Education is widely seen as a key determinant of social mobility within a person’s lifetime and across generations. For the first time, we now have high-quality data available that allows us to study social mobility in developing countries more rigorously. The 2019 RISE Annual Conference brought together four papers in its first session, doing just that. Three sets of findings emerge from the presented and other work in this area: 

1. Schooling matters for social mobility within a lifetime and across recent generations, but possibly less than often thought. Causal evidence remains scarce, but seems to confirm these findings. Educational mobility is much higher among previous generations, who seem to have reaped the fruits of early educational expansion, when education levels were much lower.
2. Variation in social mobility is very large—both geographically and across social groups. Locality is a key covariate and suggests (at least) two important channels: quality of schooling (on the supply side) and aspirations (on the demand side).
3. For policymaking a prudent lesson stands out: Targeted interventions do not necessarily work better than non-targeted ones, even though in some contexts affirmative action has been shown to work. 

Let’s go through these findings in more detail. 

Mapping social mobility

The first step in this active field of research consists of mapping how schooling during childhood is related to social background and earnings later in life. Existing work typically looks at completed years of schooling. Abhijeet Singh presented unique new findings from across five countries (Ethiopia, India, Pakistan, Peru, Vietnam) looking at learning—rather than years of schooling. The good news is that they find those who have learned more at the end of primary school to be more likely to progress to tertiary education, a strong predictor of high earnings later in life (while controlling for socio-economic status). The magnitude of the relationship is substantial: closing the learning gap between the bottom and top socio-economic quantiles at age 12 would narrow the gap in completed years of schooling at age 22 by between 25 percent and 50 percent. While this is unfulfilling from an equality of opportunity point of view, it is large from a policy perspective—it is hard to think of policies with a bigger impact. Further analysis indicates that the gaps in learning are evident early on and remain mostly unchanged between ages 8 and 15, confirming previous evidence for earlier life periods in other settings (see Figure 1 in Heckman, [2007]).

Other recent work confirms constraints to social mobility, this time across generations. Due to the lack of income data, social mobility across generations is often measured in terms of years of schooling itself (so social mobility is proxied by educational mobility across generations). A detailed study for India that allows cross-group analysis finds that although overall upward mobility remains relatively unchanged for those at the bottom, there is substantial geographic and social variation within this group: while Scheduled Castes and Tribes maintain high upward mobility, Muslims are now the least upwardly mobile (Asher et al., 2019). 

Worldwide, persistence in years of schooling across generations (which measures the lack of mobility) is also found to be high, with those in the bottom half of parental education having low chances of rising to the top quartile (Narajan and van der Weide, 2019). Persistence is found to be typically higher (so mobility lower) in low-income countries (and the gap with high-income countries seems to be rising), but some areas in parts of East Asia, Latin America and the Middle East have lower persistence; and girls are generally found to be catching up with boys. Because these studies concentrate on educational mobility across generations, they may be over- or underestimating income mobility, but it is the best we have for now. 

Does education cause social mobility in developing countries?

A next step is to assess the causal relationship. One popular approach is to study the impact of an (exogenous) change in the supply of schooling. Marieke Kleemans and co-authors find that men who benefited from Indonesia’s large school building programme in the 1970s, apart from having completed considerably higher years of schooling (0.27 for men and 0.23 for women), are more likely forty-three years later to be working in the formal sector, working outside agriculture, and to have migrated. Women are also more likely to have migrated, have fewer children, and be in a household with improved living standards. These effect sizes remain small, below 5 percentage points, with impact on women’s fertility among the largest, (confirming other recent evidence, see for instance Oye, Pritchett & Sandefur [2016] and Koppensteiner & Matheson [2019] for an in-depth study of Brasil). Benefits also carry over to the next generation, who enjoy more education (0.10 and 0.17 years for men and women respectively), in particular, secondary and tertiary education. The effect of a mother’s education is consistently and significantly larger than that of a father’s education. A key mediating factor is the characteristics of the marriage partner, in particular, spouses with more education and improved labour market outcomes  

What does intergenerational mobility look like when going back further in time? Leonard Wantchekon assesses intergenerational mobility across three generations in Benin. Exploiting the haphazard choice both of the venue where missionary posts were settled, and of the children who were taught in these schools, he compares (a) outcomes of descendants of those who had schooling and (b) companions of the same cohort who did not get but were exposed to schooling by living in a village with a school, with (c) those living in similar villages without schools (control group) to estimate direct (a-c) and indirect (b-c) treatment effects. He finds important income mobility for both the 2nd to the 3rd generation, especially for those who were exposed to schooling but did not attend school themselves (indirect effect). The effects are largest for those of lower-income categories. For the third generation, the effect is entirely at the village level, suggesting important externalities. Those exposed to (but not having attended) schooling are found to have higher risk aversion and improved self-reliance, work ethic, and life outlook. This is consistent with aspirations as a key mechanism for social mobility (see below).

These results also corroborate recent evidence from 26 countries across Africa observing substantial variation in upwards educational mobility both across and within countries, and noting an important role for geographical exposure (Alesina et al., 2019). 

It is interesting to compare these results with those from a recent US study, which also finds high educational mobility among previous generations, in this case, parents born between 1880 and 1910, and their children, born in the 1920s (<Card et al., 2018). What these studies have in common is that they look at an era of substantial expansion of primary education when levels of schooling were very low. This higher mobility early on raises the question whether a ‘golden age of upward mobility’ is mostly behind us now, in the US or elsewhere, and if so why? A likely answer lies in the relative demand for educated workers, which even if low, must have far outstripped supply in early days, thereby securing a well-paid job to these early generations who went to school. However, there is an important qualification: while the roll-out of education worldwide has been impressive, it has also been uneven. Schooling is lower among some social groups. And while enrolment rates have been high, the quality of schooling varies widely. Large numbers of children attend school but with little learning. The study by Card et al. also confirms the role of education quality for historical upward intergenerational mobility in the US. Similar evidence for developing countries is not yet available as far as we know. 

Improving upward social mobility

What is the best way to improve upward social mobility? Should interventions target specific groups or are general, non-targeted, educational improvements equally effective? David Evans and Fei Yuan, in their presentation, assessed which programmes have been most effective at reducing education and learning gaps, focusing on experimental and quasi-experimental interventions that included girls. They find general interventions to be just as effective as girl-targeted interventions at improving outcomes for girls. The best way to ameliorate girls’ access to school turns out to be cutting the cost of schooling, while the preferred way to advance girls’ learning is through improvements in pedagogy for all pupils.

This is a compelling finding. Nevertheless, it is unlikely to be a general principle. Recent work on India finds that affirmative action can have strong positive effects on enrollment outcomes for the targeted group, in this case, lower caste (Bagde, 2016). This fits with the observation that virtually all gains in upward mobility in recent times accrued to Scheduled Castes and Tribes, who have benefited from constitutional protection, reservation in education and politics, and targeted policies.  Meanwhile, the upward mobility of Muslims, who enjoy very little targeted policies, has declined (Asher et al., 2019). At the same time, affirmative action comes with challenges and pitfalls. Earlier work highlights potential undesirable consequences for non-targeted groups, for instance, reducing the number of female applicants in the presence of caste-based targeting (Bertrand et al., 2009).  

Ways forward for research on education and social mobility 

This blog started by summarising three big messages. What do they imply for future work? Improved data will undoubtedly allow better mapping of social mobility. This, in turn, will further highlight the central question: What explains the large variation in social mobility? Why do some areas, groups, individuals, or periods show strong upward mobility—within their life or across generations—while others do not? What are the moderating factors and what are the mediating ones? Existing work underlines the central role of locality (neighbourhoods, villages, regions), both for social mobility within one’s life and across generations. What do these neighbourhood effects mean? Two key factors present themselves. 

Quality of schooling is expected to play an important role on the supply side. Most work focuses on years of education (but not on learning) underestimating the variation in skills and knowledge. The study by Singh and co-authors looks at learning outcomes across five countries and finds that learning at age 12 helps explain advancement to tertiary education substantially, albeit not fully (and it does not control for locality). This confirms the expected importance of quality of schooling; further work is needed to deepen our understanding of this relationship.

On the demand side, several factors may play a role, including credit and information constraints, as shown by previous studies (see for instance Jenssen, Attanasio & Kaufmann). Recent work also underlines the role of people’s internal constraints, like attitudes, hope, and aspirations. Anthropologists argue that the poor may lack the capacity to aspire in order to overcome the conditions of their own poverty (see Appadurai, 2004). Aspirations are socially embedded: they are the product of what people see around them. Recent work in the US provides a fascinating illustration. Focusing on innovators, a study by Bell, Chetty and co-authors finds that children of the same mathematical ability (which is a strong predictor of patented inventors later in life) are much more likely to invent in a type of technology to which they have been exposed to in their childhood neighbourhood. Other recent work looks at the effect of role models and local leaders. In my own work, analysing data for India, we find a strong and meaningful relationship between mother’s educational aspirations for her child at age 12 and completed years of schooling, learning, and labour outcomes at ages 15 and 18. Work on Ethiopia demonstrates that aspirations can be moulded. An intervention that shows films with life stories of poor people who managed to improve their life through their own effort is found to alter aspirations and investment in education both six months and five years later, and has considerable spill-overs. Recent conceptual work on hope provides an attractive framework for a structured way of thinking about internal constraints. 

The increased availability of data on learning, combined with the renewed interest in long term panel data, promises exciting further work investigating these mechanisms. 

Many thanks to Lant Pritchett for comment on an earlier draft of this blog.