یادگیری اجتماعی، اثرات محله ای، و سرمایه گذاری در سرمایه انسانی: شواهدی از انقلاب سبز هند
|کد مقاله||سال انتشار||تعداد صفحات مقاله انگلیسی||ترجمه فارسی|
|18544||2007||26 صفحه PDF||سفارش دهید|
نسخه انگلیسی مقاله همین الان قابل دانلود است.
هزینه ترجمه مقاله بر اساس تعداد کلمات مقاله انگلیسی محاسبه می شود.
این مقاله تقریباً شامل 12080 کلمه می باشد.
هزینه ترجمه مقاله توسط مترجمان با تجربه، طبق جدول زیر محاسبه می شود:
|شرح||تعرفه ترجمه||زمان تحویل||جمع هزینه|
|ترجمه تخصصی - سرعت عادی||هر کلمه 90 تومان||17 روز بعد از پرداخت||1,087,200 تومان|
|ترجمه تخصصی - سرعت فوری||هر کلمه 180 تومان||9 روز بعد از پرداخت||2,174,400 تومان|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Journal of Development Economics, Volume 83, Issue 1, May 2007, Pages 37–62
This paper empirically identifies social learning and neighborhood effects in schooling investments in a new technology regime. Social learning implies that learning is most efficient when observed heterogeneity in schooling is greatest. The estimates of learning-investment rule, from farm household panel data at the onset of the Green Revolution in India, show that (i) agents learn about schooling returns from income realizations of their neighbors, and (ii) the speed of learning is high when the variation of schooling is large. Thus, schooling distribution of the parents' generation in a community has externalities to schooling investments in children. Simulations show that the variations in schooling within and across communities generate variations in child enrollment rate and average household income.
It has become increasingly recognized that technological changes affect returns to schooling in both developing and developed countries (e.g., Foster and Rosenzweig, 1996 and Juhn et al., 1993). To correctly infer new returns, however, agents face an informational problem. Since schooling investment is irreversible and also requires a long gestation period, agents cannot simply go to school to learn about schooling returns. Therefore, agents cannot rely on their own experience but must use observations from others to infer the returns. When agents learn from their neighbors, neighborhood factors influence the social learning. The neighborhood characterizes the environment in which agents learn from their neighbors. This paper examines neighborhood effects on social learning that determines schooling decisions, using household data available from the onset of the Green Revolution in India, where in some regions the diffusion of high-yielding varieties (HYVs) affected returns to schooling. The analysis shows that schooling distribution of the parents' generation in a village is important to social learning and household decisions on child schooling investments. The empirical finding that schooling decisions are correlated among neighbors can be viewed as the evidence of neighborhood effects, peer pressure, role models, norms of behavior, and social networks. The high correlation of similar decisions among neighbors has been found in many empirical studies (Case and Katz, 1991, Evans et al., 1993, Strauss and Thomas, 1995, Topa, 2001 and Conley et al., 1999). Moreover, the within-community correlations are also hypothesized to justify public subsidies for education in theoretical studies (Benabou, 1996).1 However, the process that generates the cross-sectional correlations of decision making has not been empirically identified until recently, except by Besley and Case (1994), Foster and Rosenzweig (1995), Munshi (2004), and Conley and Udry (2004).2 In this study, I attempt to empirically identify the process of social learning and neighborhood effects on child schooling investments in a Bayesian learning model. The question of whether agents know of and how fast they respond to return structures poses a more extensive but fundamental question into the way we think about economic development. For example, are observed variations of human-capital accumulation simply a consequence of different return-augmenting mechanisms in perfect information, as argued in endogenous growth theories (Lucas, 1988 and Romer, 1986)? Or are they a consequence of local environments that affect agents' learning speed under imperfect information? Even if returns are augmented, the latter would potentially generate substantial variations in investment. Though corresponding implications for development policy are different, it is not easy to identify these two cases by casual observations. Empirical findings regarding the above question are not conclusive. In his extensive survey on the rate of return to schooling investments, Psacharopoulos (1994) points out higher rates of return to private schooling investments in developing countries than in developed countries, especially from primary education. Child schooling investments are likely to be suboptimal in less-developed countries, although in most studies he surveys the sampling is not random and sometimes selective.3 The evidence on dynamic changes in enrollment rate is rare in the literature on developing economies. Foster and Rosenzweig (1996) and Rosenzweig (1990) are exceptions. They show evidence from India that private schooling investments have increased in 10 years in regions where technical change was rapid and therefore, they argue, schooling returns were augmented.4 Given the change in returns to schooling, however, it is not clear how precisely agents inferred the true returns immediately after returns changed and, if social learning was important, how agents learned about the returns and responded with investment behavior to altered environments. Among empirical tests for learning externalities, a few studies have explicitly incorporated sequential updating of agents' perception.5 In the literature, social learning was identified in the context of technology adoption in agriculture.67 To estimate the adoption rule of HYV with learning externalities, Besley and Case (1994) use a risk-neutral Bayesian framework in which agents infer the mean profitability of HYV from their neighbors. On the other hand, Foster and Rosenzweig (1995) adopt a modified target-input Bayesian model in which agents learn the best uses of inputs with the new technology and show that farmers are learning from both their own experiences and those of neighbors. In the target-input framework, Rosenzweig (1995) also shows that schooling hastens farmers' learning speed in HYV adoption. While the above studies assume that agents learn from others in their geographical cluster such as a village, Conley and Udry (2004) incorporate agents' networks explicitly in their empirical analysis, based on actual information flows in pineapple adoption behavior in Ghana. They show that it is not geographical proximity but rather information networks that significantly enhance social learning.8 Bandiera and Rasul (2005) also study the role of social network in social learning in the adoption of sunflower in Mozambique. The importance of reference group identification is emphasized by Manski, 1993a and Manski, 1993b in his seminal work. Among the literature, Munshi's (2004) study is related to this paper in an interesting way. Munshi attempts to identify the role of unobserved heterogeneity in determining the efficiency of social learning in the context of HYV adoption in rice and wheat productions. His results show that farmers can learn less from others when production is more sensitive to farm-specific idiosyncratic factors, that is, unobserved heterogeneity is important. One conclusion of my paper is that heterogeneity helps agents learn. While Munshi examines heterogeneity that is unobservable (and idiosyncratic) to agents, this paper examines observed heterogeneity. In an analogous way, error terms (i.e., unobserved heterogeneity) in econometric estimation deter precise parameter estimates, while variations in explanatory variables (i.e., observed heterogeneity) help estimate parameters precisely. The details will be described in Section 2. In this paper, I assume that households are attentive to the expected returns to schooling. In other words, it is assumed that agents are risk neutral. An alternative modeling strategy would be to use a target-input framework, as in Foster and Rosenzweig (1995). Target-input framework is suitable for identifying learning externalities if learning externalities affect, for example, input allocation decisions and therefore the actual profitability of investment in the context of HYV adoption. In the context of schooling investments in children, however, the informational spillovers from neighbors should only influence agents' perceptions on their future income gains — returns. The income gain from advancing to a higher level of education will be realized only in the future, after agents complete the education. The returns will be realized when agents accumulate their labor-market experience. Hence, learning about schooling returns does not lead to changes in profitability or income at the time the decision is made. In the framework of this paper, I therefore model social learning and investment behavior such that learning externalities change agents' perceptions of future income gain and that agents change their schooling decisions in response to changes in their perceptions but informational spillover has no effect on current incomes that directly influence the current welfare of agents. The following two points need careful attention. First, in any kind of test for externalities, it is important to exclude a possibility that observed cross-agent correlations of schooling decisions are spurious, say, driven by common unobservable factors. For example, variations in schooling investment can be attributed to unobserved heterogeneous local endowments and preference for education. Since unobserved factors are often correlated with observable factors, we can easily infer a spurious correlation between observables and schooling investments. Any empirical analysis must meet the challenge of identifying learning externalities against common unobservables. Second, both social learning and learning-by-doing lead to similar observable implications. The observation that schooling investments are positively correlated with an income gap between the educated and the uneducated does not necessarily imply social learning. Suppose, to find the best manager among household members, the household assigns each member in turn to the manager. In villages where schooling returns are increased, households eventually discover that the best farm manager is the most educated member. In this scenario, information from neighbors plays no role. To distinguish social learning from this within-family learning-by-doing, it is therefore imperative to examine not only the relationship between schooling return signals and schooling investments but also the process by which neighborhood factors affect social learning, with a theoretical framework to interpret empirical findings. In the next section, a theoretical framework is formulated to provide a basis for the following empirical analysis. Agents learn about schooling returns from income difference between the educated and the uneducated households. It is shown that agents' learning speed is influenced by neighborhood conditions such as income uncertainty and schooling distribution of the parents' generation. Section 3 describes the empirical strategy. Instead of tracing agents' learning process, I estimate schooling returns in farm profit function in each village. At the onset of the Green Revolution in India, HYVs were available to some regions, which caused changes in schooling returns in some sample villages. Section 4 describes farm household panel data from India, which I use in the empirical analysis. The Additional Rural Incomes Survey (ARIS) was conducted by the National Council of Applied Economic Research (NCAER) in three crop years, 1968–1969, 1969–1970, and 1970–1971, which correspond to the onset of the Green Revolution, when at least in some districts farmers experienced changes in schooling returns (see Rosenzweig, 1990). Section 5 summarizes empirical results. First, schooling investment is positively correlated with the income difference between educated and uneducated households. The finding is consistent with social learning. Second, schooling distribution of the parents' generation in a village influences the response of school enrollment to schooling return signals – that is, agents' learning speed – in a manner consistent with theoretical predictions on social learning. Heterogeneity in schooling increases learning speed. Consistently, the results also show that estimated uncertainty on schooling returns is minimized when observed heterogeneity in schooling of the parents' generation is maximized. Therefore, local schooling distribution of the parents' generation has intergenerational externalities to schooling investments in children. In Section 6, I simulate paths of enrollment rate and average household income, based on the estimated learning-schooling investment rule. Simulations show that school enrollment rate would increase by about 4.3 percentage points in 5 years if the proportion of educated households in a village increases from 0 to 0.65, for example. Since educated households have on average a higher income than uneducated households, a disparity of average household income would emerge over the 5 years. Thus, the initial distribution of schooling, which differs across communities, determines the evolution of income inequality over space. The analysis also has some aggregate implications: reallocating agents across communities can improve the aggregate response of schooling investments to a change in returns. This economy-wide implication is also quantified based on the estimated parameters. The final section summarizes the findings of this paper and discusses further implications.
نتیجه گیری انگلیسی
This paper shows that neighborhood factors matter in schooling investments, with evidence from farm household panel data from the Green Revolution in India. In the face of the HYV availability that altered schooling returns, agents learned of the benefits of new returns to schooling from neighbors and adjusted schooling investments over time. In this context, the empirical results clarify the importance of schooling distribution of the parents' generation within a community. Observed heterogeneity of schooling in the community increases informativeness on schooling returns, since agents easily compare income changes of differentially educated agents. The homogeneous community with small differences in schooling makes it hard to identify the effect of schooling on income. This intuitive prediction was supported in the empirical analysis of this paper. However, as discussed, identifying social learning is a challenging empirical issue. The situation where schooling returns had exogenously changed provides an ideal empirical setting for investigating the role of social learning in child schooling decisions. However, our conclusion crucially depends on the assumption that the initial prior on changes in schooling returns was uncorrelated with the schooling distribution of the parents' generation. Only if this assumption is satisfied, our empirical findings prove a unique proposition that learning is most efficient when observed heterogeneity in education is greatest. To increase learning efficiency in a society, should the educated and uneducated be integrated or segregated by communities? Our findings imply that the integration of the two heterogeneous populations in a community is more desirable than the segregation. Intuitively, given that a mixture of the two groups in a neighborhood enables the comparison between the groups – schooling returns, in this paper – all communities should have the educated and uneducated. This implication is against a common finding on positive sorting in residential choice behavior (e.g., Fernandez, 2001). If agents are sorted by their types, including education, in the choice of their residential areas, the population becomes more homogeneous in a community and weakens the response of schooling investments to a change in schooling returns. If social learning effects are not internalized in agents' location choice, the evidence of this paper justifies a socially desirable policy intervention. This implication should not only be relevant in the example of education but could be equally applicable to other issues such as social class formation and the division of labor. However, the relevance of the findings in this paper depends on the frequency of structural changes. As stated in Schultz (1975), if the benefit of education generates from situations of disequilibrium such as the Green Revolution, the augmented returns to schooling will eventually decrease as the knowledge of new technologies diffuses evenly and widely in the population. All these issues still remain unexplored and should be examined carefully in the context of developing countries.