Empirical studies of the effects of social capital

Following the econometric discussion, the literature on the effects of social capital may be divided into two types: individual and aggregate studies.

5.1. Individual-level studies

Individual-level studies of social capital may be divided into studies that focus on de­veloping societies and studies that focus on OECD societies.

A typical social capital study in this literature posits an individual outcome of the form

where, following previous notation, X_i denotes a set of individual controls, Y_g(_i) de­notes a set of group controls and SC_g(_i) denotes social capital. As such, Equation (11) corresponds to the case of exogenous social capital discussed in Section 3. Evidence for the relevance of social capital is equated with the statistical significance of the co­efficient J. In the various tables we have constructed to summarize various empirical papers, we report dependent variables and social capital measures, as well as findings based on the statistical significance standard.

5.1.1. Social capital and development

Links between social capital and development have been examined in a range of con­texts. One reason for this is that the failure of many developing economies to achieve sustained growth has led social scientists to look for previously unexplored factors in the development process.

As the table indicates, a range of alternative outcomes have been studied. Simi­larly, a range of social capital measures have been employed. While these studies are quite disparate, there are some commonalities. First, these development studies typi­cally focus on measures describing the social networks in which individuals participate. Fafchamps and Lund (2003), Fafchamps and Minten (2001, 2002), Grootaert (2000), Isham (2002) and Narayan and Pritchett (1999) all give primary focus to the role of memberships in various organization and trading networks as determinants of economic outcomes. The quite different social capital measures used by Lee and Brinton (1996) and Palloni et al. (2001) reflect the different outcomes they are measuring (immigration and placement in elite firms.) Further, the studies in Table 1 give primary focus to par­ticipation in organizations that can provide economic benefits in terms of information sharing and the production of collective goods. In this sense, these studies focus on eco­nomic benefits to organizations as opposed to more tangible psychological and social benefits.

From the perspective of the discussion of identification in Section 3, several ques­tions arise. First, how does one differentiate social capital effects from the presence of other group effects such as information spillovers, or the presence of common factors such as legal or political institutions? In the papers discussed here, relatively little at­tention has been paid to this question. Notice that the failure to consider this issue is not necessarily a damning criticism, in the sense that one may have reasons to rule out such effects in advance. However, these studies also typically fail to make good argu­ments that alternative social determinants of outcomes can be ignored. This strikes us as a more serious indictment in that social capital variables can easily proxy for such factors.

In our judgment, the more successful studies of social capital and development are those that have focused on specific phenomena that have been placed under the social capital rubric. Unsurprisingly, Fafchamps and Minten (2002) is in our view a good ex­ample of this approach. As indicated in the paper’s title, the focus of the analysis is less on social capital per se than on the role of social networks in affecting trader prof­itability. This paper focuses on agricultural traders in Madagascar. These traders are intermediaries between farmers and various markets in the country. Because the goods they sell (staples such as rice, potatoes and beans) are well defined (the basic goods are homogeneous and are distinguishable by observable features such as whether they have been milled or converted to flour, etc.), it is relatively easy to measure the value added associated with a trader’s activity. Fafchamps and Minten (2002) find that measures of the size of an individual trader’s business network are positively associated with value added and total sales. The paper argues that a relationship between networks and these economic outcomes may be understood in the context of models of imperfect infor­mation and monitoring, which provides a clear theoretical motivation for the empirical framework as well as a plausible theoretical interpretation for the various findings.

Table 1

Individual-level studies of social capital in developing countries

1674 S.N.

Ch. 26: SocialCapital 1675

Finally, it should be noted that while the different studies in Table 1 consistently support a role for social capital in facilitating various economic outcomes, two of the studies, Krishna (2001) and Varughese and Ostrom (2001), argue that there are impor­tant subtleties in this relationship that need to be accounted for. Krishna (2001) finds that for villages in Rajastan India, the relationship between conventional social capital measures and outcomes such as common land development and poverty reduction is sensitive to a notion of effective governance Krishna calls ‘capable agency’.

5.1.2. Social capital in OECD societies

Just as social capital has been used to explain a range of outcomes in developing economies, so it has been used to explain a range of US phenomena. Table 2 reports a number of such studies.

In comparing Tables 1 and 2, a number of differences may be identified. First, social capital studies for affluent societies are far more heterogeneous than those which we report for developing economies. One finds studies of social capital for the United States that explore outcomes ranging from mental health [Furstenberg and Hughes (1995)] to dropping out of high school [Teachman, Paasch and Carver (1997)] to criminal activity [Hagan and McCarthy (1995)]. We do not believe this reflects differences in our choices of what studies to report. Rather, interest in social capital in advanced societies has been motivated by different phenomena than in the case of developing economies. In particular, the focus on social capital appears to be motivated by a desire to understand how some individuals avoid self-harming behaviors of various types.

Second, social capital studies for affluent societies focus on somewhat different vari­ables to proxy for social capital than their development counterparts. This may be seen in the frequent examination of parental influences in Table 2. A common assumption in studies for the US is that the parent, child, neighborhood and school relationships are a primary form of social capital. McNeal (1999), for example, explicitly argues that par- ent/child interactions closely correspond to what Coleman originally meant by social capital.

Another feature that distinguishes the literature on OECD societies is its focus on traditionally sociological concepts in construing social capital. One important notion is intergenerational closure, which holds when parents of a given child know both his friends as well as his friends’ parents; both Morgan and Sorenson (1999a) and Sandefur, Meier and Hernandez (1999) treat closure as an important aspect of social capital. This variable arises because, as argued originally in Coleman (1988), control and monitoring of children is sensitive to the ways that a family is embedded in a community.

While OECD social capital studies typically are based on richer data sets than those available for developing countries, these studies often suffer from serious flaws. One problem is that little discipline has been imposed on the empirical proxies used for so­cial capital, which makes many of the empirical claims in this literature incredible. For example, authors such as Furstenberg and Hughes (1995), McNeal (1999) and Sandefur, Meier and Hernandez (1999) treat the number of family moves as a measure of social capital for youths. The idea is that the more a family moves, the weaker the social ties between the youth and his community. This is certainly a plausible claim. However, it does not suffice to make family moves a valid social capital measure. Since moves are endogenous, the variable in essence provides an indictor for those characteristics that determine the moves. Such characteristics can be associated with different youth outcomes for reasons that have nothing to do with social capital. For example, families who make more moves plausibly contain parents who are less interested in their chil­dren than those who make fewer, since such parents may be putting less weight on the costs to children of changing neighborhoods. Parents with less interest in their children [which can be formalized by using Loury’s (1981) model of intergenerational mobility and allowing for heterogeneity in the rates at which parents discount offspring utility] will presumably invest less in their children, altering their outcomes in ways similar to the purported effects of lower social capital. Our point is not that one explanation or the other is correct, but rather that neither is identified from the data. Put differently, there are good reasons to believe that there are systematic differences in the unexplained com­ponents of individual behavior that render standard estimation methods inconsistent; specifically, families asserted to posses high levels of social capital, from the perspec­tive of the estimated model, may be expected to be associated with higher levels of parental interest in children, which means the residuals in the associated regressions no longer have conditional expectations of 0. As such, this discussion is an illustration of an exchangeability violation of the type discussed in Section 3; Furstenberg and Hughes (1995) are especially susceptible to this criticism due to the lack of attention to control variables.

Similarly, little attention is typically given to the identification problem of distin­guishing social capital from endogenous or other group effects. This failure derives from the flexibility of the social capital definitions that are employed. Is a psychologi­cal propensity to behave similarly to one’s peers a form of social capital? The answer to this question is unclear from the literature, since such a propensity could easily count as a type of social norm.

While none of the studies in Table 2 can be said to fully address these general statis­tical questions, some of the studies are nevertheless clearly valuable contributions. One paper we would identify is Morgan and Sorenson (1999a). This paper is noteworthy for

Table 2

Individual-level studies of social capital: OECD countries

1678 S.N. Durlauf andM. Fafchamps

Ch. 26: SocialCapital 1679

its careful attention to different causal mechanisms by which social capital may matter and by the care with which empirical proxies are constructed. We would also note that the paper focuses on a very specific issue, namely why Catholic schools appear to out­perform their public counterparts, where there are good prior reasons to believe social factors matter.^[421] Palloni et al. (2001) is in many ways a very different study, yet is also very admirable. This analysis focuses on a very simple notion of social capital, in study­ing the effect on an individual’s migration decision of prior migration by a sibling. What commends this study is the immense care taken to deal with questions of unobserved heterogeneity and common factors between siblings unrelated to social capital.

Before leaving this section, we draw attention to Costa and Kahn (2003b), which provides an historical perspective on social capital. In this paper, the behavior of union soldiers in the Civil War is examined, with particular attention to rates of promotion and desertion across different companies of soldiers. Costa and Kahn find that ethnic and occupational homogeneity of companies was conducive to braver conduct by soldiers. While far removed from the types of behaviors that are usually studied using social capital, the behavior of soldiers is in fact an excellent phenomenon to examine, given the well documented role of social factors in battlefield conduct.^[422] We believe creative exploration of data sets like this can add a great deal to the understanding of social capital.

4.4. Aggregate studies

At the beginning of Section 3, we outlined the difficulty of estimating the beneficial effects of social capital from individual data. We now turn to empirical studies that rely on aggregate data and examine whether they provide more convincing evidence of social capital. Table 3 reports a number of social capital studies that employ such data. As the table indicates, a large number of aggregate level social capital studies have focused on the relationship between social capital and per capita output growth at a high level of aggregation, such as a country or region. As such, most of the studies of this type are variants on empirical growth regressions that have become a workhorse of modern growth economics.^[423] An assessment of the aggregate studies using social capital is therefore essentially equivalent to an assessment of a set of growth regressions designed to establish that a particular variable is causally related to growth.

Growth regressions of the type found in the studies of Table 3 have been subjected to very serious methodological criticisms; examples include Brock and Durlauf (2001b), Durlauf (2000), Durlauf and Quah (1999) and Temple (2000). As argued in these papers, growth regressions suffer from several fundamental problems that make implausible the types of causal inferences one typically finds in the empirical literature. First, there is the problem of the choice of control variables. Growth theories are open-ended, which means that one growth theory does not have any logical implications for the truth or falsity of another. Hence, there is no natural way, when one wishes to test the importance of a given theory, to identify the appropriate set of theories to incorporate in a correctly specified structural growth model. As Durlauf and Quah (1999) indicate, there are in fact more extant growth theories than there are countries to which they are supposed to apply. As a result, any given growth regression may be subjected to the criticism that relevant control variables have been omitted. While there are some possible ways to deal with this problem, see Fernandez, Ley and Steel (2001), this problem has not been addressed in any social capital and growth studies, as far as we know.

Second, growth regressions typically fail to account properly for parameter hetero­geneity across countries. Evidence of such heterogeneity may be found in Desdoigts (1999), Durlauf and Johnson (1995) and Durlauf, Kourtellos and Minkin (2001); the­oretical models that imply heterogeneous growth processes for different groups of countries include Azariadis and Drazen (1990) and Howitt and Mayer-Foulkes (2002). Failure to account for parameter heterogeneity calls into question the structural interpre­tation of a social capital variable as it may be proxying for this form of heterogeneity. One example that is suggestive of this possibility concerns the role of ethnic hetero­geneity in growth, a question studied by Easterly and Levine (1997).^[424] In this paper, the authors argue that ethnic conflict inhibits public good creation and so acts as an im­pediment to growth. Ethnic conflict is instrumented with a measure of ethnolinguistic diversity which proves to be strongly negatively associated with growth. Since Sub­Saharan Africa has exceptionally high levels of ethnolinguistic diversity, the authors conclude that this is an important mechanism in understanding Africa’s growth prob­lems. Brock and Durlauf (2001a) reexamine this study, allowing for various types of exchangeability violations due to parameter heterogeneity, and find that the relation­ship between ethnolinguistic diversity and growth appears only for Sub-Saharan Africa; this variable does not help explain growth patterns in the rest of the world. Brock and Durlauf’s finding illustrates how growth explanations may well not be constant across countries. And for the African case, it is unclear whether the growth findings are causal or whether ethnolinguistic diversity simply proxies for some other form of ‘African exceptionalism’.

Table 3

Aggregate-level studies of social capital

1682 S.N. Durlauf andM. Fafchamps

Ch. 26: SocialCapital 1683

Taken as a whole, these arguments imply that the social capital/growth studies do not meet the exchangeability requirements that we discussed in Section 3. While this reflects more general failings of the empirical growth literature [Brock and Durlauf (2001b)], it is also the case that growth studies using social capital have been quite insensitive to efforts in the growth literature to address these problems.

Beyond questions concerning the comparability of observations, there are unresolved issues concerning causal interpretation of growth regressions that apply to the social capital case. This is especially important given the endogeneity of aggregate measures of social capital. We are unaware of any social capital study using aggregate data that addresses causality versus correlation for social capital and growth in a persuasive way. While this is a broad brush with which to tar this empirical literature, we believe it is valid. A related problem is that we are unaware of any compelling instrumental variables for social capital in these regressions. This failure is a corollary of the absence of any strong theories of aggregate social capital determination in the social science literature that would allow one to characterize appropriate instruments.

When one turns from national-level growth studies to other aggregate studies, the plausibility of claims concerning social capital becomes stronger in some cases. Guiso, Sapienza and Zingales (2004b) find evidence that trust helps explains trading and in­vestment patterns between countries. An interesting feature of their analysis is that the correlation between levels of trade and trust cannot be explained by measurable factors such as quality of legal systems. A recent study by Goldin and Katz (1999) is partic­ularly interesting in its focus on the sources for the rise of high school attendance in Iowa in the early part of the twentieth century. By focusing on characteristics of Iowa counties, they are able to avoid some of the clear problems of exchangeability that plague studies using coarser levels of aggregation. But even here, other problems arise: more important, the data available are quite weak in the sense that the variables which suggest the presence of social capital effects could equally well suggest alternative ex­planations. The specific variables that seem most suggestive of social capital effects are the percentage of native born citizens and the population of towns; high percentages of native born and low population sizes are each associated with higher high school atten­dance. Clearly, linking these correlations to a causal role for social capital or other type of social influence is speculative. To be fair, Goldin and Katz (1999) point out that there may be alternative explanations, such as the smaller towns having fewer opportunities for those without high school educations.^[425]

Overall, we conclude that aggregate social capital studies have not been successful in providing compelling empirical evidence on the effects of social capital. These studies require identifying assumptions that are incredible by conventional social science rea­soning. We believe that research efforts should be directed towards micro-level studies as the problems with country-wide studies seem too intractable to overcome. Data at lower levels of aggregation, such as county data for a homogeneous place like 1915 Iowa, are likely to be more amenable to persuasive analysis, provided the issues of ex­changeability and identification can be addressed adequately.

6.