footnote4 among other groups. The National Unity and Reconciliation Commission reports that the majority of the population considers land disputes to be the greatest factor hindering peace (Musahara & Huggins, Citation2005; Pottier, Citation2006; Wyss, Citation2006). However, access to land is just one of the potential factors driving rising inequality. Access to social safety nets, extension services, finance, healthcare, education, and so forth, are also all potential sources for elite capture.

shows the Lorenz curves for consumption in 2000 and 2005. The Lorenz curves do not cross, implying that inequality increased across the distribution between 2000 and 2005 by whichever measure of inequality used (see Shorrocks [1982] for a formal examination of Lorenz curve domination). Given the civil war, genocide and massive displacement of the population over the 1990s, it is not surprising that inequality rose in that decade. What is more troubling, from a conflict prevention perspective, is the high level of consumption inequality and the trend of rising inequality between 2000 and 2005.

Figure 2: The Lorenz Curve, 2000 and 2005

Figure 2: The Lorenz Curve, 2000 and 2005

Consumption growth has not been broad based. shows consumption growth by decile between 2000 and 2005. Consumption growth of the top decile, at 21.1%, far exceeded that of any other decile over this period. Compared with the bottom decile, whose consumption grew at 6.7%, the top decile's consumption grew more than three times as fast. Clearly, absolute poverty reduction would have been faster if the observed income growth could have been combined with falling of inequality.

Figure 3: Consumption growth by decile

Figure 3: Consumption growth by decile

To investigate the correlates of household consumption, I use a standard Mincerian model specified as follows:

where ln Y i is the logarithm of per-capita household consumption (adjusted using equivalence scales and for regional price differences and given in 2000 prices) in Rwandan francs for household i and χij is a vector of exogenous explanatory variables that can be broadly grouped in regional, household-level and individual-level characteristics. Regional variables include urban/rural residence and province of household. Household-level characteristics include gender of household head, size of household, dependency ratio, household land cultivated, ownership of cattle, household received cash or in-kind transfers in, household sent cash or in-kind transfers to another household, household member with debt, and household member with savings. Individual-level characteristics for household head include age, education, and employment as farm labourer or waged worker.

presents descriptive statistics for the variables used in the expenditures regression analysis. As many variables were included in the poverty regression in Section 2, I will limit this discussion to variables only used for this exercise. Of particular interest are changes in correlates of consumption that may be affecting inequality. There was a slight increase in mean years of education from 3.0 years in 2000 to 3.3 years in 2005. While an improvement, this still indicates that the majority of the population had very low levels of formal education (less than four years of primary school education). Agricultural assets rose between the two years. On average, each household owned 0.53 cattle in 2000 and 0.67 cattle in 2005. Similarly, landholdings per household rose between the years. In 2000, the average land holding was 0.69 ha and by 2005 this had risen to 0.76 ha. Interestingly, the dependency ratioFootnote5 declined slightly between years (from 0.98 in 2000 to 0.90 in 2005), while at the same time household size slightly increased (from 4.96 in 2000 to 5.0 in 2005).

Table 7: Descriptive statistics

The data reveal rising levels of debt and savings. In 2000, 40% of households reported debt, which rose substantially in 2005 to 58%. At the same time, the proportion of households with savings doubled, from 21% to 44%.

3.1 Discussion of results: Household expenditure estimation

presents the results from the household expenditure estimation. Comparing female widow and non-widow households with the full sample, I find lower levels of consumption income, although this result is only statistically significant for widows in 2000 and for non-widows in 2005. The urban and rural samples show important differences for gender and household structure. In urban areas, female-headed households, both widows and non-widows, had a lower penalty in consumption in 2005 as compared with 2000. This result may indicate greater labour market opportunities for urban women compared with rural women whose livelihood strategies are primarily connected to ownership of land.

Table 8: Household expenditure estimation

Widows in rural areas faced a greater consumption penalty in 2005 while rural non-widows faced a lower consumption penalty in 2005 (both in comparison with 2000). This result is surprising because in Section 2 I found a decline in the likelihood of rural widow-headed households being poor. These results are not contradictory, however: they illustrate consumption penalties still exist for rural widow household heads, controlling for human, physical and geographical differences, but the penalty does not push them as frequently in 2005 into poverty.

The increase in magnitude of consumption penalty for rural widow heads may be explained by two factors. Firstly, the household consumption estimation explains about 30% of all variation in consumption, suggesting other important income-generating factors are not captured in the model. Secondly, if widow heads of household are employed in agricultural wage labour in a higher proportion, they will have been particularly hard hit by the decline in real wages during this period. The National Institute of Statistics of Rwanda reports that real wage rates for agricultural labourers fell in all regions by around 30% between 2000 and 2005 (Strode et al., Citation2007).

Returns to education were positively and significantly correlated with consumption and increased between 2000 and 2005. Substantially higher returns to education were present for the urban subsample, although both subsamples register increases between the years.

Land was positively and increasingly associated with higher levels of consumption income for rural households. This is not surprising given the agrarian structure of the economy and the negative relationship between land and poverty, as seen in Section 2. The increasing importance of land for levels of consumption in the rural sector indicates that any structural changes in the distribution of land will have significant implications on levels of consumption for the rural population. The importance in understanding changes in land ownership and concentration cannot be overstated. The majority of the Rwandan population relies on subsistence agriculture, and changes in ownership and distribution of agricultural assets threatens their survival.

Households in urban areas were statistically more likely to have higher levels of consumption in both 2000 and 2005. Regional dummies show no consistent or significant relationship in most regions in 2000. By 2005, however, regional differences are important in explaining overall variation in consumption. In 2005, all regions had significantly lower consumption than Kigali (the omitted variable).

There has been a recent resurgence of interest in different channels of financial resources, in particular remittances, within the development community (Adams & Page, Citation2003; Orozco, Citation2003; Johnson & Sedaca, Citation2004). In the expenditure estimation I include a number of variables to measure financial capital: whether the household receives or sends cash or in-kind transfers, whether a member of the household has a savings account, and whether a member of the household is in debt.

Whether a household has debt or not accounts for very little in the expenditure estimation. This finding is not entirely surprising as households with debt are likely to include both the very poor (those forced to borrow to try to avoid destitution) and the wealthy (those able to borrow to see higher returns in the future). In contrast, households with savings are strongly and statistically more likely to be associated with higher levels of consumption. Savings, as expected, act as a buffer against household income shocks. There has been a large increase in the magnitude of the effect of savings on consumption for urban households. This may be indicative of a rise in income-generating opportunities for urban households with access to financial capital.

Households receiving cash or in-kind transfers have lower consumption, although this variable is statistically significant only in 2000. Households sending cash or in-kind transfers to other households were much more likely to have higher levels of consumption in both years.

Examining the correlates of consumption is an important exercise in and of itself, and it gives us an idea of changes in patterns of correlates between the two surveys. Ideally, however, we would like to understand what is driving changes in inequality between the two years, as this will have important development ramifications for segments of the Rwandan population negatively impacted by these changes.

3.2 Decomposition of inequality

There is now a large literature on the decomposition of inequality (see Cowell, Citation2000; Morduch & Sicular, Citation2002). A seminal piece by Shorrocks (Citation1982) identified a methodology to decompose income inequality by the various sources of income for each individual. Over the last couple of decades there has been a resurgence of work in this area using regression-based methods for decomposing inequality (see Cowell & Jenkins, Citation1995; Morduch & Sicular, Citation2002). I use the Fields (Citation2003) methodology to decompose inequality, as is widely employed in similar exercises, to examine factor components of consumption and how they contribute to consumption inequality. presents the results for the decomposition of inequality exercise.

Table 9 Regression-based decomposition of inequality

I find little evidence that changes in consumption inequality were driven by the structure and composition of household members. The gender of the head of household explains a small and decreasing (for widows) proportion of inequality. The most significant decrease has been in the proportion of inequality explained by urban female widow-headed households (decreasing from 3.6% in 2000 to 0.6% in 2005). This supports the earlier finding that female widow-headed households (the largest proportion of female-headed households) have seen a larger than proportionate fall in poverty (although they still have higher poverty rates on average than their male counterparts).

Both the correlates and decomposition exercises found that being in a widow-headed household is becoming less important as a factor in determining a household's position in the consumption distribution, controlling for other possible determinants. My findings are encouraging in that consumption gains have been made for this vulnerable group between these two surveys. This finding does not mean that widows are no longer a vulnerable group compared with male-headed households. Besides widows experiencing lower levels of consumption, they are more likely in post-conflict settings to have lower physical and human assets (UNDP BCPR, 2002).

In examining levels of physical and human capital by gender of household head there are marked differences. Female-headed households had significantly lower average household land, cattle and education, compared with their male counterparts. For example, female-headed households in 2005 had on average 0.66 ha of land and male-headed households had 0.79 ha. In terms of education, male heads had 3.8 years of education in 2005 versus female heads with 2.1 years. For cattle, male heads in 2005 had on average 0.79 cattle compared with female heads of 0.38.

The dependency ratio explained an increasing proportion of consumption inequality, rising from 4.4% to 7.8%, with much greater rise in impact for rural households. This has occurred while the dependency ratio nationwide has declined, from 0.98 in 2000 to 0.90 in 2005. Over this time period, female-headed household experienced a decline in the dependency ratio (from 1.1 in 2000 to 0.82 in 2005). In contrast, male-headed households had a relatively stable dependency ratio (from 0.92 to 0.93 between years).

Education explained by far the largest proportion of inequality. Between 2000 and 2005, it accounted for a significant and slightly increasing proportion of explained inequality. This overall trend obscures important differences between urban and rural expenditure estimations, however. In urban areas there was a substantial decline in the amount of overall inequality that education explained, from 47% in 2000 to 39% in 2005. While education was still important in explaining urban inequality in 2005, there were a number of additional important factors, such as savings and region.

Land ownership explained an increasing proportion of overall inequality. For rural households, the contribution of land to overall explained inequality was substantial, from 4.7% in 2000 to 12.4% in 2005. This finding is particularly troubling in the context of Rwanda as there is a history of periods of social upheaval and violence corresponding to heightened tensions and conflict over land.Footnote6

The patterns of spatial inequality can be divided into two different aspects – a decreasing importance of rural–urban inequality, but an increasing importance of inequality between provinces. The proportion of inequality explained by urban residence versus rural residence was halved between 2000 and 2005. However, spatial inequality by region explains more. In 2000, province-level variation accounted for 9% of explained inequality, while by 2005 it had risen to more than 20%.

Whether or not a household member had savings explains a greater proportion of explained inequality in 2005 relative to 2000. The rise in significance of savings has been dramatic in urban areas. In the urban subsample, household savings rose from explaining 7% of explained inequality to 21%. Access to finance and to different sources of finance will probably be an increasingly important determinant of inequality in the years to come.

4. Conclusion

In any development context, but especially in post-conflict countries, it is important to understand the determinants of poverty and inequality. Despite the genocide and continued regional instability, Rwanda's economy has experienced positive growth since the late 1990s. This has occurred despite persistent structural problems in the Rwandan economy. In particular, the majority of the population remains dependent upon the agricultural sector where soil fertility, erosion, environmental degradation, plot fragmentation and variable rainfall limit the gains from investments in this sector.

This paper has looked at the changing correlates of poverty and inequality in Rwanda. There has been a small reduction in poverty, although this hides important differences based on household, education, assets and spatial characteristics. I found a substantial decline in poverty for widow-headed households, but an increase in poverty for female non-widows. Education of head of household acted as a buffer against poverty, although physical assets and geographical location overshadowed this variable. Poverty was strongly correlated with female non-widow-headed households, rural residence, size of household, lack of education and region of residence.

Taking the poverty analysis one step further, I decomposed the FGT measures of poverty by a number of important household and geographical characteristics. I found an important shift in the vulnerability of female-headed households. Widow heads were increasingly less likely to be poor, yet female non-widow heads were increasingly more likely to be poor. The decomposition of poverty by province revealed vast spatial differences in incidence, intensity and distribution of consumption income of the poor based on place of residence. The spatial patterns and intensity of poverty swamped educational returns.

Rising inequality is a pressing development concern in Rwanda, although its causes are poorly understood. In this paper I decomposed changing consumption inequality and discovered a number of important trends. There has been a decline in proportion of inequality explained by level of education, and a concomitant rise in inequality due to geographical location, land ownership and savings.

Both exercises reveal the increasing importance of physical assets versus human capital. In both cases, the distribution of assets was increasingly important in determining inter-household distribution of income. These results indicate that changes occurring in the agrarian structure and distribution of assets were dampening returns to education. They also point to the importance of examining the asset concentration occurring in rural areas, as this is an important component driving rising inequality and one that could potentially threaten long-term peace and stability in Rwanda.

Notes

2Ethnicity is not examined because the Government of Rwanda, since the 1994 genocide, made it illegal to collect information on ethnicity.

3An extensive literature on vulnerability exists (see Chaudhuri et al., Citation2002; Bourguignon et al., Citation2006; Kamanou & Morduch, Citation2005).

4Old returnees refer to individuals whose families fled Rwanda during periods of violence prior to 1990. A large proportion of these families escaped Rwanda from 1959 to 1963 during the instability and targeted attacks of Tutsi's with independence and the overthrow of the monarchy.

5The dependency ratio is calculated as number of children (age <18) divided by number of adults.

6Pottier (Citation2006) cites three additional sources of conflict around land since the genocide, due to female-headed households' need for land, return of new and old caseload refugees, and acquisition of land by the new political elite.

References

  • Adams, RH & Page, J, 2003, International Migration, Remittances, and Poverty in Developing Countries. The World Bank, Washington, DC.
  • Bourguignon, F, Goh, C & Kim, DI, 2006. Estimating individual vulnerability to poverty with pseudo-panel data. In Morgan, SL, Grusky, DB & Fields, GS (Eds), Studies in Social Inequality. Stanford University Press, Stanford, CA, pp. 349–69.
  • Chaudhuri, S, Jalan, J & Suryahadi, A, 2002. Assessing household vulnerability to poverty from cross-sectional data: A methodology and estimates from Indonesia. Discussion Papers 0102-52, Columbia University, Department of Economics.
  • Cowell, FA, 2000. Measurement of inequality. In Atkinson, AB & Bourguignon, F (Eds), Handbooks in Economics, vol. 16. Elsevier Science, North-Holland, Amsterdam, pp. 87–166.
  • Cowell, FA & Jenkins, SP, 1995. How much inequality can we explain? A methodology and an application to the United States. Economic Journal 105(429), 421–30. doi: 10.2307/2235501
  • Date-Bah, E, 2001. Jobs after War: A Critical Challenge in the Peace and Reconstruction Puzzle. International Labour Office (ILO), Geneva.
  • Deaton, A, 1997, The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Johns Hopkins University Press for the World Bank, Baltimore, MD.
  • Fields, GS, 2003. Accounting for income inequality and its change: A new method, with application to the distribution of earnings in the United States. Research in Labor Economics 22, 1–38.
  • Foster, J, Greer, J & Thorbecke, E, 1984. A class of decomposable poverty measures. Econometrica 52(3), 761–6.
  • Johnson, B & Sedaca, S, 2004. Diasporas, amigos and development: Economic linkages and programmatic responses. A Special Study of the USAID Trade Enhancement for the Services Sector (TESS) Project, USAID, Washington, DC.
  • Justino, P & Verwimp, P, 2006. Poverty Dynamics, Violent Conflict and Convergence in Rwanda. Households in Conflict Network, Institute of Development Studies at Sussex.
  • Kamanou, G & Morduch, J, 2005. Measuring vulnerability to poverty. In Dercon, S (Ed.), A Study Prepared by the World Institute for Development Economics Research of the United Nations University (UNU-WIDER). Oxford University Press, Oxford, pp. 155–75.
  • Koster, M, 2008. Linking Poverty and Household Headship in Post-genocide Rwanda. Households in Conflict Network (HiCN) Fourth Annual Workshop, Yale University, 5–6 December, New Haven, CT, USA.
  • Lindsey, C, 2001. Women facing War. International Committee of the Red Cross (ICRC), Geneva.
  • McKay, A & Greenwell, G, 2006, EICV-2005 Methodological Report for Poverty Line Construction. National Institute of Statistics Rwanda, Kigali, Rwanda.
  • MINECOFIN (Ministry of Finance and Economic Planning), 2000. Participatory Poverty Assessment 2000. Republic of Rwanda, Kigali.
  • MINECOFIN (Ministry of Finance and Economic Planning), 2002a. A Profile of Poverty in Rwanda: An Analysis Based on the Results of the Household Living Conditions Survey 1999–2001. Republic of Rwanda, Kigali.
  • MINECOFIN (National Poverty Reduction Programme Ministry of Finance and Economic Planning), 2002b. Government of Rwanda Poverty Reduction Strategy Paper. Republic of Rwanda, Kigali.
  • MINECOFIN (Ministry of Finance and Economic Planning), 2007, Economic Development and Poverty Reduction Strategy 2008–2012. Republic of Rwanda, Kigali.
  • Morduch, J & Sicular, T, 2002. Rethinking inequality decomposition, with evidence from rural China. The Economic Journal 112, 93–106.
  • Musahara, H & Huggins, C, 2005. Land reform, land reconstruction: A case study of Rwanda. In Huggins, C & Clover, J (Eds), From the Ground Up: Land Rights, Conflict and Peace in Sub-Saharan Africa. Institute of Security Studies, Pretoria and Cape Town.
  • Orozco, M, 2003, The future trends and patterns of remittances to Latin America. Inter-American Dialogue, Mexico.
  • Pottier, J, 2006. Land reform for peace? Rwanda's 2005 Land Law in context. Journal of Agrarian Change 6, 509–37. doi: 10.1111/j.1471-0366.2006.00133.x
  • Ravallion, M, 1999. Issues in measuring and modeling poverty. Policy Research Working Paper Series 1615, The World Bank, Washington, DC.
  • Sen, AK, 1976. Poverty: An ordinal approach to measurement. Econometrica 44(2), 219–31. doi: 10.2307/1912718
  • Shorrocks, AF, 1982. Inequality decomposition by factor components. Econometrica 50(1), 193–211. doi: 10.2307/1912537
  • Sorensen, B, 1998. Women and Post-conflict Reconstruction: Issues and Sources. United Nations Research Institute for Social Development (UNRISD), Geneva.
  • Strode, M, Wylde, E & Murangwa, Y, 2007, Labour Market and Economic Activity Trends in Rwanda: Analysis of the EICV2 Survey. Republic of Rwanda, Kigali.
  • Uchida, H & Nelson, A, 2008, Agglomeration Index: Towards a New Measure of Urban Concentration. World Bank, Washington, DC.
  • UNDP, 2013. Human Development Report 2013. The Rise of the South: Human Progress in a Diverse World. United Nations Development Programme, New York.
  • UNDP BCPR (Bureau for Crisis Prevention and Recovery), 2002. Gender Approaches in Conflict and Post-Conflict Situations. United Nations Development Programme, New York.
  • UNIFEM (The United Nations Development Fund for Women), 1998. Women's Land and Property Rights in Situations of Conflict and Reconstruction: A Reader. The United Nations Development Fund for Women (UNIFEM), Kigali, Rwanda.
  • World Bank, 2005. Introduction to Poverty Analysis. World Bank, Washington, DC.
  • World Bank, 2013. World Development Indicators. World Bank, Washington, DC.
  • Wyss, K, 2006. A Thousand Hills for 9 Million People. Land Reform in Rwanda: Restoration of Feudal Order or Genuine Transformation. Swiss Peace Foundation, Bern.

Reprints and Corporate Permissions

Please note: Selecting permissions does not provide access to the full text of the article, please see our help page How do I view content?

To request a reprint or corporate permissions for this article, please click on the relevant link below:

Academic Permissions

Please note: Selecting permissions does not provide access to the full text of the article, please see our help page How do I view content?

Obtain permissions instantly via Rightslink by clicking on the button below:

If you are unable to obtain permissions via Rightslink, please complete and submit this Permissions form. For more information, please visit our Permissions help page.