SEARCH

SEARCH BY CITATION

Keywords:

  • refugee;
  • rural poverty;
  • socio-economic position;
  • South African;
  • Mozambican

ABSTRACT

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

The assimilation of refugees into their host community economic structures is often problematic. The paper investigates the ability of refugees in rural South Africa to accumulate assets over time relative to their host community. Bayesian spatial–temporal modelling was employed to analyse a longitudinal database that indicated that the asset accumulation rate of former Mozambican refugee households was similar to their host community; however, they were unable to close the wealth gap. A series of geo-statistical wealth maps illustrate that there is a spatial element to the higher levels of absolute poverty in the former refugee villages. The primary reason for this is their physical location in drier conditions that are established further away from facilities and infrastructure. Neighbouring South African villages in close proximity, however, display lower levels of absolute poverty, suggesting that the spatial location of the refugees only partially explains their disadvantaged situation. In this regard, the results indicate that the wealth of former refugee households continues to be more compromised by higher mortality levels, poorer education, and less access to high-return employment opportunities. The long-term impact of low initial asset status appears to be perpetuated in this instance by difficulties in obtaining legal status in order to access state pensions, facilities, and opportunities. The usefulness of the results is that they can be used to sharpen the targeting of differentiated policy in a given geographical area for refugee communities in rural Africa. Copyright © 2011 John Wiley & Sons, Ltd.


INTRODUCTION

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

The assimilation of refugees into their host community economic structures is often problematic from a number of perspectives. Firstly, the traumatic nature of their transition is exacerbated because these groups arrive with few assets and experience difficulties with respect to their initial assimilation into the economic and social structures of the host country (Chambers, 1986; Kuhlman, 1990; O'Brien et al., 2009). Secondly, the difficulties of the transition are aggravated by high levels of poverty in rural Africa that have increased in recent decades as a result of globalisation, disease, climate change, and urban migration (Bryceson, 2004; Mertz et al., 2005; Rodgers, 2008; Sherbinin et al., 2008). Finally, the disadvantaged economic situation of refugee communities is perpetuated because of an inability to access institutions and invest in new opportunities (Binswanger 2004; Kydd et al., 2004; Sherbinin et al. 2008).

Mozambican refugees in rural South Africa arrived in the 1980s to 1990s with very few assets as a result of fleeing from a civil war (1975–1992). In addition, their situation was compromised by high levels of poverty in rural South Africa because of historical legacies that have fundamentally skewed the wealth and population demographics of the country (Bryceson, 2002a, 2002b; Kariuki, 2004; Machethe, 2004). In this regard, South Africa's highly skewed bi-modal economy has exacerbated the plight of the rural poor as a result of a focus on urban industrial development at the expense of the rural economy (Bryceson 2002a, 2002b). Furthermore, despite consistent GDP growth from 1994, income inequality in rural South Africa has deepened (Armstrong et al., 2008). As a result of these circumstances, this group – from here referred to as ‘former Mozambican refugees’ – experience higher levels of poverty and mortality nearly 20 years after their arrival (Rodgers, 2008; Collinson, 2010).

This paper investigates the ability of refugees in a rural South African community to improve their asset wealth [socio-economic position (SEP)] relative to their host community over a long tenure in the community. In particular, the paper investigates the dynamics of asset accumulation in a refugee community in order to target as well as make policy suggestions for their assimilation. The first research question tests whether the asset accumulation rates of former refugees and local households are different, as well as whether there is a persistent wealth gap (SEP) between the two sets of households. The second research question tests whether there is a spatial element that influences the household wealth of refugees. The third research question investigates what variables influence the asset wealth (SEP) of households in a rural community, as well as whether these variables show significant differences between former refugee and local households.

The importance of the study is underlined by the World Bank's increasing concern about rural poverty in Africa that has contributed to rapid urbanisation, overcrowding and a fierce competition for scarce resources. This problem is particularly problematic in South Africa (Bryceson, 2004; Machethe, 2004; Mertz et al. 2005). In particular, the study makes a contribution by extending the rural poverty debate to include a refugee context (Onyu et al., 2009; Schatz, 2009). The paper also makes a methodological contribution as a result of the application of a Bayesian multivariate model to track socio-economic phenomena such as poverty dynamics. In particular, this method ensures that the significance of predictor variables is not overstated because of spatial correlation (Elliot et al., 2000). Finally, the use of simulation-based Bayesian kriging maps makes a contribution by their ability to predict wealth (poverty) at village level in order to sharpen the focus of geographical targeting for policy makers (Gelfand et al., 1999).

The remainder of the paper is structured as follows: A Conceptual Framework outlines a conceptual framework that explains the dynamics of asset accumulation rates and their impact on SEP. Data and Method describes the data and method employed. Results and Discussion present and discuss the results, respectively. Finally, Conclusion and Recommendations concludes the study and makes some policy recommendations.

A CONCEPTUAL FRAMEWORK

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

This section develops a conceptual framework, illustrated in Figure 1, to explain the dynamics and direction of asset accumulation in a rural community. In this regard, the historical legacies of communities influence the initial asset status of households, as well as their characteristics. These factors, in turn, influence their choice of household strategies that largely determine income earning opportunities, asset accumulation rates and their outcome, namely, asset wealth, or SEP.

image

Figure 1. The dynamics of socio-economic position (SEP).

Download figure to PowerPoint

Household Asset Wealth (SEP)

Rural households are likely to experience different levels of asset wealth over time as the family producer/consumer ratio changes (Chayanov, 1986; Krishna, 2006). Household wealth or SEP incorporates some measures of physical and social resources, as well as household status in a local social hierarchy (Howe et al., 2008). In Figure 1, initial SEP status is explained as a long-term effect of inter-generational transfers that include natural, social, human, physical, and financial assets (Wu and Pretty 2004; Barrett, 2005). SEP is often measured by using a bundle of household assets (Booysen et al., 2008) and can be disaggregated into four possible categories of rural wealth status. Two of these categories include ‘always poor’ (poor then poor now) and ‘descending into poverty’ (not poor then, poor now). The other two groups are ‘never poor’ and ‘ascending out of poverty’ (poor then, not poor now). Changes in SEP can be calculated by estimating the aggregated difference between the number of households that ascend out of poverty and those that descend into poverty during a given period (Sen, 2003; Krishna, 2006).

Historical Legacies and Household Characteristics

Historical legacies have fundamentally influenced many African countries that have been subjected to decades of colonial rule, inept governments, corruption, succession wars and agro-climatic change (Bryceson, 2004; Mertz et al., 2005; Rodgers, 2008; Sherbinin et al., 2008). In South Africa, for instance, the 1913 Land Act largely determined where Black citizens could live for the next nine decades. Furthermore, historical economic policies have influenced the development of a dual economy that is largely urban based. These legacies, illustrated in Figure 1, have had a pervasive long-term effect on the initial SEP of rural households, institutions, and infrastructure (Williamson, 2000; Barrett, 2005). In this regard, initial asset status (SEP) is a critical factor that perpetuates a household's options (Barrett, 2005), as well as impacts on the household SEP growth rate and strategy choices, household size, education, social status, and employment opportunities (Schwarze and Zeller, 2005; Anriquez and Valdes, 2006; Vermeulen et al., 2008; Xing et al., 2008).

Household characteristics, strategy, and SEP

The dynamics of asset accumulation rates are primarily influenced by existing household assets and characteristics that are mediated by the presence of institutions, shocks, and opportunities. A household's size, health, education levels, and social networks, illustrated in Figure 1, largely determine its income generating strategies. These strategies, in turn, influence asset accumulation rates that translate into the SEP of the household. SEP, however, also has a reverse (cyclical) influence (see direction) on household characteristics in that wealth created is often re-directed at improving social status, education, or health (Sen, 2003; Krishna, 2006).

A wide range of household strategies influence SEP in rural Africa and the opportunity led strategies of wealthier rural households are very different from the survival led strategies of the poor (Barrett, 2005). Wealthier households (never poor), for example, are better able to adopt high-return strategies (pull led) because they have the necessary assets, ability, and social networks to access opportunities. Wealthier households (high SEP), moreover, are able to fund investment opportunities such as new technology or diversification that further accelerates asset accumulation and raises SEP. Improved SEP, moreover, allows a household to invest in healthcare needs and education, to elevate social status, and facilitates the future adoption of high-return strategies (Barrett et al., 2001; Barrett, 2005; Ardington et al., 2009). Conversely, poorer households (always poor) are more likely to have higher numbers of dependents, a poor education, and limited social status and networks. These households are, in effect, forced into adopting complimentary type low-risk, low-return strategies in order to survive because they lack both the assets and the ability to fund high-return options. They also have limited funds to invest in healthcare and education, and they are less able to fund shocks. These households, in effect, remain locked into low-income opportunities because they are unable to fund investments that would improve their strategic options (Sen, 2003; Elmquist and Olsson, 2006; Krishna, 2006; IIYama et al., 2008; Lay et al., 2008; Mendola, 2008; Wouterse and Pieterse, 2008). Because of a combination of a lower initial asset status (SEP) and less lucrative strategy options, poorer rural households are unable to close the wealth gap (SEP) with richer households (Barrett, 2005). Poorer households, moreover are also more vulnerable to natural disasters, illness, and death than richer households (Schatz and Ogunmefun, 2007; Goudge et al., 2009a: 2009b; Schatz et al., 2011), as well as less able to sustain economically inactive dependents (Urassa et al., 2001; IIYama et al., 2008; Lay et al., 2008; Wouterse and Pieterse, 2008).

DATA AND METHOD

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

This section briefly describes the location and type of data used to test the research questions before outlining how the data were analysed.

The Data

The Agincourt Health and Socio-Demographic Surveillance Site (HDSS) constitutes a rural sub-district of the Bushbuckridge municipal area that is located in Mpumalanga Province in the northeast of South Africa. A national park, forming the border with Mozambique, is situated on the eastern boundary of the site. Furthermore, the site occupies 400 km2 that includes 21 villages with a combined population of 70,000 inhabitants living in 12,167 households. A third of this population are of Mozambican origin, many of them arriving as refugees in the early to mid-1980s during the civil war (Schatz, 2009). Although we have differentiated South African households from their former refugee (Mozambican) counterparts, all Shangaan-speaking people in South Africa can be traced back to the colonial or local succession wars in Mozambique dating back to the period 1830–1890. This initial influx of refugees assimilated themselves into predominantly Sotho-speaking South African communities. In 1970, the Shangaan leadership assumed formal control of the area when a separate Bantustan called Gazankulu was created. It is, therefore, somewhat ironic that households in the area are now classified as South African or former Mozambican (Polzer, 2004). In fact, the latter category relates to more recent refugees that fled from Mozambique as a result of a civil war (1975–1992). For the purposes of this paper, we use the term former (Mozambican) refugee for those inhabitants who arrived in the mid-1980s as a result of the civil war in Mozambique. The original inhabitants of this area are referred to as South African.

A household level (unit of analysis) panel data structure was used for the years 2001, 2003, 2005, and 2007. The data recorded the establishment of new households established after 2001, as well as recorded dissolutions between 2001 and 2007. In order to track households identified as former Mozambican refugees, we established this status on the registration of a new household for each household member. Furthermore, the description of former refugee was retained for the full period of the study, regardless of changes in their legal status, which could include the acquisition of a permanent residence document or South African citizenship (Landau, 2005). To ensure that we did not overestimate significance because of households that dissolved and reformed within the site under another identifier, we checked for households with common individual identifiers, established linkages if individuals were shared and clustered upon this in the analysis. The data used were largely categorical and were gathered as a result of censuses that included a detailed survey of the demographics of the households. In this regard, a GIS system exists for all households within the site that is updated each year. An asset survey was also conducted every 2 years between 2001 and 2007. The assets recorded in the surveys included a range of appliances, telecommunications facilities, transport, and toilet facilities. Further assets included water supply type, the main source of cooking power and lighting, and the size and structure of the family swelling. Finally, we also included the number of livestock in response to the recommendations of other researchers (Jabbar et al., 2002; Berzborn, 2007). The predictor variables influencing SEP were selected on the basis of the theoretical relationships described in the conceptual framework (Fig. 1). These included household head demographics (age, gender, nationality, mortality due to HIV/TB or other causes), household mortality, duration of household existence at first asset observation, dependency ratio (<18:≥18 years old), migration patterns of household occupants, labour types, education, ownership, or access to farming land. It should be noted that some of these predictors are not tracked annually at each census update and hence could not be included in the final multivariate model because of missing observations in certain of the panel years.

The Methods

The paper incorporates a combination of static and dynamic analysis. These methods include descriptive analysis, a combination of classical and Bayesian univariate, and multivariate models and simulation-based Bayesian kriging.

The First Research Question

The first research question, namely, whether the asset accumulation rate of former refugee and local households are different, as well as whether there is a persistent wealth gap between the two sets of households, was tested as follows. Firstly, a table of assets was constructed for both former refugee and South African households for the four census years (2001, 2003, 2005, 2007). The rate of asset accumulation was then tested by developing line plots on the basis of regression models for both South African and Mozambican households using the year as a predictor. In order to test if there was a significant difference in the mean wealth (assets) of the two categories of household, we ran a series of two sample t-tests for each year.

The Second Research Question

The second research question tests whether there is a spatial element with respect to household wealth. In order to track the location, extent, and persistence of household poverty in the Agincourt HDSS site, we produced a series of spatial maps for the period 2001 to 2007. A multiple correspondence analysis (MCA) index was developed as a proxy for SEP (Howe et al., 2008) and simulation-based Bayesian kriging (Gelfand et al., 1999) was used to produce maps for each of the four panel years of the study period. The smoothed maps predicted the MCA asset score at numerous prediction points (regular grid) within the site for each of the panel years ( WinBUGS software). This approach allowed us to predict the asset count at un-sampled locations throughout the site and thus generate the smoothed maps. A baseline linear model (see Appendix 2) was used, which included no covariates except a constant and spatial (village level) random effect modelled as a parametric function of distance between pairs of village centroids (Diggle et al., 1998). Village centroids were superimposed on each map. Lighter areas represent higher levels of poverty (or lower SEP) whereas, conversely, the darker areas represent higher levels of SEP. In order to analyse areas of persistent poverty, as well as better interpret the smoothed maps, we calculated the percentage of households below the absolute poverty by using a set of baseline assets (Booysen et al., 2008). These assets included ownership of a radio and bicycle, a cement floor in the house, and access to public water and a pit latrine. This information was then used to compare poverty over the full period in South African versus former Mozambican refugee households, as well at each of the 21 villages in the site.

The Third Research Question

The third research question investigates what variables are associated with wealth (SEP), as well as whether these variables show significant differences between former refugee and local households. This question was tested by a range of univariate and multivariate classical and Bayesian ordinal regression models. A classical model (with no random effects) was run for comparative purposes.

We chose a Bayesian methodology to predict SEP in order to address a number of spatial and temporal problems. In this regard, objects in close proximity are often more alike. Thus, common exposures (measured or unmeasured) may influence SEP similarly in households of the same geographical area, introducing spatial correlation in the outcome. Repeated data are also expected to be correlated in time. Standard statistical methods, however, assume independence of outcome measures, thus ignoring correlation bias for two reasons. Firstly, the standard error of the covariates is underestimated, thereby overestimating the significance of the risk factors. Secondly, estimates of SEP outcomes would be incorrect at locations where data are lacking. Geo-statistical Bayesian models relax the assumption of independence and assume that spatial correlation is a function of distance between locations. They are highly parameterised models, and their full estimation has only become possible in the last decade by formulating them within a Bayesian framework (Hedeker and Gibbons, 1994) and estimating the parameters via Markov chain Monte Carlo (MCMC) simulation.

Construction of the dependent variable (SEP indices)

A set of assets, employed by the World Bank, was used to develop a wealth (SEP) index for each household (Gwatkin et al., 2000; Howe et al., 2008). These assets included a balance of purchased household assets, housing quality, water and sanitation. Livestock ownership was also included in our set of assets because of their importance in rural African communities. The use of household assets (wealth indices) to determine SEP has been adopted by a number of other studies (Gwatkin et al., 2000; Sahn and Stifel, 2003). In this regard, household asset status is a reliable proxy for long-term consumption expenditure (Booysen et al., 2008). Furthermore, measures of income and consumption expenditure are difficult and costly to obtain in a rural setting, and measures of income can vary widely from season to season (Howe et al., 2008). SEP indices were constructed using an absolute asset count, principal component analysis (PCA) and MCA. The weights used for both the PCA and MCA indices were those from the first dimension. We chose to use an MCA-based index (see APPENDIX 1) after a detailed evaluation of the properties of the three measures, as well as because it is better suited to categorical data (Howe et al., 2008).

Explanatory variables

The explanatory variables used in the univariate and multivariate analysis were based on well tested issues in the literature (Sen, 2003; Malmberg and Tegenu, 2007; Sherbinin et al., 2008; Collinson, 2010). These variables included year, household head demographics (age at panel year, gender, nationality, death), household factors (duration or years existed at asset observation, number of individuals in household 18+ or <18, hospital admissions in a given panel year, the migrant proportion, and the number of working individuals in the preceding year). Household head status was classified as former refugee or South African, and a variable was developed to track the proportion of the household that were former refugees. Other predictors included the death of household heads and other household members in both the preceding and current years, education years of household occupants and household child grants received, ownership or access to and usage of farming land, and selected employment opportunities in the professional and government sectors. The definition of migration discriminates between permanent and temporary to reflect a high dependency on temporary labour migration in this population (Collinson et al., 2007). We also use temporary migration in the household migrant ratio because this has been shown to have significant positive linkages with SEP (Collinson et al., 2006).

Univariate and multivariate analysis

A preliminary analysis of the univariate relationships was carried out using Stata (STATACORP LP, College Station, Texas). In this regard, ordinal regression techniques (clustered on household level) were used to assess the relationship between the relevant ordinal wealth category and each covariate. Significant covariates (10% level) emerging from the univariate analysis were then selected as inputs for the multivariate models. These models included two non-spatial models. The first was a classical non-spatial model ( Stata). The parallel regression assumption was tested using the Brant test. The second was a non-spatial [ WinBUGS (MRC Biostatistics Unit, Imperial College of Medicine, Cambridge, UK)] using a Bayesian modelling approach with the third model being a spatial (or geo-statistical) model also incorporating an unstructured household level random effect. This multilevel random effects ordinal model (Hedeker and Gibbons, 1994) incorporated a Bayesian geo-statistical approach in order to account for spatial correlation in the panel data set whereas the unstructured household level random effect was included to adjust for repeated measurements on a given household. Spatial correlation was measured at village level and a household level random effect for repeated longitudinal measurements on households. Further details of the Bayesian statistical model can be seen in Appendix 2.

RESULTS

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

The results investigate the three principal research questions. These included whether the asset accumulation rare (SEP) of refugee households was different from local households, investigating the variables that are associated with SEP and determining whether differential levels of asset accumulation (poverty) can be mapped over time and space in a rural community.

Is the Asset Accumulation Rate of Former Refugee and South African Households Different, and Is There a Persistent Wealth Gap?

The results indicate that the average number of assets owned, illustrated in Table 1, increased for both South African and former refugee households over the period 2001 to 2007. The increase in overall asset status is supported by a considerable increase in high cost items such as fridges, stoves, TV sets, video machines, cellular phones, and cars. Conversely, the number of radios, fixed line telephones, bicycles, carts, and livestock has declined. The results also indicate an increased access to electricity for lighting and cooking and to pit toilets and a marked increase in the number of plans to extend the main dwelling. Contrary to the general increase in assets, regular water supply appears to have decreased in place of irregular water supply. The general increase in assets is supported by other studies in the province (DSP, 2006), and the positive influence of an electrification program is also commonly linked to an increase in assets (Kanagawa and Nakata, 2008).

Table 1. Asset status-number (percent) of households in possession of item.
FactorSouth AfricanMozambican
2001200320052007Overall2001200320052007Overall
  • 1

    Significant increasing trend (assets) for both South African, as well as for former refugee households (β = 0.03, p < 0.001) between 2001 and 2007.

Number of households7,2437,7627,7867,65530,4462,9173,2083,1513,24112,517
Ave number of assets (std. dev.)17.73 (2.83)7.71 (2.72)8.35 (2.50)9.10 (2.45)8.23 (2.69)5.83 (2.41)5.71 (2.42)6.47 (2.44)7.27 (2.66)6.33 (2.57)
Asset totals (%)          
Fridge3,526 (48.7)4,272 (55)5,157 (66.2)5,746 (75.1)18,759 (61.6)710 (24.3)786 (24.5)1,241 (39.4)1,707 (52.7)4,458 (35.6)
Stove3,577 (49.4)3,991 (51.4)4,878 (62.7)5,750 (75.1)18,196 (59.8)503 (17.2)509 (15.9)894 (28.4)1,463 (45.1)3,369 (26.9)
TV4,236 (58.5)4,778 (61.6)5,083 (65.3)5,388 (70.4)19,485 (64)1,305 (44.7)1,313 (40.9)1,526 (48.4)1,652 (51)5,796 (46.3)
Video machine544 (7.5)718 (9.3)1,149 (14.8)3,269 (42.7)5,680 (18.7)69 (2.4)89 (2.8)170 (5.4)1,010 (31.2)1,338 (10.7)
Satellite dish23 (0.3)26 (0.3)48 (0.6)157 (2.1)254 (0.8)3 (0.1)3 (0.1)10 (0.3)22 (0.7)38 (0.3)
Radio3,048 (42.1)1,878 (24.2)1,878 (24.1)1,885 (24.6)8,689 (28.5)1,148 (39.4)921 (28.7)795 (25.2)639 (19.7)3,503 (28)
Fixed phone325 (4.5)216 (2.8)166 (2.1)212 (2.8)919 (3)19 (0.7)12 (0.4)18 (0.6)41 (1.3)90 (0.7)
Cellular phone2,839 (39.2)4,348 (56)6,082 (78.1)6,646 (86.8)19,915 (65.4)944 (32.4)1,376 (42.9)2,209 (70.1)2,627 (81.1)7,156 (57.2)
Car1,177 (16.3)1,184 (15.3)1,316 (16.9)1,356 (17.7)5,033 (16.5)298 (10.2)306 (9.5)327 (10.4)344 (10.6)1,275 (10.2)
Motorbike57 (0.8)29 (0.4)27 (0.3)58 (0.8)171 (0.6)9 (0.3)6 (0.2)14 (0.4)14 (0.4)43 (0.3)
Bicycle942 (13)789 (10.2)803 (10.3)692 (9)3,226 (10.6)430 (14.7)363 (11.3)328 (10.4)237 (7.3)1,358 (10.8)
Cart329 (4.5)257 (3.3)187 (2.4)193 (2.5)966 (3.2)15 (0.5)15 (0.5)27 (0.9)20 (0.6)77 (0.6)
Ave number of rooms (std. dev.)2.85 (1.58)2.72 (1.29)2.83 (1.33)2.86 (1.35)2.81 (1.39)2.52 (1.48)2.48 (1.38)2.59 (1.47)2.5 (1.42)2.54 (1.44)
Power for lighting (%)          
Electricity5,835 (80.6)6,849 (88.2)7,437 (95.5)7,357 (96.1)27,478 (90.3)1,347 (46.2)1,485 (46.3)2,326 (73.8)2,485 (76.7)7,643 (61.1)
Battery/generator4 (0.1)10 (0.1)10 (0.1)8 (0.1)32 (0.1)4 (0.1)1 (0)6 (0.2)3 (0.1)14 (0.1)
Solar power5 (0.1)5 (0.1)1 (0)58 (0.8)69 (0.2)2 (0.1)4 (0.1)2 (0.1)50 (1.5)58 (0.5)
Paraffin387 (5.3)203 (2.6)69 (0.9)47 (0.6)706 (2.3)632 (21.7)535 (16.7)211 (6.7)138 (4.3)1,516 (12.1)
Candles1,011 (14)689 (8.9)266 (3.4)185 (2.4)2,151 (7.1)928 (31.8)1,179 (36.8)602 (19.1)562 (17.3)3,271 (26.1)
Other1 (0)6 (0.1)3 (0)0 (0)10 (0)4 (0.1)4 (0.1)4 (0.1)3 (0.1)15 (0.1)
Power for cooking (%)          
Electricity1,383 (19.1)1,899 (24.5)2,045 (26.3)3,039 (39.7)8,366 (27.5)111 (3.8)130 (4.1)195 (6.2)476 (14.7)912 (7.3)
Gas bottle225 (3.1)181 (2.3)215 (2.8)141 (1.8)762 (2.5)21 (0.7)22 (0.7)26 (0.8)22 (0.7)91 (0.7)
Paraffin508 (7)446 (5.7)350 (4.5)137 (1.8)1,441 (4.7)79 (2.7)82 (2.6)108 (3.4)40 (1.2)309 (2.5)
Wood5,088 (70.2)5,219 (67.2)5,170 (66.4)4,329 (56.6)19,806 (65.1)2,692 (92.3)2,968 (92.5)2,817 (89.4)2,697 (83.2)11,174 (89.3)
Other39 (0.5)17 (0.2)6 (0.1)9 (0.1)71 (0.2)14 (0.5)6 (0.2)5 (0.2)6 (0.2)31 (0.2)
Toilet facility type (%)          
Modern21 (0.3)17 (0.2)27 (0.3)26 (0.3)91 (0.3)1 (0)0 (0)1 (0)2 (0.1)4 (0)
VIP76 (1)39 (0.5)193 (2.5)284 (3.7)592 (1.9)14 (0.5)15 (0.5)55 (1.7)100 (3.1)184 (1.5)
Pit toilet5,044 (69.6)5,404 (69.6)5,689 (73.1)5,862 (76.6)21,999 (72.3)1,128 (38.7)1,298 (40.5)1,502 (47.7)1,777 (54.8)5,705 (45.6)
None2,095 (28.9)2,300 (29.6)1,874 (24.1)1,470 (19.2)7,739 (25.4)1,773 (60.8)1,894 (59)1,593 (50.6)1,354 (41.8)6,614 (52.8)
Unknown7 (0.1)2 (0)3 (0)13 (0.2)25 (0.1)1 (0)1 (0)0 (0)8 (0.2)10 (0.1)
Water availability (%)          
Always796 (11)978 (12.6)873 (11.2)747 (9.8)3,394 (11.1)237 (8.1)204 (6.4)674 (21.4)436 (13.5)1,551 (12.4)
Most of the time2,833 (39.1)1,854 (23.9)2,160 (27.7)2,493 (32.6)9,340 (30.7)1,550 (53.1)535 (16.7)1,247 (39.6)985 (30.4)4,317 (34.5)
Few hours a day347 (4.8)295 (3.8)587 (7.5)561 (7.3)1,790 (5.9)95 (3.3)69 (2.2)197 (6.3)268 (8.3)629 (5)
Irregular3,258 (45)4,634 (59.7)4,165 (53.5)3,847 (50.3)15,904 (52.2)1,035 (35.5)2,400 (74.8)1,030 (32.7)1,551 (47.9)6,016 (48.1)
Unknown9 (0.1)1 (0)1 (0)7 (0.1)18 (0.1)0 (0)0 (0)3 (0.1)1 (0)4 (0)
Water supply (%)          
Tap in house72 (1)48 (0.6)60 (0.8)41 (0.5)221 (0.7)10 (0.3)10 (0.3)7 (0.2)16 (0.5)43 (0.3)
Tap in yard1,574 (21.7)866 (11.2)1,512 (19.4)2,093 (27.3)6,045 (19.9)181 (6.2)110 (3.4)181 (5.7)254 (7.8)726 (5.8)
Tap in street4,661 (64.4)5,524 (71.2)5,982 (76.8)5,379 (70.3)21,546 (70.8)2,402 (82.3)2,898 (90.3)2,888 (91.7)2,955 (91.2)11,143 (89)
Other936 (12.9)1,324 (17.1)232 (3)142 (1.9)2,634 (8.7)324 (11.1)190 (5.9)75 (2.4)16 (0.5)605 (4.8)
Livestock (%)          
Cattle1,503 (20.8)1,331 (17.1)1,307 (16.8)1,254 (16.4)5,395 (17.7)80 (2.7)89 (2.8)87 (2.8)81 (2.5)337 (2.7)
Poultry4,552 (62.8)4,307 (55.5)3,268 (42)3,035 (39.6)15,162 (49.8)1,965 (67.4)1,826 (56.9)1,203 (38.2)1,314 (40.5)6,308 (50.4)
Pigs409 (5.6)284 (3.7)270 (3.5)240 (3.1)1,203 (4)30 (1)14 (0.4)20 (0.6)12 (0.4)76 (0.6)
Goats973 (13.4)789 (10.2)770 (9.9)710 (9.3)3,242 (10.6)346 (11.9)347 (10.8)340 (10.8)334 (10.3)1,367 (10.9)
Main dwelling still under construction (%)1,253 (17.3)1,519 (19.6)1,613 (20.7)1,419 (18.5)5,804 (19.1)332 (11.4)427 (13.3)685 (21.7)546 (16.8)1,990 (15.9)
Plans to extend main dwelling (%)217 (3)135 (1.7)113 (1.5)395 (5.2)860 (2.8)84 (2.9)27 (0.8)40 (1.3)287 (8.9)438 (3.5)

The results in Table 1 indicate that the average number of assets owned increased by 17.7% for South African households and by 24.7% for former refugee households. In order to assess the asset accumulation rates of the two groups, we developed simple bivariate regression models for South African and former refugee households by using time as a predictor (census year). SEP or asset wealth, moreover, was calculated using an MCA index (Booysen et al., 2008) in order to develop a more accurate measure of wealth than the number of assets illustrated in Table 1. A temporal line curve, illustrated in Figure 2, indicates a highly significant increasing relationship between time and SEP for both (pooled) South African and Mozambican households (p < 0.001). The two models, moreover, had almost identical coefficients (exp(β) = 1.20) for the predictor (time), indicating a similar asset accumulation rate for both sets of households from 2001 to 2007. A consistent asset gap is, therefore, maintained for the full period.

image

Figure 2. Comparative model of predicted asset score (MCA weighted) by household head nationality and panel year.

Download figure to PowerPoint

A series of two sample t-tests for each of the four census years confirms a highly significant (p < 0.001) difference between the wealth (SEP) levels of the two sets of households with mean difference in absolute asset count of approximately 1.8. Finally, the asset gap (asset count-not MCA) between South African and former refugee households appears to widen in 2003 (Table 1), but restabilises in 2005 through to 2007, and we conclude a persistent constant gap (Fig. 2) on the basis of the weighted asset index (MCA).

Is There a Spatial Element with Respect to Household Wealth?

The four spatial–temporal maps, illustrated in Figure 3, show the spatial proximity of 21 villages, as well as indicate the level of assets owned (MCA index) for each of the four census years. The maps indicate a general increase in asset ownership (maps darken) for the period except for 2003, which indicates a decrease in asset wealth in some of the villages in the south-east. In particular, the maps indicate a cluster of five poorer former Mozambican villages (12, 17–20) in 2001, which reduce to two in 2007 (18, 19), which have markedly lower levels of SEP than the other 19 villages. These villages all had a high percentage (>75%) of former Mozambican refugee households and are mostly (except for village 18) clustered in the south-east sector of the site. Villages 17 and 19, moreover, are situated in close proximity to the border of a national park that runs approximately south-east through the site. Conversely, the South African villages in the south-east sector of the site (5, 8, 11, 15) had lower absolute poverty levels than the neighbouring refugee villages.

image

Figure 3. The mean predicted or kriged MCA SEP score (darker colour implies increasing wealth): 2001–2007.

Download figure to PowerPoint

In order to demonstrate the differential wealth of former refugees, Table 2 shows that former refugee households retained twice the level of absolute poverty compared with South African households.

Table 2. Households (HH) below the absolute poverty line by nationality and village.
FactorCategory2001200320052007
MozambicanPoor HH1,4811,7041,3431,260
Total HH3,0413,2403,2083,520
% HHs48.70%52.59%41.86%35.80%
95% CI46.92–50.48%50.87–54.31%40.16–43.57%34.21–37.38%
South AfricanPoor HH2,4702,2011,8071,404
Total HH7,6947,8517,9288,153
% HHs32.10%28.03%22.79%17.22%
95% CI31.06–33.15%27.04–29.03%21.87–23.72%16.40–18.04%

Further analysis of absolute poverty levels, illustrated in Table 3, show that the poorer villages indicated as former Mozambican in 2001 (villages 12, 17–20) all had in excess of 75% Mozambican households. These villages all had absolute poverty level percentages (columns 2001 to 2007) of more than 50% in 2001.

Table 3. Households below the absolute poverty line by village.
VillageNumber of households 1Mozambican (%) 12001 (%)2003 (%)2005 (%)2007 (%)
  • 1

    Based on estimates for 2001–2007.

11,34716.333.932.725.117.6
260520.035.829.221.121.7
31,17215.940.428.827.317.7
46848.938.325.726.417.8
562228.031.537.231.221.3
672325.330.732.422.923.6
743840.242.539.832.028.0
81,13847.728.831.031.625.3
993312.028.423.721.217.6
1089648.530.527.827.217.9
111,25331.521.725.915.515.0
1246377.157.260.732.025.1
1365810.933.830.323.019.4
144044.042.935.527.221.4
1560838.839.352.133.421.7
168575.538.435.730.225.7
1746897.665.068.741.837.7
1824295.088.680.082.981.4
1926296.655.266.074.176.4
2021881.756.063.350.435.7
217037.374.652.139.428.9

Villages 18 and 19 retained absolute poverty level percentages of 81% and 76%, respectively, in 2007. Interestingly, many of the poorer South African villages (villages 5–8) were also situated in close proximity to the border of the national park running south-east through the site. Village 21 (South African), however, has higher levels of absolute poverty because it was established as a low-cost housing site for disadvantaged South African citizens. In general, more infrastructure and facilities are located in the western sector of the site. It is interesting to note the villages with a very high Mozambican content (>50%) also indicated a lag factor that was maintained throughout the period 2001 to 2007. The results, therefore, demonstrate that there is a spatial element with respect to the distribution of SEP across the site and that, despite an overall improvement in SEP, households in former refugee villages are more likely to fall under the absolute poverty line than those in South African villages.

What Variables Influence Asset Wealth (SEP)?

The bivariate results, illustrated in Table 4, indicated a number of highly significant relationships (<0.001) between SEP and a range of predictor variables. In this regard, SEP was significantly positively influenced by older male household heads, as well as older households. Furthermore, SEP was positively associated with the number of working individuals, by the proportion of temporary migrants and especially by access to farming land. Households with higher levels of secondary and tertiary education, moreover, had a 4.8-fold higher chance of being wealthier. The biggest influence on wealth creation, however, was a situation when more than one household member was employed in the government, professional or skilled private sector that positively increased the odds of increased SEP by 10.38 fold.

Table 4. Bivariate ordinal regression analysis of variables influencing household SEP (asset wealth) using a household level yearly panel data structure.
FactorNOdds ratio95% CIp value
  • SEP, socio-economic position.

  • 1

    Similarly non-significant when comparing these numbers in preceding year with current year SEP.

  • 2

    Large percentage of missing data (due to nature of certain specialised modules) hence not included in final multivariate models.

Household head age (continuous)45,3461.0081.006–1.010<0.001
Household head 40 or more years of age31,085/45,3461.631.51–1.77<0.001
Male household head28,037/45,3441.471.38–1.55<0.001
Mozambican head12,654/45,3130.280.26–0.30<0.001
South African head32,6591.00  
Mozambican head (in migrated post-1993)620.720.34–1.510.382
Mozambican head (in migrated pre-1993)12,5920.300.28–0.32<0.001
% Mozambican content of household45,4010.270.25–0.29<0.001
Duration of existence of household at first asset status45,4111.141.13–1.15<0.001
Household existed for 2 or more years before first asset observation44,271/45,4112.502.14–2.90<0.001
Number of individuals in household 18+45,4111.281.25–1.29<0.001
Number of individuals in household <1845,4111.091.08–1.11<0.001
Number of working individuals in preceding year22,5891.471.43–1.51<0.001
Proportion of household that are migrants45,1281.521.34–1.72<0.001
Average 4+ migrant months per household individual per year11,334/45,1281.231.17–1.29<0.001
Household head death227/45,4110.580.48–0.71<0.001
Household head death in given year (pre-September)1113/45,4110.890.63–1.250.493
Household head status    
Alive45,1841.00  
Died (non-HIV/AIDS)2670.650.57–0.73<0.001
Died (HIV/AIDS)950.300.20–0.44<0.001
Household deaths in given year145,4111.100.96–1.270.163
Cumulative household deaths to given year145,4111.010.98–1.050.475
Number of household hospital admissions11,6201.000.89–1.130.968
Number of education years in household8,9481.011.00–1.010.022
Average secondary plus education level years per household individual520/8,9484.814.51–5.14<0.001
Ownership or access to and usage of farming land29,209/11,5962.292.10–2.49<0.001
Number of child grants received23,9010.840.76–0.92<0.001
Number of child grants received in previous year23,9010.900.83–0.980.017
Labour type43,821   
None involved in government, professional, or private sector skilled work23,8781.00  
At least one individual involved in only one of the above17,5042.442.29–2.59<0.001
Individuals involved in two or more of the above2,43910.389.26–11.62<0.001

Household SEP was negatively influenced if it was headed by a former refugee, and the larger the percentage of former refugee household members, the lower the SEP. Household deaths also had a major negative effect on SEP, especially if the household head died because of HIV/AIDS infection. SEP also appears to be negatively associated with the number of child grants received, indicating rather that poorer households are more likely to access child grants.

The multivariate results, illustrated in Table 5, incorporate three different models. All three models are reasonably consistent with respect to their prediction of the odds ratios for a majority of the predictors. The difference in the odds ratios between Models 1 and 2–3 is explained by the fact that the latter two models incorporate an unstructured household level random effect. The models confirmed a consistent significant positive relationship between SEP and time, which supports the contention of an overall increase in asset wealth (SEP) between 2001 and 2007. In this regard, SEP growth in 2003 was only marginal for South African households and negative for former refugee households.

Table 5. Multivariate multilevel ordinal (MCA SEP tertiles as outcome) regression analysis using a household level yearly panel data structure comparing different modelling approaches: classical, Bayesian unstructured household level random effects, and parametric distance spatial random effect.
FactorModel 1: Non-spatial model1 ( Stata – classical fixed effects model)Model 2: Non-spatial random effect model ( WinBUGS)Model 3: Spatial random effect model2 ( WinBUGS)
OR95% CIOR95% CIOR95% CI
  • MCA, multiple correspondence analysis; SEP, socio-economic position; AIC, Akaike's information criterion; DIC, deviance information criterion.

  • 1

    Brant test of parallel regression assumption: χ2: 19.6, p > χ2: 0.108, that is, sufficient evidence to suggest that the parallel regression assumption has not been violated

    .
  • 2

    Model also includes an unstructured household level random effect (pooled).

  • 2

    Significant at the 10% level.

Year1.131.12, 1.151.221.2, 1.251.231.20, 1.25
Relative SEP loss (or gain) of Mozambican-headed households versus South African-headed by panel year relative to 2001 baseline      
20011   
20030.780.72, 0.83
20051.010.93, 1.09
20071.030.94, 1.12
Household size of 3 or more2.842.6, 3.093.963.52, 4.434.083.64, 4.55
Household existed for at least 2 or more years before first asset observation1.591.35, 1.862.021.56, 2.581.721.31, 2.20
Household head age of 40 or more years1.291.21, 1.371.51.38, 1.641.471.35, 1.60
Male household head1.571.48, 1.672.121.91, 2.342.131.93, 2.35
Mozambican household head0.360.33, 0.380.170.15, 0.20.200.17, 0.23
Household head alive1 1   
Died (non-HIV/AIDS)0.810.71, 0.920.670.54, 0.820.650.53, 0.79
Died (HIV/AIDS)0.500.33, 0.760.320.13, 0.630.310.13, 0.63
Average 4+ migrant months per household individual per year1.010.96, 1.071.071.00, 1.1521.081.00, 1.15
Average of secondary or higher education level years per household individual2.572.39, 2.765.314.72, 5.994.544.05, 5.09
None involved in government, professional or private sector skilled work1 1   
At least one individual involved in only one of the above1.831.72, 1.953.022.71, 3.362.882.59, 3.18
Individuals involved in two or more of the above5.134.55, 5.7718.7015.26, 2316.2013.36, 19.00
Constant−3.1−3.37, −2.82−3.30−4.06, −2.00
inline image (household)5.124.87, 5.394.714.48, 4.96
inline image (spatial)1.220.62, 2.34
Range (m)2,37817,794, 269
AIC( Stata)/DIC( WinBUGS)79,282.361,918.961,670.7

An interactive effect or empirical estimation of relative gains (or losses) by household nationality and panel year (using 2001 as the reference) underlines the relative loss of asset wealth (SEP) of former refugee households compared with their host community in 2003 (see Model 1: OR = 0.78, p < 0.001). This confirms the previous observations with respect to the trend of assets owned (Table 1), as well as the results of the reported SEP changes in the 2003 spatial map (Fig. 3). Thereafter, there was no significant gain in the years 2005 and 2007 as shown by the odds ratios that were slightly above 1 (see Model 1).

All three models indicate that larger (more than three members), older households and households headed by a man, were wealthier (MCA asset score). In particular, all the models indicated that SEP was positively influenced by higher levels of secondary and tertiary education, as well as significantly positively influenced if household members were employed in the government, professional or private skilled sectors. The positive impact of these categories of employment on SEP was specifically marked (5.13- to 18.70-fold increase) if a household had more than one member employed in these high-return sectors. Conversely, the death of the household head, especially due to HIV/AIDS, has a profound negative impact on household SEP in all the models.

South Africans and former refugee comparisons

The results indicate no significant difference in household head death proportions between South African-headed and Mozambican-headed households (OR = 1.01; p = 0.934). However, Mozambican-headed households did appear to have a significantly higher overall household proportion of deaths, which supports the findings of Collinson (2010). The results also indicated that South Africans had a significantly higher number of secondary and tertiary education years (p < 0.001), as well as a significantly higher proportion of labour migrants than their Mozambican counterparts (p < 0.001) as confirmed in other studies (Collinson et al., 2007; Rodgers, 2008). Presumably, because of the reduced ability of former refugees to access labour migrant options, Mozambican households had a significantly higher proportion of male-headed households; however, they were younger (<40 years) than male-headed South African households (p < 0.001). Further analysis of household age (duration) indicates a significant difference (p < 0.001) between South African households who had an average duration of 13.15 years compared with 11.59 years for former refugees. Finally, the results indicate that South African households had a significantly higher proportion of individuals involved in government, professional or private sector skilled work when compared with Mozambican-headed households (p < 0.001).

DISCUSSION

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

This section further discusses the results with respect to the persistence of the wealth gap, the spatial nature of poverty and the dynamics of asset accumulation.

The Persistent Wealth Gap

Recent evidence in Southern Africa suggests that refugees can be exposed to multiple stressors that impact negatively on their ability to integrate into the economic activities of their host community (O'Brien et al., 2009). Our results show that there was a persistent (constant) wealth (SEP) gap between former refugee and South African households despite the fact that former refugee households had slightly higher asset accumulation rates in terms of the number of assets (Table 1). In this respect, the refugees arrived with very few assets (Dolan et al., 1997; Rodgers, 2008) and were unable to close the wealth gap because their asset accumulation rate was influenced by lower return strategies levied off a lower initial asset base. The long-term effect of a low initial asset status (Barrett, 2005), therefore, continues to affect their asset accumulation rate despite having arrived over 20 years ago (Schatz, 2009; Collinson, 2010). This inability to close the wealth gap, moreover, has been compounded by difficulties with respect to attaining legal status, as well as social discrimination with respect to labour opportunities (Landau, 2005). Finally, former Mozambican refugees were unable legally to access social grants until 2006 (Schatz, 2009) and were, thus, precluded from this important stabilising source of income that was available to South African residents (Booysen and Van Der Berg, 2005). The question remains, however, whether there was a spatial component with respect to the wealth gap.

The Temporal–Spatial Nature of Persistent Differential Poverty

The spatial location of the South African villages can be traced back to the colonial wars in Mozambique. Conversely, the more recently established refugee villages were determined 20 years ago when land was allocated to them on their arrival from Mozambique by the local authorities (see The Data section). In this regard, the temporal–spatial maps illustrate that the five predominantly former refugee villages (12, 17–20 in Fig. 3) had high absolute poverty levels in 2001, and this reduced to two in 2007 (18, 19), which are located in close proximity to the border of a national park. All of these villages are (largely) situated in the south-east sector of the site. Interestingly, some of the poorer South African villages also border the national park (6–8). Conversely, village 11(South African), despite being in the south-east, has lower levels of poverty than all of its neighbouring villages (7, 8, 17). In this regard, village 11 is the oldest in the area, is situated closest to a road and has a health clinic and, generally, more services and infrastructure than the neighbouring villages. A common denominator with respect to all the villages in the south-east is that they are located in a drier area that is more suited to livestock and shows little potential for intensive agriculture. The south-eastern side of the site, moreover, also has lower levels of infrastructure and facilities. There are, however, degrees of remoteness due to access to different roads in the vicinity of the south-east sector of the site. For example, two South African villages (5, 15) are situated relatively close to a tar road (R536) to the south of the site. The difference in poverty levels of former refugee villages, therefore, cannot be ascribed to the spatial location alone. In this regard, former refugee villages have significantly higher levels of mortality (Collinson, 2010), lower levels of education, less access to employment, and lower levels of service provision such as water and sanitation (Hargraeves et al., 2004; Kahn 2006; Twine et al., 2007; Schatz, 2009). Conversely, many of the less poor South African villages are mostly located further to the west and enjoy better natural resources, rainfall and access to infrastructure, facilities, and services (Collinson, 2010). The higher poverty levels in former refugee villages, in general, can possibly be explained as a combination of spatial location influencing access to natural resources and infrastructure, compounded by refugee status, which has led to differential access to opportunities, social networks, and state pensions (Hargraeves et al., 2004; Hunter et al., 2007; Polzer 2007; Rodgers, 2008; Schatz 2009). Finally, the poverty gap between certain villages is perpetuated because of a type of ‘neighbourhood effect’ that results in wealthier former refugees migrating out of the poorer villages (e.g. 18, 19) and settling in wealthier villages (Collinson, 2010).

The Dynamics of Asset Accumulation (SEP)

Household SEP is largely influenced by the long-term effect of the inter-generational transfer of assets (Wu & Pretty 2004; Barrett 2005). SEP, moreover, influences a household's characteristics, in combination with a wide range of other variables, to determine its initial asset status at the starting point of an asset accumulation cycle. In particular, the results also show that older, larger South African households, headed by a man, have higher rates of asset accumulation. Male household heads, in particular, improve a household's economic prospects in rural areas because of improved access to communal resources and social networks (Sen, 2003; Krishna, 2006; Goudge et al., 2009a, 2009b). Furthermore, the receipt of state grants has stabilised the income of many rural South African households (Landau, 2005; Sherbinin et al., 2008; Goudge et al., 2009a, 2009b). A combination of factors, therefore, appears to trigger the ability to invest in higher return opportunities, and wealthier households are the most likely to respond (Binswanger 2004; Barrett, 2005; Dimba and K'Obongo, 2007). Wealthier households also have the resources and social contacts to adopt pull led or high-return strategies, as well invest in education, healthcare, and social status, thus perpetuating their competitive advantage into the future (Kanagawa and Nakata, 2008; Vermeulen et al., 2008; Xing et al., 2008; Goudge et al., 2009a, 2009b). Finally, wealthier (South African) households are better able to fund the costs of shocks such as death and disease and are not forced into survival strategies as a result of them (Sen, 2003; Krishna, 2006; Hosegood, 2009; Schatz et al., 2011).

Conversely, the results indicate that poorer former refugee households have been more recently established and have lower levels of secondary and tertiary education and fewer working individuals. They also had a low initial asset status, and they are physically located in drier, more marginal areas. These households also struggled to secure the necessary legal status to access institutions and employment (Rodgers, 2008; Schatz, 2009; Collinson, 2010) and were unable legally to access state grants until 2006 (Schatz, 2009). A combination of factors, therefore, forced the former refugees to adopt push led strategies that included low-return options such as farm wages, herding, and petty trading (Sen, 2003; Elmquist and Olsson, 2006; Wouterse and Pieterse, 2008). Their relative disadvantage, moreover, was perpetuated by their inability to invest in healthcare and education, as well as the bigger impact of shocks such as death, medical expenses, and drought (Twine et al., 2007).

CONCLUSION AND RECOMMENDATIONS

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

The paper illustrates the complex, and persistent, nature of asset accumulation in a refugee community in rural South Africa. The results of the first research question indicate that, even though the asset accumulation rate of former refugees was equal to that of its host community, they were unable to close the wealth gap. The usefulness of the results is that they empirically demonstrate the persistent long-term impact of refugee status in a rural African community. The usefulness of the results, moreover, are that they develop and compare a comprehensive set of assets generated by a rural community that is host to a refugee community. In this regard, we believe that the construction of a MCA wealth index that includes livestock contributed to the development of a more reliable proxy for SEP in an African context.

The results of the second research question illustrate that the spatial location of the poorer former refugee villages does not fully explain the higher levels of absolute poverty experienced in these households. In this regard, some South African villages in close proximity also display higher absolute poverty levels, but others do not. The usefulness of the results is that geo-statistical-based mapping can pinpoint poverty at an acute local level where neighbouring villages demonstrate significantly different levels in wealth (SEP). In a broader sense, the results illustrate how the dynamic nature of SEP can be modelled by integrating a GIS system with a combination of classical and Bayesian techniques, thus expanding the methodological options for investigating social problems with spatially and temporally correlated data. In particular, the paper recommends the use of Bayesian-based kriging to provide a cost-effective way of geographically targeting villages that require incremental levels of policy intervention. The usefulness of the results, furthermore, is that they demonstrate that spatial location alone is unlikely to have perpetuated the wealth gap experienced by the former refugees.

The results of the third research question illustrated that a range of household characteristics significantly influenced the ability to accumulate assets. Former refugee SEP accumulation rates, for example, were more likely to be compromised by higher mortality rates, as well as poorer education levels and less access to health and high-return local and migrant employment opportunities. The usefulness of these results is that they identify, quantify, and differentiate the impact of certain household characteristics on the asset accumulation rates of former refugees and their host community. In this regard, household characteristics largely determine whether a household has the necessary assets to invest in a high-return versus low-return strategy. Finally, the results of the third research question suggest that differential household characteristics probably better explains the significantly different wealth levels of the villages rather than their spatial location.

The results and discussion can be employed to suggest a number of policy interventions in a rural community that is host to a refugee community. In the short term, the results indicate that the former refugee villages (18, 19) should be targeted for expanding current levels of support through feeding programs, as well as the provision of basic services. In this regard, the results also indicate that policy is required to ensure former refugees' access to state grants, facilities, and employment. Extending the results of the paper, it becomes clear that policy interventions are needed to ensure full legal status for former refugee communities that have settled in communities in order for these residents to access facilities and employment. Furthermore, the different factors (variables) influencing poverty indicate that policy will need to be coordinated across a number of government departments in order to implement rural development initiatives that alleviate poverty. For example, the results indicate that mortality and education are key determinants of wealth, so policy interventions, in this regard, would suggest the need for programs coordinated by the health, education, and finance ministries.

The reliability of our conclusions, however, is influenced by a number of limitations. Firstly, the SEP increases indicated in our results could be inflated because of exogenous factors such as a government electrification program that appears to have contributed to the acquisition of a range of assets for cooking, lighting, and communications in our survey. Secondly, the study coincided with a period of sustained GDP growth in South Africa that is also likely to have stimulated asset accumulation.

REFERENCES

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2
  • Anriquez G, Valdes A. 2006. Determinants of farm revenue in Pakistan. Pakistan Development Review 45(2): 281301.
  • Ardington C, Case A, Hosegood V. 2009. Labor supply responses to large social transfers: longitudinal evidence from South Africa. American Journal of Applied Economics 1(1): 2248.
  • Armstrong P, Lekezwa B, Siebrits K. 2008. Poverty in South Africa: a profile based on recent household surveys. A Working Paper 04/08, Department of Economics, University of Stellenbosch.
  • Barrett CB. 2005. Rural poverty dynamics: development policy implications. Agricultural Economics 32: 4560.
  • Barrett CB, Reardon T, Webb P. 2001 Non-farm income diversification and household livelihood strategies in rural Africa: concepts, dynamics and policy implications. Food Policy 26(4): 315331.
  • Berzborn S. 2007. The household economy of pastoralists and wage laborers in the Richtersveld, South Africa. Journal of Arid Development 70(4): 672685.
  • Binswanger H. 2004. Empowering rural people for their own development. Farewell lecture from the World Bank, November 30.
  • Blasius J, Greenacre M. 2006. Correspondence analysis and relatedmethods in practice. In: Multiple correspondence analysis and related methods, Greenacre M. & Blasius J. (eds). Chapman & Hall: London; 340.
  • Booysen F, Van Der Berg S. 2005. The role of social grants in mitigating the socio-economic impact of HIV/AIDS in two Free State communities. The South African Journal of Economics 75: 545563.
  • Booysen F, Van Der Berg S, Burger R, Von Maltitz M, Du Rand G. 2008. Using an asset index to assess trends in poverty in seven Sub-Saharan African countries. World Development 36(6): 11131130.
  • Bryceson DF. 2002a Multiplex of livelihoods in rural Africa: recasting the terms and conditions of gainful employment. Journal of Modern African Economics 40(1): 128.
  • Bryceson DF. 2002b. The scramble in Africa: re-orientating rural livelihoods. World Development 30(5): 725739.
  • Bryceson DF. 2004. Agrarian vista or vortex: African rural livelihood policies. Review of African Political Economy. 31(102): 617629.
  • Chambers R. 1986. Hidden losers? The impact of rural refugees and refugee programs on poorer hosts. International Migration Review 20(2): 245263.
  • Chayanov AV. 1986. On the theory of peasant economy. Madison: The University of Wisconsin Press.
  • Collinson MA. 2010. Striving against adversity: the dynamics of migration, health and poverty in rural South Africa. Global Health Action 3: 5080. DOI: 10.3402/gha.v3i0.5080
  • Collinson MA, Tollman SM, Kahn K. 2007. Migration, settlement change and health in post apartheid South Africa: triangulating health and demographic surveillance with national census data. Scandinavian Journal of Public Health 35(Suppl. 69): 7784.
  • Collinson MA, Tollman SM, Kahn K, Clark SJ, Garenne M. 2006. Highly prevalent circular migration: households, mobility and economic status in rural South Africa. In: African on the Move: Migration in Comparative Perspective, Tienda, M, Findley SE, Tollman, SM, Preston-Whyte, E. (eds). Wits University Press: Johannesburg; 194216. ISBN 1-86814-432-1
  • Diggle P, Tawn J, Moyeed R. 1998. Model-based geostatistics. Applied Statistics 47: 299350.
  • Dimba B, K'Obongo P. 2007. Influence of culture on strategic human resource management (SHRM) practices in multinational companies (MNC) in Kenya: a critical literature review. A working paper, College of Humanities and Social Sciences, University of Nairobi: Kenya; 19.
  • Dolan CG, Tollman SM, Nkuna VG, Gear VSS. 1997. The links between legal status and environmental health: a case study of Mozambican refugees and their hosts in Mpumalanga (Eastern Transvaal), Lowveld, South Africa. Health and Human Rights 2(2): 6284.
  • Elliot P, Wakefield J, Best JN, Briggs D. 2000. Spatial Epidemiology-Methods and Applications. Oxford University Press: London.
  • Elmquist B, Olsson L. 2006. Livelihood diversification: continuity and change in the Sahel. GeoJournal 67: 160180.
  • Gelfand A, Vounatsou P. 2003 Proper multivariate conditional autoregressive models for spatial data analysis. Biostat 4: 1125.
  • Gelfand A, Ravishanker N, Ecker M. 1999. Modeling and inference for point-referenced binary spatial data. In: Generalized Linear Models: A Bayesian Perspective, Dey D, Ghosh S, Mallick B (eds). Marcel Dekker Inc.: New York; 373386.
  • Goudge J, Gilson L, Russell S, Gumede T, Mills A. 2009b. The household costs of health care in rural South Africa with free primary care and hospital exemptions for the poor. Tropical Medicine & International Health 14(4): 458467.
  • Goudge J, Russell S, Gilson L, Molyneaux C, Hanson K. 2009a. Household experiences of ill health and risk protection mechanisms. Journal of International Development 21(2): 159168.
  • Gwatkin D, Rutstein S, Johnson K, Suliman E, Wagstaff A, Amouzou A. 2000. Socio-economic differences in health, nutrition and population: Malawi 1992, 2000. HNP/Poverty Thematic Group of the World Bank: Washington, DC.
  • Hargraeves J, Collinson MA, Kahn K, Clark SJ, Tollman S. 2004. Childhood mortality among former Mozambican refugees and their hosts in rural South Africa. International Journal of Epidemiology 33: 12711278.
  • Hedeker D, Gibbons D. 1994. A random-effects ordinal regression model for multilevel analysis. Biometrics 50: 933944.
  • Hosegood V. 2009. The developing impact of HIV/AIDS across family household lifecycle implications for efforts to strengthen families in Sub-Saharan Africa. AIDS Care 21(Suppl. 1): 1321.
  • Howe L, Hargreaves J, Huttly S. 2008. Issues in the construction of wealth indices for the measurement of socio-economic position in low-income countries. Emerging Themes in Epidemiology 5: 3.
  • Hunter LM, Twine W, Patterson L. 2007. Locusts are now our beef: adult mortality and household dietary use of local environmental resources in rural South Africa. Scandinavian Journal of Public Health 35(suppl. 69): 165174.
  • IIYama M, Kariuki P, Kristjansen P, Kaitibie S, Maitima J. 2008. Livelihood diversification strategies and soil management strategies: a case study from Kerio Valley, Kenya. Journal of International Development 20: 380397.
  • Jabbar MA, Ehui SK, Von Kaufmann R. 2002. Supply and demand for livestock credit in Sub-Saharan Africa: lessons for designing new credit schemes. World Development 30(6): 10291042.
  • Kahn K. 2006. Dying to make a fresh start: mortality and health transition in a new South Africa [dissertation]. Umeå University: Faculty of Medicine, Public Health and Clinical Medicine.
  • Kanagawa M, Nakata T. 2008. Assessment of access to electricity and the socio-economic impacts in rural areas of developing countries. Energy Policy 36: 20162029.
  • Kariuki SM. 2004. Creating the black commercial farmers in South Africa. ASC Working Paper 56/2004, Department of Sociology, School of Social Sciences, University of the Witwatersrand, Johannesburg, South Africa.
  • Krishna A. 2006. Pathways out of and into poverty in 36 villages of Andhra Pradesh, India. World Development 34(2): 271288.
  • Kuhlman T. 1990. The economic integration of refugees in developing countries: a research model. Journal of Refugee Studies 4(1): 120.
  • Kydd J, Dorward A, Morrison J, Cadisch G. 2004. Agricultural development and pro-poor economic growth in sub-Saharan Africa. Oxford Development Studies 32(1): 3757.
  • Landau L. 2005. Urbanization, nativism and the rule of law in South Africa's forbidden cities. Third World Quarterly 26: 11151134.
  • Lay J, Mahmoud TO, M'Mukaria GM. 2008. Few opportunities, much desperation: the dichotomy of non-agricultural activities and inequality in Western Kenya. World Development 36(12): 27132732.
  • Machethe CL. 2004. Agriculture and poverty in South Africa: can agriculture reduce Poverty? Paper presented at the Overcoming Underdevelopment Conference held in Pretoria, 28–29 October 2004.
  • Malmberg B, Tegenu T. 2007. Population pressure and dynamics of household livelihoods in an Ethiopian Village: an elaboration of the Boserup–Chayanovian framework. Population and Environment 29: 3967.
  • Mendola M. 2008. Migration and technological change in rural households: complements or substitutes? Journal of Development Economics 85: 150175.
  • Mertz O, Wadley RL, Christensen AE. 2005. Local land use strategies in a globalizing world: subsistence farming, cash crops and income diversification. Agricultural Systems 85(3): 209215.
  • O'Brien K. Quinlan T, Ziervogel G. 2009. Vulnerability interventions in the context of multiple stressors: lessons from the Southern African Vulnerability Initiative (SAVI). Environmental Science and Policy 12: 2332.
  • Onyu LP, Neuner F, Ertl V, Schauer E, Odenwald M, Elbert T. 2009. Trauma, poverty and mental health among Somali and Rwandese refugees living in African refugee settlements- an epidemiological study. Conflict and Health 3: 6. DOI:10.1186/1752-1505-3-6
  • Polzer, T. 2004. “We are all South Africans Now:” The integration of Mozambican refugees in rural South Africa. Forced Migration Working Paper #8, Forced Migration Studies Programme, University of the Witwatersrand.
  • Polzer, T. 2007. Adapting to changing legal frameworks: Mozambican refugees in South Africa. International Journal of Refugee Law 19(1): 2250.
  • Rodgers G. 2008. Everyday life and the political economy of displacement on the Mozambique–South Africa borderland. Journal of Contemporary African Studies 26(4): 385399.
  • Sahn D, Stifel D. 2003. Exploring alternative measures of welfare in the absence of expenditure data. Review of Income and Wealth 49(4): 463489.
  • Schatz E. 2009. Reframing vulnerability: Mozambican refugees access to state funded pensions in rural South Africa. Journal of Cross-Cultural Gerontology 24(3): 241258.
  • Schatz E, Ogunmefun C. 2007. Caring and contributing: the role of older women in multi generational households in the HIV/Aids era. World Development 35(8): 13901403.
  • Schatz E, Madhavan S, Williams J. 2011. Female headed households contending with HIV/AIDS related hardship in rural South Africa. Health & Place 17: 598605.
  • Schwarze S, Zeller M. 2005. Income diversification of rural households in Central Sulawesi, Indonesia. Quarterly Journal of International Agriculture 44(1): 6173.
  • Sen B. 2003. Drivers of escape and descent: changing household fortunes in rural Bangladesh. World Development 31(3): 513534.
  • Sherbinin A, Van Wey WL, McSweeny K, Aggarwal R, Barbieri A, Henry S, Hunter LM, Twine W, Walker R. 2008. Rural household demographics, livelihoods and the environment. Global Environmental Change 18: 3853.
  • Twine R, Collinson MA, Polzer T. 2007. Evaluating access to a child-oriented poverty alleviation intervention in rural South Africa. Scandinavian Journal of Public Health 69: 118127.
  • Urassa M, Boerma JT, Isingo R, Ngalulu J, Ngweshemi J, Mwaluko G, Zaba B. 2001. The impact of HIV/AIDS on mortality and household mobility in rural Tanzania. AIDS 15(15): 20172023.
  • Vermeulen S, Nawir AA, Mayers J. 2008. Rural poverty reduction through business partnerships? Examples of experience from the forestry sector. Environment, Development and Sustainability 10: 118.
  • Williamson OE. 2000. The new institutional economics: taking stock, looking ahead. Journal of Economic Literature XXXVII(September): 595613.
  • Wouterse F, Pieterse JE. 2008. Migration and income diversification: Evidence from Burkina Faso. World Development 36(4): 625640.
  • Wu B, Pretty J. 2004. Social connectedness in marginal rural China: the case of farmer innovation circles in Zhidan north Shaanxi. Agriculture and Human Values 21: 8192.
  • Xing L, Shenggen F, Xiaopeng L, Zhang X. 2008. Community poverty and inequality in Western China: a tale of three villages in Guizhou Province. China Economic Review (2008). Doi: 10.1016/j.chieco.2008.10.006

APPENDIX 1

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

Assessing the asset-based wealth indices

Measurements of agreement are of great importance for assessing the acceptability of a new process. The three indices were compared with each other for misclassification of households between tertiles of indices. Kappa statistics (measure of reliability based on the agreement expected on the basis of chance) were calculated to assess the agreement of classification between indices. Kappa is a preferred statistic to estimate agreement for nominal or ordinal scale data (in our case, the variables are ordinal SEP indices). A kappa statistic of 1 indicates perfect agreement and a value of zero indicates no agreement. In general, a kappa statistic of less than 0.5 indicates poor agreement. Misclassification between tertiles was used as the measure of agreement (Howe et al., 2008).

The relationship between the methods, illustrated in Table 6, indicates a significantly high level of agreement between the multiple correspondence analysis (MCA) and principal component analysis (PCA) indexes (87%) and slightly less with the absolute count analysis (ACA) index (82%). For those where there was disagreement, all were only one tertile shift away. When comparing this disagreement for MCA versus ACA, we found that the majority of this shift was for ACA score to be one tertile below (6,152 of 8,043 or 76.5%) whereas for MCA versus PCA, this was similarly distributed above or below at 50.9% (3,081 of 6,054). This leads to the conclusion that the ACA generally underscores relative to MCA and PCA indices.

Table 6. Percentage of households (n = 45,411) in the same tertile and kappa statistics of agreement between pairs of indices.
Wealth indexAbsolute asset countPCAMCA
  • PCA, principal component analysis; MCA, multiple correspondence analysis.

  • *

    p < 0.001, expected agreement = 33.3%.

Absolute asset count
PCA83%; κ = 0.748*
MCA82%; κ = 0.734*87%; κ = 0.800*

A decision must be made about which weights to assign to each indicator variable when constructing a wealth index. Using an equal weights approach (e.g. ACA), although simple, may be too arbitrary and simplistic, as different assets are unlikely to have equal meaning in terms of SEP. It is for this reason that an MCA tertile index was selected for use in the final univariate and multivariate analysis, as it is it more appropriate for the analysis of the categorical data commonly collected on most assets (Howe et al., 2008) and reduces the computation time of the Bayesian model compared with a tertile index. Moreover, a PCA index was designed for use with continuous variables whereas MCA makes fewer assumptions about the underlying distributions of the indicator variables and is more suitable with categorical or discrete data such as those in our study (Howe et al., 2008).

The direction of poverty trends can also sometimes be different when using a PCA versus an MCA-based asset index (Booysen et al., 2008). The most important difference, however, is that PCA requires linear constraints (i.e. assumption that the distance between categories are the same and ordered) whereas MCA imposes fewer such constraints (Blasius and Greenacre, 2006). It should be noted that all the asset indicator variables tracked at the Agincourt field site are categorical which, as mentioned above, reinforces the use of MCA.

APPENDIX 2

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. A CONCEPTUAL FRAMEWORK
  5. DATA AND METHOD
  6. RESULTS
  7. DISCUSSION
  8. CONCLUSION AND RECOMMENDATIONS
  9. REFERENCES
  10. APPENDIX 1
  11. APPENDIX 2

Multivariate statistical model

The Bayesian ordinal model we used is based on that developed by Hedeker & Gibbons, 1994. It assumes that the observed socio-economic tertile Yi,t of household i (i = 1,…,) at year t (2001, 2003, 2005, 2007) follows a categorical distribution, Yi,t ~ Categorical(pi,t, 1:C) with probability pi,t. A continuous scale may be envisaged to underlie the SEP tertile, with a series of thresholds τ1, τ2,…, τc − 1 defining which of C categories (tertile) a household lies in.

The likelihood model for household i at year t is then

  • display math

with the various models as follows

  1. multivariate non-spatial model μi, t = β0 + βXit + ui
  2. mutlivariate spatial model μi, t = β0 + βXit + ui + φit

The spatial kriging model was specified differently as follows:

  • Continuous MCA asset score for household i and year t is normally distributed
  • MCAi,t ~ Normal(mui,t, τ) where
  • mui,t = β0 + φit
  • and τ represents the precision (1/variance) of the Normal distribution and assumed ~ Gamma(1,1).

For all models, β0 is the intercept, Xit is the covariate vector (time constant and varying), and β is a vector of unknown regression coefficients. The regression coefficients (β) as well as the constant (β0) was given non-informative normal priors centred around zero, namely ~ Normal (0,0.1). To account for the spatial correlation, we introduced a village-specific random effect, φit. A household level random effect (ui) for repeated longitudinal observations on a household was also introduced.

The household level random effect, μi, was modelled as a Normal distribution, μi ~ N(0, inline image).

The village-specific random effect has a multivariate normal distribution, φi ~ MVN (0,Σ), with variance–covariance matrix Σ expressed as a parametric function of distance between pairs of the 21 village centroids points. We also assume an isotropic stationary spatial process, where inline image, dkl is the Euclidean distance between villages k and l, inline image models the geographic variability, and ϕ is a smoothing parameter that controls the rate of correlation decay with increasing distance and measures the range of geographical dependency. We specified ϕ as a uniform distribution between ϕmin and ϕmax (Gelfand & Vounatsou, 2003). The range is defined as the minimum distance at which spatial correlation between locations is below 5%. This distance can be calculated as 3/u meters.

We choose vague Normal distributions for the β parameters centred around zero with variance 0.1 and non-informative gamma distribution priors for all σ2 with means and variances of 1.

MCMC simulation was applied to fit the models. We run a single chain sampler with a burn-in of 5,000 iterations. Convergence was assessed by running the simulation until the Monte Carlo error for each parameter of interest was less than 5% of the sample standard deviation. The chains thereafter were sampled every single iteration until a sample size of 10,000 had been attained.