شبکه های اجتماعی، اشتغال و دلسردی کارگر : مدارک و شواهد از آفریقای جنوبی

Social networks are increasingly being recognized as having an important influence on labour market outcomes, since they facilitate the exchange of job related information. Access to information about job opportunities as well as perceptions about the buoyancy of the labour market depend critically on the social structures and the social networks to which labour market participants belong. In this paper, we examine the impact of information externalities generated through network membership on labour market status. Using Census data from South Africa, a country characterized by high levels of unemployment and worker discouragement, we adopt an econometric approach that aims to minimise the problems of omitted variable bias that have plagued many previous studies of the impact of social networks. Our results suggest that social networks may enhance employment probabilities by an additional 3–12%, and that failure to adequately control for omitted variables would lead to substantial over-estimates of the network co-efficient. In contrast, the impact of social networks on reducing worker discouragement is much smaller, at between 1 and 2%.

Networks matter for economic and social outcomes, since they allow for complex social interactions to take place between individuals, thereby facilitating information spillovers and learning between network members, as well as the transmission of norms and values (Banerjee, 1992 and Bikhchandani et al., 1992). While information spillovers have been widely examined by economists, as affecting fertility decisions, education decisions (Coleman et al., 1966), participation in criminal activities, (Besley and Coate, 1992, Borjas, 1995 and Case and Katz, 1991), and consumption (Abel, 1990), in this paper we focus primarily on information externalities generated through network membership as they pertain to employment prospects. Arguably, it is now a stylized fact that individuals rely on friends, family and acquaintances (or “weak ties” in the parlance of Granovetter, 19741) to find jobs, and while reliance on these social networks may vary by location and demographic characteristics2, these channels are usually productive (Calvo-Armengol and Jackson, 2004, Blau and Robbins, 1992 and Montgomery, 1991). Moreover, reliance on networks to find employment typically results in better matching between the job candidate and the available job, reduces employer uncertainty about worker productivity, and may even enhance job satisfaction and employee loyalty (Datcher, 1983, Devine and Kiefer, 1991 and Marsden and Gorman, 2001). Obviously, this is conditional on the “quality” of one's network. When an individual's contacts are unemployed, the likelihood of getting information about jobs through contacts is reduced, thereby impeding such network benefits, reducing active search behaviour (Devine and Kiefer, 1991 and Kingdon and Knight, 2001) and increasing the duration of unemployment. Thus, heterogeneity in network effects is important since it can explain changes in wages and employment inequality over time. Arrow and Borzekowski (2004) show that networks explain 15% of the unexplained variation in wages and a substantial portion of the disparity between black and white income distributions. This is exacerbated if one allows for the fact that social networks are largely endogenous or self-selected. Within the context of very high levels of unemployment and worker discouragement in South Africa (see Table 1) combined with relatively low absorption of the unemployed into the informal sector (Kingdon and Knight, 2001), understanding the effect of networks on employment prospects is arguably critical, especially since job search methods in South Africa are predominantly passive, with most jobs being obtained through word-of-mouth and other informal recruitment methods (Kingdon and Knight, 2001). Using the Special Retrospective Survey of Employment and Unemployment (SRS) dataset, Statistics South Africa (1998) reports that just under 20% of individuals cite high costs as the reason they do not engage in active job search. This decision not to search owing to high costs may be exacerbated for individuals living in areas characterised by high levels of worker discouragement. Such individuals may believe that there is little chance of finding work, even if they spend the necessary money to reach the job market, since this is the experience of others in their community, and so, may rationally decide to not actively search as the opportunity cost of the transport is too high. At the same time, large numbers of firms rely on informal recruitment methods. Standing et al. (1996) use firm level data from South Africa to show that 41.4% of all firms studied relied on friends and relatives of their existing employees to recruit new workers. This reliance on informal methods may be attributable to the poor signaling content of the school-leaving exam qualification, given the continuing discrepancies in educational quality, as well as to the high costs associated with formal recruitment procedures, especially for relatively low skilled jobs.Yet, until relatively recently, the South African literature on job search and employment has focused on individual job search and work choices in isolation, ignoring the potential influence that social networks, constituted by families, peers and acquaintances, might have on these individual decisions. Networks facilitate the exchange of job related information (Ioannides and Loury, 2004), be it about actual job openings or about the paucity of such opportunities. Thus, access to information about job opportunities as well as one's perceptions about the buoyancy of the labour market may depend critically on social structures and the social networks to which individuals belong. The work that has been done examining the impact of social networks on employment status in South Africa has mostly defined the network as the number of other household members who are employed. (see Wittenberg and Pearce, 1996, Mlatsheni and Rospabe, 1999 and Schoer and Leibbrandt, 2006). Wittenberg and Pearce (1996) find that in South Africa, these networks positively influence access to jobs, while Mlatsheni and Rospabe (1999) find that they significantly increase the probability of youth being in wage employment. Wittenberg's (1999) non-parametric analysis is also suggestive of a household network effect as it indicates clustering of employed and unemployed individuals3. Using a non-representative survey for a specific magisterial district located near Cape Town, Schoer and Leibbrandt (2006) demonstrate that unemployed individuals who report that they have contacts in the labour market are significantly more likely to rely on social networks in searching for a job, and that the difference between relying exclusively on social networks as opposed to more active search methods depends crucially on contact availability. Moreover, they also show that unemployed individuals who report that they rely on social networks to search for jobs are more likely to live in households where at least one household member is employed compared to unemployed individuals who rely on more active search methods, such as place-to-place searches or relying on newspapers. However, all of these studies are plagued by omitted variable bias (which we discuss in more detail below). Specifically, it is difficult to disentangle whether unemployed individuals relying on social networks to find jobs are simply more likely to live with employed individuals who support them while they engage in search behaviour, or whether the presence of an employed individual in the household lowers job search costs for the unemployed and improves access to job market information. Schoer and Leibbrandt (2006) argue that their data suggest that household employment provides access to financial resources which unemployed household members use to pursue active search methods, but that when unemployed individuals are constrained in their ability to engage in active search due to domestic duties, other employed individuals in the household then act as transmitters of information. Recognising that most studies of network effects suffer from omitted variable bias, Bertrand et al. (2000) devise an estimation strategy to try to minimise these biases in their work examining the impact of social networks on the take up of welfare grants in the USA. They use language groups to proxy for social networks, arguing that information flows more easily between those who speak the same language, and include language fixed effects as well as geographic area fixed effects to minimise problems arising from omitted variable bias. They find that networks may increase the responsiveness of welfare use to policy shocks by an additional 15–27%. Our empirical work draws directly on their approach, but we differ in that we use age–language4 cohorts to define the social network, and we focus on the impact of social networks on labour market status. We include age–language cohort fixed effects, as well as geographic area fixed effects in order to minimise potential biases that might arise from omitted variable bias. Our network estimates, while small are not insubstantial, and suggest that social networks alone may enhance employment probabilities by an additional 3–12%. In contrast, the impact of social networks on reducing worker discouragement is much smaller, at between 1 and 2%.

In this paper, we have presented estimates of the magnitude of social networks on employment probabilities and worker discouragement, controlling for as many sources of omitted variable bias as possible. Our results illustrate the importance of properly controlling for omitted variable bias in this type of work, since failure to do so clearly results in an overstatement of the magnitude of the network effects. Even after controlling for various fixed effects and controls for personal and household characteristics, our network co-efficients remain significant, although they have a substantially smaller impact on reducing the incidence of worker discouragement amongst the unemployed than they do on affecting employment probabilities. Indeed, our results suggest to us that there is a lot about worker discouragement that remains unexplained through the usual socio-economic analysis, and may point to the need for a more psychologically grounded behavioural analysis in order to explain this phenomenon more fully. Secondly, while it is clear that informational externalities provided through other employed adult household members are important for enhanced job prospects, information and job opportunities acquired through this means accounts for only a small proportion of the measured social network effect, suggesting an equally important role for “weaker ties” within the social network. However, at the same time, our results also suggest that informational externalities provided through other employed adult household members do not affect the incidence of discouragement among unemployed individuals. This is a phenomenon worthy of further investigation. Our results confirm that the importance of social networks in terms of employment probabilities and in affecting worker discouragement varies demographically. Social networks significantly enhance employment prospects for women, for individuals who did not relocate in the 1996–2001 period, and for those with access to telephones. In contrast, social networks appear to significantly reduce worker discouragement amongst individuals who did relocate in the 1996–2001 period, and those with access to telephones. However, it is important to note that the absence of a significant network effect for specific groups of individuals in our results may reflect lower usage of networks by the specified group to find jobs or to gather information about jobs, but it could also reflect differences in the likelihood of the individual in the specific group being made a job offer for jobs that one hears of through word-of-mouth. It is also the case that employer characteristics will determine to some extent the context in which different types of job search methods operate. In some cases, worker characteristics may be easily observable, while in other cases, recommendations from existing employees may be especially important or valuable. The strength of network effects will vary in these circumstances. In addition, job seekers may vary in terms of their contact availability and in terms of the quality of their networks. In other words, the impact of the social network on individual outcomes depends not only on the size and quality of the network, but also on whether the individual is able to access it effectively. For example, migrants may have greater difficulty accessing the social network since they are relative newcomers to the area, and it is perhaps unsurprising that social networks have smaller effects on the employment probabilities of those individuals who had relocated. However, social networks significantly reduce worker discouragement for those who did relocate. Since it is plausible that these individuals may have relocated to look for work, it is unsurprising that they are significantly more likely to be engaged in active job search. This suggests to us that individuals who move away from their own age–language cohorts may well be qualitatively different from those who stay behind, and again underscores the importance of trying to control for this source of bias in network studies. In this paper, we do not identify the specific ways in which the effectiveness of social networks depend on differences in the characteristics of job seekers, in the characteristics of the contacts they use, or the relationship between job seekers and their contacts and features of the job environment. We simply document the existence of network effects for various groups. Thus, a large number of questions remain unanswered, and in the absence of more detailed survey information of job search and recruitment methods, may remain so. Providing convincing explanations for the difference in magnitude of network effects for different groups is an example of one such question. For example, is it that groups that rely more heavily on social networks to access employment opportunities or gather job information do so because they have skills which are difficult to observe, with the result that employers rely more heavily on social networks in their hiring decisions, or is it that these groups face explicit labour market discrimination in formal hiring processes and thus, have to rely on networks to access employment. Or is it that some groups, for example women, gain more from their social networks than men because their contacts have better information about job opportunities and do a better job of passing this information on. Moreover, while our statistics describing the sample characteristics suggest that some age–language cohorts may find it more difficult to access opportunities in the labour market than others, an unanswered question is whether ethno-linguistic job niches exist in the South African labour market, in much the same way as documented by Waldinger and Bailey (1991) for the USA. To the extent that such ethno-linguistic niches exist, this will reinforce the importance of social ties in job search and employment prospects, but may also serve to inhibit social mobility. Arguably, understanding these kinds of dynamics, which will hinge critically on collecting better data on social networks, will contribute substantially to a better understanding of the complexities inherent in the South African labour market.