Does poverty constrain migration in South Africa? Evidence, explanations and implications

Abstract

There are indications that poor people may face obstacles to their rural–urban migration. This article pursues the question of whether this is the case in the South African context. It argues for the importance of longitudinal data, which are not available at present, to answer this question conclusively. Levels of education can be used as a proxy for income levels, and the article examines recently published multivariate data in this regard. However, using education as a proxy for income is problematic, because education has an independent effect on migration rates through its selection of those with the skill levels demanded by the labour market. The article develops an argument about the constraining effects of the costs of migration and the role of social networks in migration and ends by demonstrating how the costs of migration can solve a number of puzzles presented by empirical research on migration.

1. Introduction

Moving from a poor area with few opportunities to a wealthier area is one way poor people can improve their situation. A positive outcome is by no means guaranteed, of course, given the risk that the resources invested in migration may go to waste as a result of unsuccessful migration; nevertheless, most migrants report that their situation has improved since migrating (Population Information Programme, 1983).

In the light of the importance of migration, it is significant that there seem to be constraints on the migration of the poor in South Africa. This observation is borne out by the disjuncture between the intentions to migrate and actual migration in South Africa. In a nationally representative sample survey in 2001/2002, 25.4 per cent of the population expressed an intention to migrate over the following five years (De Jong & Steinmetz, 2006: 263). However, until 2001 the overall migration rate remained relatively stable, at between 11 and 13 per cent every five years for a number of decades (Kok et al., 2003: 55). This rate prevailed despite a reorientation of migration streams from a rural–rural to a rural–urban pattern over this period. Unless there has been a dramatic increase in the migration rate since 2001, this indicates that only about half of the intentions to migrate are acted upon. Although we do not know for sure, it makes sense to think that the poor, who have less control over their lives than others, will be more likely to have their migration intentions thwarted. This has indeed been the experience elsewhere in the developing world, where the poor are generally under-represented in migration streams (Skeldon, 2003). Obstacles to the migration of the poor could also provide an explanation for why rates of rural–urban migration increased less dramatically than commonly expected after influx control was lifted in 1986. According to Barbara Anderson (2006), during the 1990s rates of rural–urban migration did increase, but they were still lower than in the rest of sub-Saharan Africa.

This paper seeks to establish whether poverty affects migration rates. If it has a negative effect on migration rates it also reduces the chances for self-improvement implied by migration. This will create a positive feedback cycle in which poverty reduces migration, which will in turn reinforce previously existing poverty (see Gelderblom, 2006a). The effect that poverty has on migration occurs within a context where other factors such as level of education and social networks also come into play. These two factors consequently also receive attention in this paper, and an attempt is made to point to the interactions between poverty on the one hand and education and social networks on the other. The focus of the paper is in addition more on voluntary migration, thereby excluding movements caused by war (refugees and internal displacement), environmental change (global warming and land degradation) and epidemics.

The paper departs from the assumption that geographical location influences access to employment and the nature of the jobs on the market. In other words, the labour market in South Africa is regionally stratified. Evidence for this is provided by Dinkelman and Piroux (2001), who claim that location – urban or rural – is one of the most important determinants of whether a working age unemployed person will be searching for a job or not. The rural location of many unemployed people helps to explain why such a large proportion of the unemployed are not actively searching for employment; the idea being that the distance from employment opportunities, as well as the large numbers of unemployed living there, are obstacles to search activities. Relocation to an urban area will therefore facilitate job search. Rural–urban migration is not the only type of migration that can improve the situation of the poor. Any kind of relocation to an area that provides better life chances for a poor person with particular skills and contacts would be favourable, almost by definition. In the South African context an example would be migration out of declining mining or industrial towns such as the Free State goldfields towards a developing industrial centre such as Richard's Bay.

Geographical location is of course only one factor out of many that has an influence on people's life chances and, at that, not the most important one. It certainly plays second fiddle to factors such as class, race and gender. My claim about the impact of migration on the life chances of the poor is therefore a modest one. However, migration is a frequently neglected factor. It also interacts with these more important factors in often unexpected ways. It consequently deserves closer scrutiny, which is what I attempt here.

After clarifying the concepts of poverty and migration as they are used in this paper, I look at recently published multivariate data as they relate to poverty and migration. I note though with concern the unavailability of longitudinal data about pre-migration income levels, which are important for answering this question conclusively. Levels of education are often used as a proxy for income levels. It is, however, far from ideal to use education as a proxy for income. Education affects migration rates independently from income through its selection of those with the skill levels demanded by the labour market. The effects of income and education will thus be scrambled if education is used as a proxy of income. In the next section I develop arguments I have made elsewhere (Gelderblom, 2006b; Gelderblom & Adams, 2006) about the constraining effects of the costs of migration and the role of social networks in migration. I end by demonstrating how the costs of migration can solve a puzzle presented by Kok et al.'s (2003) analysis of the role of distance in rates of migration in South Africa.

2. Poverty and Migration: Defining the Concepts

The notion of ‘poverty’ has proven to be remarkably difficult to define (Gordon, 2000; Saunders, 2000; Townsend, 2000; Conley, 2005). It refers in general to an inability to buy the necessities of life due to a low income, or a state of having difficulty in making ends meet (Saunders, 2000). It has been difficult to operationalise this definition, however, because what are to count as necessities of life depend on the general standard of living and therefore vary from society to society. Some critics also claim that there is an inevitably subjective dimension to what is regarded as an acceptable standard of living (Conley, 2005). The conceptual waters have been further muddied by Amartya Sen's attempt to define poverty in terms of limitations to a wider basket of human capabilities, of which material deprivation is only one. Sen's emphasis on factors such as health, political capabilities and gender relations as additional obstacles to human flourishing is certainly valid. However, it does not remove the need to have a concept to refer to the specifically material dimension of deprivation (Conley, 2005). Sen's wider conception of poverty will also inevitably compound the difficulties of arriving at an acceptable operationalisation of poverty. It is therefore desirable to restrict poverty to material deprivation in order to arrive at an empirically useful conceptualisation. It is in addition clear from the work of Townsend (2000) and Gordon (2000) that it is possible (and politically necessary) to arrive at reasonably objective definitions of poverty lines.

My interest in poverty in this paper concerns the extent to which it can reduce the affordability of the costs of migration. I will therefore focus exclusively on the material dimension of poverty. I conceive of it as a state where an individual and/or household controls too few material resources (conceived here widely as income in cash and kind derived from all sources, including transfer payments, as well as liquid assets) to partake in the normal lifestyle of a society. It is of course not easy to operationalise this concept, but this is not a concern in the present context where my interest is more theoretical. In implementing this concept cognisance will have to be taken of redistribution relationships within households, based on age and gender, which may limit the access that any specific household member may have to household resources.

Like all forms of mobility, migration is literally a movement in space at a particular time, and having a particular duration. The dimensions of space and time are therefore essential to describing migration and to distinguishing it from other forms of movement. This has been pointed out most elegantly by the Swedish geographer, Torsten Hägerstrand (1969; also see Pred, 1977), who is the father of time-geography. He developed a technique for sketching the time–space paths of individuals with great physical realism. This technique tracks individuals as they move between ‘stations’ such as home and work over different time scales, and across social transitions such as the ending of the labour tenancy system. His methodology is not suitable, however, for statistical analysis at the macro-scale.

The typology advanced by Pieter Kok (1999) and Kok et al. (2003: 9) provides a good compromise between the demands of quantitative data collection on the one hand and close attention to the dimensions of space and time on the other. In terms of the space dimension, migration is regarded as a movement over a relatively large distance, unlike local residential mobility, for example. Because distance is so difficult to account for in large-scale studies, the question of whether a migration-defining boundary has been crossed or not is used as an indicator of longer distance. The time dimension is captured by the duration of the change in residence implied by the movement. Weekly movement across a migration-defining boundary is regarded as short-term labour migration. Movements across such a boundary of longer than a week but not implying a permanent change of residence is called long-term labour migration, while a permanent change of residence is regarded as permanent migration.

3. Poverty and Migration: Recent Empirical Evidence

Kok et al.'s (2003) multivariate analysis of the 10 per cent sample of the South African census of 1996 seems to provide evidence that poverty does indeed limit migration, but unfortunately this conclusion does not survive critical scrutiny. Following the census, these authors distinguish between labour migrants and migrants proper (Kok et al., 2003: 61). Migrants, in their view, are those people who have made a permanent change in address during the last five years. Labour migrants, by contrast, are those people who are absent from their homes for more than a month each year for the purpose of employment. In their logistic regression, they find that the relative probability of having been a migrant increases significantly if individual monthly income rises above R3000, and reaches its maximum at a monthly income of around R13 500 (Kok et al., 2003: 62). Almost all labour migrants, by contrast, have individual monthly incomes lower than R3000. The probability of being a labour migrant is most strongly predicted by a low level of education, with a negative relationship between level of education and the probability of being a labour migrant. The probability of being a permanent migrant, by contrast, is most strongly predicted by the possession of a post-secondary education.

Kok et al.'s analysis highlights some of the problems involved in testing the effect of income on migration. We can say for certain that labour migrants are now (that is after migrating for the first time) poorer and less educated than permanent migrants, but their analysis does not provide information about the situation before the move. The current values of these variables (which is what were measured by the census) do not necessarily correspond to their values at the time that migration took place, especially if that was long ago, because both income and (to a lesser extent) educational levels change over time. It is therefore impossible to make any causal connections between income/education and the decision to engage in either migration or labour migration. The best solution to the problem of knowing whether income differentials have manifested themselves before or after migration is to study the income-selectivity of migration in a longitudinal fashion (Connell et al., 1976; Davanzo, 1981: 121). In this kind of study, a particular sending area is surveyed at one stage. At a later stage, an attempt is made to find out which households have since relocated. The impact of income on migration can then be estimated by comparing the pre-migration income of migrant households with that of non-migrant households in the community of origin.

Van der Berg et al.'s (2004) analysis of the 10 per cent sample of the 1996 census tries to overcome this problem by ignoring the income variable altogether and concentrating on levels of education. In addition, they restrict their analysis to those who have migrated in the nine months of 1996 before the census. As a result, the education variable (which is less likely to have changed over only nine months) becomes a useful predictor of who will migrate and who will not. This will not work for income, as we saw above, because the act of migration is likely to change income levels.

Van der Berg et al. (2004) specifically look at migration from the former Transkei homeland (the poorest part of the Eastern Cape) to Cape Town and other metropoles, and to a lesser extent at movement from Limpopo Province to any metropole. Unfortunately, they do not provide information about the other provinces. They found that, in general, migration from Limpopo Province and the former Transkei homeland to any metropole is higher among people with secondary education than among people with primary or no education, after the effect of other variables has been controlled for. It is also higher among those with tertiary education than among those with primary education or none, but less dramatically so. These trends are finally much stronger in the case of migrants from Limpopo Province than among those from the former Transkei homeland. Unfortunately we do not know which of the two provinces is the more typical as far as the impact of education on migration is concerned, or why there is such a big difference in the impact of educational levels on the propensity to migrate.

When Van der Berg et al. (2004) disaggregate the Transkei figures, a clearly gendered trend appears. Secondary and tertiary education have opposite effects for males and females on the propensity to migrate: men with a tertiary education are more likely to migrate to any metropole than men with a secondary education, while women with a tertiary education are less likely to migrate to a metropole than women with a secondary education.

Because longitudinal data on income levels are not presently available in South Africa, Van der Berg et al.'s (2004) strategy to use data on education can be seen as the next best solution. Because there is generally a good correlation between educational levels and income, their data can also be interpreted to mean that people with a low income will be less likely to migrate, which provides (indirect) evidence for the claim that a low income is an obstacle to migration. However, this is not an unproblematic assumption, since education has an impact on the propensity to migrate that is independent from income, among other things. As a result, it is impossible to say what amount of variation in the propensity to migrate can be attributed to a low educational level and what to a low income when education is used as a proxy for income. In such an analysis the effect of income and education will be scrambled together. In the next section I demonstrate the likely impact of education on the propensity to migrate, which will provide a clearer statement of the problems involved.

4. The Changing Labour Market in South Africa

The impact of education on the propensity to migrate operates through the demand for particular kinds of labour, and by implication the nature of the post-apartheid labour market. The analysis conducted by Burger (2004), among others, has indicated that since 1994 those sectors of the economy that have employed a high proportion of unskilled labourers, such as mining and agriculture, have shed large numbers of jobs. And within those sectors the skill demands for the remaining jobs have been climbing. The sectors employing more skilled workers, such as finance, real estate and business services, by contrast, have been adding large numbers of new jobs. Consequently the demand for unskilled workers has slumped and the demand for semi-skilled and skilled ones has boomed. Only the better educated are therefore in a position to benefit from relocating to areas where their skills are highly valued. By contrast the unskilled are unlikely to find their labour being demanded elsewhere.

Van der Berg et al. (2004) investigated this as an explanation for the limitations placed by a low education on migration by analysing the 1993 October household survey data. They attempt to arrive at a conception of how a person with the human capital attributes of a typical rural dweller in the Eastern Cape will fare in both rural and urban labour markets. They come to the conclusion that ‘the rural employed make up the bulk of those favoured by the urban labour market while the rural unemployed make up the greater share of those favoured by the rural market. Aside from a small pocket, there seems to be little scope for improving the dire predicament of the rural unemployed by moving to an urban area’ (Van der Berg et al., 2004: 25). As can be expected, the better-educated rural people will do best in the urban areas, while the least educated will fare the worst.

These findings are very interesting. However, as I said above, the effects of education and income are likely to be scrambled together. Because Van der Berg et al. (2004) did not control for the effect of household income it is impossible to say to what extent they are observing the impact of education as opposed to that of income.1 Finding a way to account for the effect of household income is therefore essential if we are going to build a proper understanding of the determinants of migration in South Africa.

5. The Costs and Risks Of Migration as an Obstacle to the Migration of the Poorest

While the education variable affects the propensity to migrate through the demand for particular types of labour, the income variable operates through the inability of the poor to bear the costs and risks of migration. In this section I further develop arguments I have made elsewhere (Gelderblom, 2006b; Gelderblom & Adams, 2006). The potential migrant has to consider the following monetary costs in the decision to migrate: transport and relocation costs, the cost of lodging in town while looking for work, and the costs associated with acquiring information about work and housing opportunities. If these costs are too high relative to the resources of the poor, they will act as constraints on their migration.

In addition to the above-mentioned costs, there are also more subtle costs that are not so easily quantifiable. These concern what Fischer et al. (1997: 75–6) call ‘insider advantages’. According to these authors, ‘part of the abilities and assets of every human being are location-specific’. This means that they can only be used within a specific place and are not transferable to other places. For middle-class people they may include information about how to navigate in a specific city, which hairdresser, dentist, etc. to use, the location of good restaurants, which neighbourhoods to avoid at night, and so on. Similar examples for poor people are knowledge about which shops provide credit, where to find water and firewood, which neighbours or patrons to approach for help and how to access services such as clinics and grants. These abilities and assets include skills such as cattle herding or bead working, and knowledge about local customs, which may only be appropriate to the local environment. They take time, effort and money to acquire and if they are not transferable moving away means that this investment is lost. For Fischer et al. (1997) the costs of having to replace these advantages with new ones in a new environment provide an explanation for why so few people move despite the potential advantages that can be derived from migration. Because the poor are less educated, and because they have less information about other places, it is likely that more of their abilities and assets will be place-specific, thus increasing the opportunity costs of moving away and rendering them less likely to move.

For the migrant, the move to the destination area may be quite risky if the new environment is unknown, or even potentially dangerous. Information reduces risks (DaVanzo, 1981: 96), and the unequal access to information to reduce these risks will act as an additional bias against migration among the poorest. Poor people are also often very risk averse, because their coping mechanisms are so limited (Lipton, 1980). This consideration of risk of course has to be balanced by the observation that, if people's current situation is desperate, future risks may seem insignificant compared to current risks of destitution (Fischer et al., 1997: 59).

Transport and information costs increase with the distance of the move (DaVanzo, 1981: 110–111; Population Information Program, 1983: M255; Massey, 1988). So does the amount of risk involved in the move. The further away a migration destination is, the less risk-reducing information is available about it, and the more difficult is it to organise assistance from local sources if something goes wrong while living there. Since the journey itself is risky to some extent (depending on the means of travel), longer distances imply more risk. One can therefore assume that the constraining impact of costs and risks would be more apparent in the case of people who live far from migration destinations. This statement is confirmed by the fact that the poorest are more likely to migrate over short distances (Lipton, 1980; DaVanzo, 1981: 110–11).

The relationship between distance and the costs of migration is modulated by the availability and quality of transport and telecommunication networks between places. As a result, there is no simple one-to-one relationship between the two variables. It can thus happen that it is cheaper to travel a large distance between a regional centre and a metropole than a much smaller distance between two small villages. According to Brown and Lawson (1985: 425), the effect of distance is significantly stronger in the case of rural-to-rural migration. The reason for this is that ‘knowledge of rural economic conditions is more severely attenuated by distance; thus increasing both monetary and psychic movement costs (per unit distance)’.

Another factor modulating the impact of distance on the costs of migration is the existence of migrant networks between the origin and destination areas. If a migrant has friends and family members at destination B but not destination C, the costs faced in moving from place A to place B will be lower than in moving from A to C. Migrant networks therefore mean there will be more migration among the poor than would otherwise have taken place (Massey et al., 1994; Massey et al., 1998). This effect may be limited, however, owing to constraints on network functioning. There are a variety of reasons for this (cf. Gelderblom & Adams, 2006).

The first is that the networks of the poorest may be less able to offer help because they control fewer resources. Bhorat et al.'s (2001: 37–9) analysis of the clustering of unemployment in households in South Africa gives one example of how this can happen. These authors found that the majority (56.5 per cent) of the unemployed live in households that contain more than one unemployed person. The average formal employment rate of households varies from 40.6 per cent for those with one unemployed member to only 14.4 per cent for those households with more than two unemployed members. Because so few of the immediate family members of potential migrants in the latter group of households are employed themselves, little help will be forthcoming from immediate family members with regard to contacts for job placements. I base this on the assumption that the employed are a far better source of knowledge about job opportunities than the unemployed.

Poor people may, in addition, be unable to maintain their network connections in the face of disruptions brought about by disease, death and family conflict, as well as the attenuation of relationships through distance. In their case study of a Ceres township, Arnall et al. (2004) found that people prefer to call on family members for assistance because this help was seen to be more unconditional. However, most of their family members lived in the Eastern Cape and were too far away to be of much assistance. They therefore had to ask their neighbours and friends for help. These help where they can, but expect to be helped in return when they are in need. People who are too poor to reciprocate in future find it more difficult to remain part of such mutual help networks, and can therefore expect less help to begin with.

These examples demonstrate that migrant networks may be of quite limited help to the poorest who want to migrate, and the costs of migrating are indeed an obstacle to them, with the result that there will be less migration. Not all poor individuals will be equally discouraged by these costs, however. Those who are prepared to suffer more severe deprivation and to carry higher risks in pursuit of a life elsewhere will be less discouraged than others. Poor people can, if they are desperate, carry the costs of migration on their bodies, so to speak. I call this internalising the costs of migration. If they cannot afford transport costs, they can walk or cycle to employment possibilities. If they do not have money to tide them over while they are looking for work, they can go hungry and sleep rough. This, however, increases the physical risk of migration exponentially. Exposing oneself in this way increases the likelihood of being attacked and robbed along the way, or suffering illness and disability due to exposure to the elements. Women and older people are especially vulnerable to these increased risks, and can be expected to be more easily discouraged. The class-differentiated constraints imposed on female migration by increased physical risk are a special case of the more general limitations imposed on female mobility by threats to their safety. From this discussion we can conclude that poor people therefore do not altogether lose their agency as a result of the costs of migration, but that their ability to internalise these costs is limited.

The notion of internalisation of the costs of migration helps to explain why the poor may be involved in distress migration away from disaster areas, even though the monetary costs of doing so may appear to be high. I am thinking here in particular of the movement of refugees from places such as Mozambique to South Africa during the 1980s. People often walked on foot through the Kruger National Park, risking injury and death from electric fences and wild animals.2 These were cases where the migrants were prepared to internalise the costs of migration because the alternative of continuing to live in Mozambique was considered to be so much worse. It has to be pointed out that these are extreme circumstances where the alternative to migration is possible death from starvation or conflict.

To summarise the argument thus far: some of the poor may face obstacles to rural–urban migration. Those who live in more distant areas that are far from transport and communication links will be more constrained by the costs of migration than others who are more favourably located. This statement applies particularly to those whose social network connections to people who have previously migrated are weaker, such as the poorest, because they will be less likely to receive assistance with their migration. This does not mean that people in this category of potential migrants are completely powerless to change their situation, however. For those who cannot afford the long haul of migration to the major metropolitan areas, step migration is an option (Lipton, 1980). This takes the form of migration within the former homeland areas from more isolated to more centrally located settlements. Catherine Cross (2006), among others, has remarked on the large amount of migration that has taken place over the last few decades within the former homeland areas of South Africa. People have congregated in settlements that are on transport routes to employment opportunities in the metropolitan areas and that enjoy better access to services. Besides a search for access to services, which is the motive suggested by Cross, this movement can also be interpreted as step migration by the poor in their march to the major metropolitan areas. It should be clear that the costs of migration provide a ready explanation for this movement.

6. The Impact of the Costs and Risks of Migration in South Africa: Evidence

The question now arises as to how one would go about testing the assertion that the costs and risks of migration are an obstacle to the migration of the poorest. Given the broadly positive relationship between distance and the costs and risks of migration, evidence that the poor tend to migrate over shorter distances in South Africa will provide support for this explanation. As we saw above, such a tendency has indeed been observed elsewhere in the developing world. I emphasise the broad nature of the relationship between distance and costs because, as we saw above, the quality of the transport and telecommunications infrastructure modulates the impact of distance on costs, as do social networks. In addition, not all migration costs are distance related (e.g. the cost of board and lodging while looking for work). As a result the fit between distance and costs is imperfect.

There is some evidence that costs are an obstacle to migration in the South African context. Mlatsheni (2004: 65) observes that the distance of areas of black settlement from employment opportunities and the costs of looking for work are perceived by young people as obstacles to job search. The impact of the costs of job search is also attested to by Dinkelman and Piroux (2001). They find proof for this in the large numbers of discouraged work seekers in South Africa.

In this country there is a large gap between the narrow and broad unemployment rates. Narrow unemployment is defined as the state where people will accept work if it is offered, and where they have actively sought employment in the last two weeks. Broad unemployment is all those cases where people would accept work if offered, whether they have searched or not. According to the 1997 October household survey, 21.2 per cent of people were unemployed and looking for work. The broad unemployment rate, however, was far larger at 36 per cent, indicating that large numbers of people were no longer searching for work, or had never done so (Dinkelman & Piroux, 2001). The same tendency is apparent from more recent figures, with unemployment rates increasing but the gap remaining the same. According to Du Toit (2005), the narrow unemployment rate was 26.2 per cent in 2005 and the broad unemployment rate 41.2 per cent. In the former homeland areas of South Africa this gap becomes even larger (Dinkelman & Piroux, 2001).

The challenge now is to explain why there is such a large difference between the two unemployment rates, especially in the former homeland areas. Based on data from the 1997 October household survey, Dinkelman and Piroux (2001) develop a model to predict in which of three states a non-working individual will find him or herself: unemployed and searching for work, unemployed and not searching for work, or out of the labour force. Besides the obvious spatial element to these distinctions, with rural districts with high unemployment rates emerging as a major predictor of non-searching unemployment, the possession of a matriculation certificate (rather than educational levels in general) features strongly in predicting why an individual will be looking for work. This agrees with the findings of Van der Berg et al. (2004) cited above. Dinkelman and Piroux (2001) also test for the impact of household income. Because we are dealing with unemployed individuals who are presumably not earning an income, we do not face the problems referred to earlier with regard to modelling the impact of household income on the propensity to migrate. It seems that household income has a contradictory effect on distributing people between these three states. On the one hand, an increase in this income makes people less likely to be part of the labour force, presumably because it reduces the imperative to look for work and earn an income. On the other, it changes the distribution between the searching and non-searching unemployed, suggesting that an increased household income facilitates job search. This would be consistent with the hypothesis that a low income is a constraint to job search because of an inability to fund the costs associated with it. Despite being statistically significant, it has to be pointed out, however, that the effect of household income is not very strong.

The relevance of Dinkelman and Piroux's (2001) work to this article should be apparent. The cost of job search is one of the costs typically involved in migration, and if that is a constraint to low income households it will reduce the amount of migration taking place.

Besides these studies, the closest researchers have come to an empirical test is Kok et al.'s (2003: 44–7) finding about the importance of distance in selecting which magisterial districts will provide migrants to Gauteng Province. They find evidence of two distinct patterns. In the first there is a strong negative correlation to the effect that distance explains 40 per cent of the drop in the contribution rate of a district to Gauteng. This applies to all districts that were not previously part of a homeland. The second pattern is found in the former homeland districts, where this relationship practically disappears. Here, distance explains only 3 per cent of the drop in the contribution rate. Kok et al.'s (2003) analysis is limited by the fact that they did not investigate the effect of district distance from Gauteng for different socio-economic classes of migrants. The effect of distance is expected to be much stronger for the poorest because they are least able to afford the costs and risks of migration. Part of the effect of migration costs will therefore be obscured by reliance on figures not disaggregated by socio-economic background. It does, however, suggest that, at least for those poor people who do not live in the former homeland areas, the costs of migration are a significant obstacle.

The striking difference observed by Kok et al. (2003) between the effects of distance in the former homeland districts and districts in the rest of the country in terms of reducing migration to Gauteng is nevertheless puzzling. In my view, three possible explanations for this difference suggest themselves. In the first place, the migrants from the former homeland areas to Gauteng may be from a higher socio-economic background and thus be less affected by the costs of distance than Gauteng-bound migrants from the rest of the country. In the absence of disaggregated figures we can only speculate about what the situation is, but this is unlikely. The rest-of-the-country-to-Gauteng category contains inter-metropolitan migrants of all race groups (that is from Cape Town, Durban, etc.) which one can safely expect to be the migrants with the highest socio-economic background. Because of this, we can expect the average socio-economic status of rest-of-the-country migrants to be higher that of former homeland migrants. It is thus very unlikely that the difference between former-homeland and rest-of-country migrants to Gauteng with regard to the effect of distance can be explained as a result of the former having a higher socio-economic status.

The second explanation for the difference is that the extensive long-distance commuting from former-homeland peri-urban settlements to Gauteng reduces the need to migrate to Gauteng to access work there. One of the enduring legacies of the apartheid settlement system is the large number of people who commute to work in Gauteng over long distances (up to 80 km one way) every day from a home base in the former homeland areas. This pattern is made possible by subsidised public transport. As a result, it is not necessary to relocate to Gauteng to access work there, which reduces the incentive to migrate and thus the proportion of ex-homeland moves up to the first 80 km or so away from Gauteng. This reduction in the number of close moves in the ex-homeland category will have some effect in reducing the negative correlation between distance and contribution rates. It is, however, unlikely to be more than a partial explanation for the patterns observed by Kok et al. (2003).

Kok et al. (2003: 48) refer in passing to a third possible explanation, which is the effect of social networks. They do not develop this explanation, however, and I will try to do so here. I have mentioned above that the effect of distance on the costs of migration is modulated by the effect of migrant networks. These networks reduce the costs of migration between a specific area of origin and a specific destination area for those with access to the networks. The existence of pre-existing migrant networks between the ex-homeland areas and Gauteng province will help to explain why the impact of distance is less severe for migrants from the former areas, and why the contribution rate does not fall steeply as the distance from Gauteng increases.

This still leaves a few questions unanswered, specifically: ‘Why is there such a big difference between the patterns observed for the ex-homeland areas and the rest of the country?’ and ‘What made the initial migration possible on which the eventual migrant network was built?’ The latter question becomes relevant if we consider that a migrant network is based on social connections between previous and potential migrants between two areas. Saying that migration is now possible because networks subsidise their costs does not answer the question of how the first migrant from a distant area managed to migrate; it merely transfers the question about how the obstacles imposed by distance were overcome from current migrants to the situation of the original migrants.

To answer these questions necessitates a closer historical understanding of the migrant labour system in South Africa. There exists a voluminous literature on the origins of this system, from which we can extract two relevant, if somewhat over-generalised, points. The first is that there was a more extensive history of migrant labour from the ex-homeland areas to Gauteng than from the rest of the country to Gauteng. This division between the homeland areas and the rest of the country became more pronounced as the labour tenancy system came to an end on the white farms towards the middle of the 20th century.

Labourers and their families who had previously engaged in some migrant labour in the off-season now left the white farms to settle in the homelands, using their new homes as a base to intensify their involvement in the migrant labour system (Gelderblom, 2004). Consequently these areas have long-standing and well-developed migrant networks compared to the rest of the country. This helps to explain the first question about the differences in the migration patterns between the two parts of the country. The second point is that employers experienced a severe shortage of cheap black labour from the beginning of capitalist development towards the end of the 19th century until the 1970s. The situation was therefore completely unlike the present with its glut of unskilled labour. As a result, big employers of unskilled labour such as the mines, parastatals and local authorities engaged in extensive recruitment of labour from the homeland areas. They made use of a network of recruitment agents and offices spread across the homelands, regularly gave cash advances to potential migrants and arranged transport to as well as free accommodation at the place of employment. This reduced the money costs of migration carried by the pioneers in labour migration to practically nothing (this of course did not offset the harsh conditions and low wages under which they had to work). As a result, distance did not affect the extent of initial migration between the homeland areas and Gauteng province, which will help to explain the current absence of such a relationship.

In the context of the rest of this article it is appropriate to emphasise that Kok et al.'s (2003) analysis does not say anything about the impact of distance on the migration of the poorest per se. Their analysis is not disaggregated on class lines, which is the information that is needed here. The historical explanation advanced above does, however, suggest that migrant networks play a stronger role in reducing the costs of migration from the ex-homeland areas to Gauteng than from the rest of the country to Gauteng, which should increase the amount of migration in which the poor from these areas can engage.

Kok et al.'s (2003) finding that the poor are more likely to be migrant labourers, referred to above, provides support for the contention that, in this case at least, the risks of migration are an obstacle. There is a considerable literature on migrant labour as a risk-reducing strategy for rural households (cf. the essays contained in Stark, 1991, and Lipton et al., 1996). The emphasis in this literature is generally on the impact that an urban income can have in reducing the rurally generated risks faced by a rural household, in the sense that the urban income helps to increase the household's overall number of income generating activities. The obverse of this is just as true: by retaining access to rural resources and the (limited) amount of security they provide, a migrant can reduce the urban generated risks attendant upon rural–urban migration. The migrant is therefore less exposed to threats arising from unemployment in town. Because the risks of migration are higher for the poor, they have an incentive to migrate as migrant labourers, rather than as permanent migrants.

7. Conclusion

In this article I have presented a theoretical argument, as well as some evidence that the poor are constrained in their migration owing to the costs and risks of migration. This applies especially to those who live in more distant areas, far from transport and telecommunication links, with a more recent history of labour migration and thus weakly developed social networks. It also applies more particularly to the poorest, whose networks will be more weakly developed.

I have also argued that more empirical work needs to be done on this topic, using longitudinal data. Information based on data about lifetime migration, or migration in the recent past, can only provide indirect indicators of the influence of poverty on migration. It is also unsatisfactory to use proxies for income, such as level of education, because it becomes impossible to disaggregate the effects of education from that of income.

The question arises of what policy implications flow from the statement that the poor may be constrained in their migration. If it is indeed the case that the poorest do not stand much of a chance in the urban labour market because they are unskilled, as we saw above, it could be argued that, strictly from a migration-for-jobs perspective, it does not make that much difference if they are currently constrained from migrating or not. This is certainly the kind of argument that one can make based on Van der Berg et al.'s (2004) analysis (although it is not an argument that they make themselves). In this respect it is significant, however, that they do identify a residual group of low-skilled rural residents who would be able to improve their situation by migrating. It is also the case that constraints on migration help to make the labour market less able to respond to changes in the supply of labour. Because it reduces the supply of low-skilled labour at the workplace, this is one more factor leading to rigidities in the labour market (Dinkelman & Piroux, 2001). It is also the case that people migrate not only to be closer to employment but also to be closer to services, housing and transport networks, as Cross et al. (1999) have pointed out. Of course, if the labour market situation changes, with more jobs becoming available to the poor, questions about the capacity of the poor to migrate will assume much more urgency.

Whatever the policy implications, however, this paper has aimed to show that these issues are central to our understanding of the underlying dynamics of migration in South Africa.

Additional information

Notes on contributors

Derik Gelderblom

Associate Professor, Department of Sociology, University of South Africa (UNISA), Pretoria. A version of this article was presented at the Society in Focus seminar series of the Department of Sociology and Social Anthropology, University of Stellenbosch, 29 July 2005. The author wishes to thank participants at the seminar, particularly Simon Bekker, who also commented on an earlier draft of this article. The author expresses gratitude for the two anonymous reviewers' comments and suggestions.

Notes

¹I am indebted to Pieter Kok for this insight.

²‘The border of the Kruger Park with Mozambique was patrolled by the military to prevent infiltration of anti-apartheid forces until the early 1990s and was controlled by an electric fence that gave lethal jolts until the early 1980s when it was turned down.’ http://www.wits.ac.za/tpari/Tele-Seminars/Bushbr%20-%20Thornton.pdf Accessed 21 February 2007.