Social capital and growth: causal evidence from Italian municipalities

Social capital and growth: causal evidence from Italian municipalities Abstract The macroeconomic effects of social capital are typically studied using data at country, region or, at most, province level of aggregation. However, social capital is defined by connections among agents who know each other and its effects, if any, should be detected at a more detailed level of spatial aggregation. To the best of our knowledge, this is the first study using longitudinal municipality-level data to investigate the causal link between social capital and growth. We extend earlier research by accounting for the endogeneity of all the covariates as well as unobserved heterogeneity. The evidence suggests that social capital has been a source of growth inequality in Italy between 1951 and 2001. The causal effect of social capital on growth is positive, on average, and stronger in the Centre-North of Italy. In addition, it was higher in the 1950s. The paper also presents local estimates of the growth return to social capital, which are of interest for specific sub-populations of municipalities. 1. Introduction The role of institutions in shaping economic performance has been emphasised at least since North (1991), if not earlier (Hayek, 1949; Myrdal, 1957). Institutions are seen as humanly devised constraints, which affect political, social and economic interaction. These constraints can be both formal (constitutions, laws, property rights) and informal. Customs, traditions and codes of conduct belong to the latter category. Communities may well be different for their customs, and these differences may well explain their relative stages of economic development. For example, in societies where individuals are used to trust each other, trade is likely to be easier and more developed. Trust, cooperation and participation in the community life are important characteristics of what is usually referred to as ‘social capital’. This paper looks at the link between social capital and economic growth. The specific interest on this link follows the contributions by Putnam (1993, 2000), arguing that civic engagement promotes the formation of trust, which in turn improves the quality of governance and thus economic performance. Analogously, Knack and Keefer (1997) argue that social capital may affect growth in a variety of ways. Economic activities that require some agents to rely on others are accomplished at lower cost in environments with higher social capital. Individuals spend less time to protect themselves from being exploited, thus being more productive. Entrepreneurs devote less time to monitor their employees and suppliers, thus having more time to devote to innovation. Governments are more credible, and people adopt more appropriate horizons in making investment decisions. Publicly provided goods, including education, have higher quality. Hiring decisions are more transparent and based on education credentials.1 The above view implicitly assumes that social capital has a positive effect on growth. The validation of this hypothesis has inspired several empirical studies, which typically rely upon data available at country, region or, at most, province level of aggregation (e.g., Algan and Cahuc, 2010; Tabellini, 2010). However, as argued by Putnam and Feldstein (2003, 9), ‘social capital is necessarily a local phenomenon because it is defined by connections among people who know one another’. In line with Putnam and Feldstein’s (2003) intuition, this paper investigates the link between social capital and growth at a more detailed level of spatial aggregation. In particular, we use municipality-level data for Italy. The research questions that we address are the following: What is the average elasticity of social capital to real growth over a long time span? Is it positive? Is there a causal link from social capital to real growth? Is there time heterogeneity in the link? Is there spatial heterogeneity? The answers to these questions are important because they shed further light on what lies behind the Solow residual, which is ‘a measure of our ignorance’ in standard empirical growth models, as emphasised by Abramovitz (1956). Most economies show a decreasing trend in the growth rate of real GDP per capita over time. Is this pattern related to a decreasing level of social capital? Or to a decreasing return to social capital? Or to both? Some economies show regional imbalances. Are they related to social-capital levels? Or to returns to those levels? Or to both? Although some of these questions are not novel, this paper makes five contributions. First, there is no systematic study of the social capital vs. growth nexus which answers all the above questions using longitudinal municipality-level data.2 These data have several advantages. Being longitudinal, they allow us to separate the growth effects of unobserved time-invariant city characteristics,3 such as the average quality of formal local institutions, from those of social capital attributes. As municipality-level data, they allow us to better approximate the local nature of social capital as well as to increase the number of cross-sectional units, making the empirical analysis more robust to the presence of outliers. Second, we innovate by accounting for the endogeneity of all the explanatory variables, including social capital, by means of internal instruments. In particular, we investigate whether a change in the social capital endowment of a municipality affects the growth rate of its per capita value added over the 1951–2001 period. Third, we investigate the extent to which lower growth in the South of Italy can be, in part, attributed not only to a lower level of social capital but also to a lower return to social capital. Fourth, we consider the possibility of a time-varying relationship between social capital and economic growth. If social capital behaves as a substitute of formal institutions in achieving economic growth (Ahlerup et al., 2009), its growth-enhancing role is likely to decline over time as formal institutions develop. In the specific case of Italy, the formal institutional system of the country was pretty much under construction during the 1950s, after the end of the Monarchy and the enforcement of the Republican Constitution in 1948. It is, therefore, likely that social capital played a more important role in that period than in others. Fifth, we provide novel local estimates of the impact of social capital on growth, using an identification approach based on two municipality-level external instruments (here, the term ‘local’ is used in the spirit of Angrist and Pischke, 2009). The rest of the paper is organised as follows. Section 2 provides reasons to use municipality-level data. Section 3 discusses our measures of social capital and growth. Section 4 describes the dataset and the empirical model. Section 5 presents results and robustness checks. Section 6 deals with local estimates. Section 7 concludes. 2. Municipalities as units of analysis The economic literature on social capital has discussed the interaction between various forms of social capital and economic development. For example, Sabatini (2008) has performed an empirical study using 200 variables on different aspects of economic development and four dimensions of social capital: strong family ties (generally referred to as ‘bonding’ social capital), weak informal ties (‘bridging’ social capital), voluntary organisations (‘linking’ social capital), and political participation. Each dimension includes from 4 to 25 variables. This should immediately alert the reader about the multiplicity of the definitions of social capital and outcome indicators that have been explored in earlier studies so far. One feature of this heterogeneity has to do with the nature of the data used, and specifically with the unit of analysis. With reference to the latter, the existing studies on social capital and economic outcomes can be broadly divided in two main categories: the ‘macro’ studies typically using data at regional or country level; and the ‘micro’ studies using data on individuals and firms. A striking feature of the literature is that the ‘micro’ studies are not necessarily consistent with the ‘macro’ studies, or vice-versa. On the one hand, many ‘macro’ studies report zero or even negative effect of social capital on economic performance, with Italy being a ‘special case’ where the link looks stronger and positive (Westlund and Adam, 2010).4 On the other hand, the ‘micro’ studies find clear evidence that social capital affects firm performance (e.g., Laursen et al., 2012) and people behaviour (e.g., Guiso et al., 2004).5 It follows that the evidence on the nexus between social capital and economic performance or behaviour at the micro (firm or individual) level, while having the expected sign, does not necessarily imply that social capital matters at the macro (GDP growth) level, which is the level investigated in this paper. To the extent that spatial proximity is crucial even in the Internet era (Rutten et al., 2010) and influences the cost of social capital investment (Glaeser et al., 2002), the step from firm or individual outcomes to regional or even national outcomes may be too big. This is one reason why a number of recent studies (Rutten and Gelissen, 2010; Corazzini et al., 2011; Eliasson et al., 2013; Albanese and de Blasio, 2014; Westlund et al., 2014) have focused on smaller spatial units, such as municipalities.6 However, there are four additional reasons for using municipality-level data: Disaggregation allows for the use of primary data on social capital. Our measure of social capital is voter turnout in political elections (Putnam, 1993). In Italy, measures of turnout rates in elections and referenda are produced at municipality level because the law (Decreto del Presidente della Repubblica del 20 marzo 1967) attributes to the municipalities the role of organizing and managing the voting operations, regardless of the type of popular consultation.7 It follows that provincial and regional data on voter turnout, extensively used in earlier studies, are inevitably secondary data, more likely to be measured with error. Aggregation can affect the level of social capital. Since the concept of social capital itself can only be defined within a given community, a change in the administrative definition of the community (e.g., municipality, province, region or nation) may affect the level of social capital. For instance, an inhabitant of a city may be very cooperative with an inhabitant of the same city, but not cooperative at all with an inhabitant of another city in the same province. Thus, cooperation may be high at city level, but low at province level. Aggregation can affect the estimate of the growth return to social capital. After Openshaw and Taylor (1979), who argued that correlations in Iowa between Republican voting and percentage of old people could vary a lot depending on how counties were aggregated, the modifiable area unit problem (MAUP) literature has expanded in several directions, including intersections with the growth literature. An example is Resende (2011) who has recently shown that ‘the determinants of economic growth in Brazil may manifest themselves differently on various spatial scales’ (Resende, 2011, 639), namely, states, micro-regions, spatial clusters and municipalities. Disaggregation provides both reliable generalized method of moments - system (GMM-SYS) estimates and novel two stage least squares (2SLS) estimates of the growth return to social capital. As we will see, the aggregation of data at provincial and regional level does not allow us to obtain reliable GMM-SYS estimates because identification criteria are not passed. In contrast, municipality-level data help us to obtain both reliable GMM-SYS estimates and novel 2SLS estimates of the causal effect of interest. The overall conclusion from the above discussion is that an analysis based on more disaggregated data is important because (dis)aggregation matters, or does it in principle, both from an empirical and a theoretical point of view. 3. Measures of growth and social capital The analysis of the link between social capital and growth with municipality-level data presents at least two empirical challenges. On the one hand, in the absence of a direct measure of economic growth at municipality level, we need to compute a proxy for the growth rate of the real per capita value added at such a detailed level of aggregation. We do this by using the municipality-level sectoral employment shares as weights to spread the value added at province level on all the municipalities of the same province. On the other hand, among the various definitions and related measures of social capital, we need to pick one definition whose measure is available at municipality level. In particular, we use the notion of civic capital (Guiso et al., 2011) proxied by the voter turnout at political elections (Putnam, 1993), i.e., the percentage of eligible voters who actually cast their vote. This measure captures the citizen engagement in the life of the community (Almond and Verba, 1963; Fowler, 2006) even though its use as a proxy of social capital may be controversial. This section discusses our measures of growth and social capital in detail. 3.1. Growth Our measure of economic development cannot be directly observed because, to the best of our knowledge, data on the value added are only available, at most, at the province level of aggregation. However, as province level data are available for each sector j, we can compute a proxy for the per capita value added at municipality level by using the sectoral employment shares—available at municipality level—as weights to spread the value added at province level on all municipalities of the same province. The sectoral employment shares are computed as the number of employees of a given sector in a given municipality over the total number of employees of the same sector in the entire province.8 In practice, we multiply the value added in sector j in province p (say vapj) times the employment share in sector j of municipality i in province p (say emplijemplpj). The result is the value added in sector j of municipality i (say vaij=vapj·emplijemplpj). This approach implicitly assumes that the average labour productivity in a municipality in a given sector is equal to that of the province in the chosen sector ( vaijemplij=vapjemplpj). This assumption is consistent with the idea that labour mobility within a province implies an equalisation of wages and productivities in a sector among municipalities in the same province, which is in turn consistent with the long-run approach taken in this paper.9 Adding up the value added of each sector at municipality level, we get the total value added at municipality level (say vai=∑jvaij). We use the population to compute the per capita value added. The real per capita value added is obtained using the national price index as deflator (base year = 1951). Finally, as we can only compute the value added at Census dates (i.e. every 10 years), its growth rate is obtained as difference between the natural logarithms of real per capita value added at Census dates. 3.2. Social capital But, what do we mean by social capital here? And, how do we measure it? The rest of this section provides an answer to both these questions. 3.2.1 Defining social capital as civic capital After its introduction by Bourdieu (1972) at the beginning of the 1970s,10 the concept of social capital has gained wide acceptance in social sciences, including economics, over the past four decades. Yet, its success has been proportional to the ambiguity of the term.11 In this paper, we follow the literature originated by Solow (1995) who described the specific characteristics that social capital must have ‘to be more than a buzzword’. First, to retain the term ‘capital’, likewise other forms of capital, social capital must have a dynamics of accumulation and depreciation. Second, it must have a non-negative economic payoff. Third, it has to be measurable, even inexactly. Fourth, to retain the term ‘social’, social capital must be clearly distinguishable from other forms of capital accumulated at firm (physical capital) or individual (human capital) level. As convincingly argued by Guiso et al. (2011), a concept of social capital which fits all the Solow criteria is the concept of ‘civic capital’, defined as ‘those persistent and shared beliefs and values that help a group overcome the free rider problem in the pursuit of socially valuable activities’ (Guiso et al., 2011, 419).12Guiso et al. (2011) go one step further suggesting that civic capital is not just one, but it is the only concept of social capital satisfying all the Solow criteria. Despite the latter may be controversial, and a less radical view is advisable, this paper follows the approach by Guiso et al. (2011) using a proxy of civic capital in the empirical analysis. 3.2.2 Measuring civic capital as voter turnout From now on, we will use the expressions ‘social capital’ and ‘civic capital’ interchangeably. Guiso et al. (2011) provide an outstanding overview of all the possible direct and indirect measures of civic capital. Direct measures of civic capital are not currently available at municipality level. So, the analysis must rely on indirect measures. According to Guiso et al. (2011), the most important indirect measures of civic capital are two: anonymous blood donations and participation in referenda. The former is an ideal measure of social capital because an anonymous blood donation is a free individual choice giving no personal returns and supporting the community life. The latter measure, instead, though widely used in the earlier research, is subject to two types of criticism. First, low participation in a referendum does not necessarily mean low participation in the community life—i.e., low social capital in a particular community—because people who are in favour of a particular law may choose not to vote in order to avoid that a referendum reaches the 50% cutoff needed to repeal a law.13 Second, people may have a personal return from a particular law under scrutiny (e.g., married couples and divorce, Catholic women and abort and so on). Both the above two indirect measures of civic capital are not available at municipality level for our period of interest. Thus, we must rely on a different indirect measure of civic capital. In particular, we use voter turnout at political elections as a proxy of civic capital (Putnam, 1993). The usual argument against our proxy is that voting at political elections may be a constrained choice, not a free one as an anonymous blood donation. In addition, it may imply personal returns as politicians may provide benefits to their electorate in exchange for votes. However, both these two arguments do not seem to apply to the case of this paper. The dutifulness of the vote was debated in the Assemblea Costituente. The compromise between those who wanted to qualify the exercise of voting as a legal duty and those who, instead, wanted to qualify it as a moral duty (Solenne and Verrilli, 2012) led to the qualification of voting as a civic duty in the constitutional chart (Costituzione della Repubblica italiana, art. 48). The constitutional norm allowed for the legislator to set only weak penalisations for voter abstention in 1957 (Decreto del Presidente della Repubblica del 30 marzo 1957, art. 115). Abstention should have been justified to the city mayor, who in turn had to publish a list of unjustified non-shows in the municipal albo for 1 month. In addition, the expression ‘he/she did not vote’ had to be recorded for 5 years in the certificato di buona condotta issued by the municipality. However, the application of the 1957 law was progressively disregarded (Cuturi et al., 2000), until the electoral reform of 1993 (Decreto Legislativo del 20 dicembre 1993) fully recognised the abstention as an expression of political choice (Cassese et al., 2009). Hence, even though voting is still intended as a civic duty in the Constitution—something that one should do—it is reasonable to think that voting at political elections has always been a substantially free choice because no punishment for abstention has ever been set, besides a mere register of non-voting. As for the personal returns, the possibility of a benefit-vote link is a real issue and should not be disregarded. However, the argument is more likely to apply to the administrative elections (e.g., city mayor) than to the political ones. Indeed, elected deputies in Italy have the constitutional right to operate without a binding mandate (Costituzione della Repubblica italiana, art. 67), meaning that, once elected, they can do whatever they like, even against the interests of their party or electorate. In addition, despite in the elections covered in our sample people could vote both for parties and specific persons, parties and elected typically had to set coalitions in order to achieve a Parliamentary majority and form a Government. Therefore, compared with a referendum where the potential outcomes are known with certainty (abrogation or not of a specific law), the vote at political elections provided, at best, non-predictable personal returns to the average voter.14 In contrast, the potential troubles could be easily forecasted: the law established—and still does—strong criminal sanctions for both those who promise and those who accept benefits in exchange for votes at political elections (Decreto del Presidente della Repubblica del 30 marzo 1957, art. 96). In sum, voter turnout at political elections may not be the best measure of social capital15 but we use it in this paper because: (i) it has already been used by many authors including Putnam (1993), (ii) it suffers from fewer distortions than referenda turnouts and (iii) to the best of our knowledge, it is the only measure available at municipality level over the long time span that we investigate. The ‘smoking gun’ is the relationship between data on anonymous blood donations (see Guiso et al., 2004), available only for the year of 1995 and at province level, and data on voter turnout at Parliamentary elections (Chamber of Deputies) for the closest year of elections in our dataset, i.e., 1992. Figure 1 suggests that an increase in voter turnout is associated with a more-than-proportional increase in blood donations. The estimated elasticity is, indeed, equal to 7.2 and statistically significant at 1% level. Figure 1 View largeDownload slide Correlation between voter turnout and blood donations. Notes: Data are at province level. Blood donations are measured as the number of blood bags (each bag contains 16 ounces of blood) per million inhabitants in the province. Voter turnout refers to the Parliamentary elections (Chamber of Deputies). Figure 1 View largeDownload slide Correlation between voter turnout and blood donations. Notes: Data are at province level. Blood donations are measured as the number of blood bags (each bag contains 16 ounces of blood) per million inhabitants in the province. Voter turnout refers to the Parliamentary elections (Chamber of Deputies). In addition, the evidence presented in the Appendix Table A1 supports our choice because voter turnout at political elections is, in general, linearly correlated with other proxies of social capital available at province level, with the expected sign (note that imposing a linear association between two variables may be rather restrictive, as documented in Figure 1). 4. Empirical strategy Our main data sources are: (i) the Istituto Tagliacarne for data on the value added at province level; (ii) the Istituto Nazionale di Statistica (ISTAT) for the Census data on surface, population and employment at municipality level; (iii) the Atlante Storico Elettorale d’Italia provided by the Istituto Cattaneo for data on political participation, measured as voter turnout at Parliamentary elections (Chamber of Deputies) at municipality level and (iv) the Associazione Nazionale dei Comuni Italiani (ANCI) for data on geographic characteristics of cities. We use data on about 2100 municipalities with at least 5000 inhabitants in 1951 and we consider the 1951–2001 period, divided in time intervals that correspond to the Census decades. The analysis is not extended further in time because of data constraints. Table 1 provides the summary statistics for the sample. On average, each city has a population of 20,349 inhabitants and extends over 71 km2. The 17% of cities in the sample is located close to the seaside and the 40% is located in the South. The real per capita value added is on average equal to 797,000 constant Italian liras (base year = 1951) and displays a 39% average growth rate over 10 years. The first growth rate refers to the 1951–1961 period. The last is for 1991–2001. Our measure of social capital displays an average of 90%. Ideally, we would like to have data on social capital every 10 years since 1951. In practice, we use data on voter turnout at political elections held in 1953, 1963, 1968, 1979 and 1992. Table 1 Summary statistics Variable Observation Mean Standard deviation Min. Max. Real growth rate (10 years) 10,145 0.39 0.37 −2.18 3.28 Social capital 10,145 0.90 0.07 0.41 0.99 Population 10,145 20,349.42 82,210.04 771 2,546,804 Real income (base year = 1951) 10,145 797.58 525.78 35.56 6613.81 Altitude 10,145 233.42 229.10 0 1211 Difference altitude 10,145 532.60 547.14 1 3282 Coast location 10,145 0.17 0.37 0 1 Surface 10,145 71.10 77.87 1.62 1307.71 Provincial capital 10,145 0.05 0.22 0 1 South 10,145 0.40 0.49 0 1 Variable Observation Mean Standard deviation Min. Max. Real growth rate (10 years) 10,145 0.39 0.37 −2.18 3.28 Social capital 10,145 0.90 0.07 0.41 0.99 Population 10,145 20,349.42 82,210.04 771 2,546,804 Real income (base year = 1951) 10,145 797.58 525.78 35.56 6613.81 Altitude 10,145 233.42 229.10 0 1211 Difference altitude 10,145 532.60 547.14 1 3282 Coast location 10,145 0.17 0.37 0 1 Surface 10,145 71.10 77.87 1.62 1307.71 Provincial capital 10,145 0.05 0.22 0 1 South 10,145 0.40 0.49 0 1 Notes: Data sources are: ANCI, ISTAT, Istituto Cattaneo, Istituto Tagliacarne. Social capital is measured with voter turnout at the Parliamentary elections (Chamber of Deputies). Table 1 Summary statistics Variable Observation Mean Standard deviation Min. Max. Real growth rate (10 years) 10,145 0.39 0.37 −2.18 3.28 Social capital 10,145 0.90 0.07 0.41 0.99 Population 10,145 20,349.42 82,210.04 771 2,546,804 Real income (base year = 1951) 10,145 797.58 525.78 35.56 6613.81 Altitude 10,145 233.42 229.10 0 1211 Difference altitude 10,145 532.60 547.14 1 3282 Coast location 10,145 0.17 0.37 0 1 Surface 10,145 71.10 77.87 1.62 1307.71 Provincial capital 10,145 0.05 0.22 0 1 South 10,145 0.40 0.49 0 1 Variable Observation Mean Standard deviation Min. Max. Real growth rate (10 years) 10,145 0.39 0.37 −2.18 3.28 Social capital 10,145 0.90 0.07 0.41 0.99 Population 10,145 20,349.42 82,210.04 771 2,546,804 Real income (base year = 1951) 10,145 797.58 525.78 35.56 6613.81 Altitude 10,145 233.42 229.10 0 1211 Difference altitude 10,145 532.60 547.14 1 3282 Coast location 10,145 0.17 0.37 0 1 Surface 10,145 71.10 77.87 1.62 1307.71 Provincial capital 10,145 0.05 0.22 0 1 South 10,145 0.40 0.49 0 1 Notes: Data sources are: ANCI, ISTAT, Istituto Cattaneo, Istituto Tagliacarne. Social capital is measured with voter turnout at the Parliamentary elections (Chamber of Deputies). Figure 2 suggests that voter turnout in Italy has decreased over time.16 This pattern is consistent with the one found in Albanese and de Blasio (2014). Figure 2 View largeDownload slide The evolution of voter turnout at political elections. Figure 2 View largeDownload slide The evolution of voter turnout at political elections. Interestingly, Figure 3 depicts a pattern of growth which is similar to the one observed for our proxy of social capital. What is really intriguing, however, is the fact that growth has decreased proportionally more than social capital, which opens the door to the possibility that the growth return to social capital has decreased over time. Figure 3 View largeDownload slide The evolution of real growth. Figure 3 View largeDownload slide The evolution of real growth. As for the empirical strategy, our idea is that social capital may constitute a pre-condition for future growth. Hence, we estimate a model of the following form: gi,t=μi+μt+αln yi,t−10+βln sci,t−10+γln popi,t−10+∑k=1Nδk Xk,i+ei,t, (1) where gi,t is the growth rate of real per capita value added between year t and year t−10 for the unit of interest i (the first observed growth rate refers to the 1951–1961 period); μi is the vector of the city-fixed effects (FEs); μt represents the time effects between year t and year t−10 (1951–1961, 1961–1971 and so on); ln yi,t−10 is the log of real per capita value added at the beginning of the period (the first observation is for 1951); ln sci,t−10 is the log of social capital endowment at the beginning of the period (the first observation should be for 1951, but the closest observation we have is for 1953)17; ln popi,t−10 is the log of population at the beginning of the period (the first observation is for 1951); X1 to XN are observable, mainly geographical, control variables and ei,t is an i.i.d. disturbance. Our model is a ‘social capital and growth’ version of a famous ‘financial intermediation and growth’ model proposed by Levine et al. (2000) in an article in the Journal of Monetary Economics. This article has been very influential in the finance–growth literature and its title has inspired the title of our manuscript. Following Levine et al. (2000), our model can be written as a dynamic panel-data model by noting that gi,t≅ln yi,t−ln yi,t−10. As our model is specified in logarithms, the parameter of interest β should be interpreted as the elasticity of the growth rate of real per capita value added with respect to social capital.18 Our empirical model imposes a given direction of causality, which goes from social capital to GDP growth. This hypothesis is supported by the panel VAR estimates proposed in Appendix A. We control for city-FEs to capture unobserved time-invariant city characteristics, such as the average quality of formal local institutions. We control for the lagged log of population and time effects to capture agglomeration economies and decade heterogeneity, respectively. The variables X1 to XN are as follows: a dummy equal to one if the municipality is located in the South; a dummy equal to one if the municipality is close to the seaside; a dummy equal to one if it is a province capital; the log of surface in square kilometres; the log of altitude from the townhouse in metres and the log of difference in altitudes within the municipality. We control for geographical variables for three main econometric reasons: Turnout is potentially correlated with geographical factors. For example, altitude may be correlated with turnout through weather conditions (the probability of raining in the day of the elections is higher at higher altitudes).19 Hence, not controlling for geographical factors would introduce a new source of social-capital endogeneity in the residuals of the growth regression. Geographical variables are useful for identification as instruments in dynamic panel-data models because they can be treated as exogenous covariates. Geography explains a share of growth variability in regression analyses using country-level data, as stressed by Gallup et al. (1999), among others. Controlling for geographical variables in studies using more disaggregated data is in line with Gennaioli et al. (2013) and Ketterer and Rodríguez-Pose (2018), among others. Our estimation strategy is as follows. We start by disregarding the presence of city-FEs and estimate model (1) by ordinary least squares (OLS). We first control for initial income only; we then add time dummies and finally, we add both time dummies and the control set. Our definition of ‘control set’ includes lagged population and all the variables X1 to XN. Afterwards, we control for the role of unobserved heterogeneity through a FEs approach. Note, however, that we expect the FE estimates to be biased due to the presence of both unobserved heterogeneity and lagged value added among regressors. This source of bias can be only resolved by estimating model (1) with a dynamic GMM approach. In the dynamic GMM framework, a GMM-system (GMM-SYS) approach (Arellano and Bover, 1995; Blundell and Bond, 1998) should be the preferred estimation method for a variety of reasons. First, like the GMM-difference (GMM-DIF) approach (Arellano and Bond, 1991), it allows for dealing with the potential endogeneity of all explanatory variables, rather than only that of social capital. Second, it is more efficient than GMM-DIF as it uses more moment conditions. Indeed, likewise GMM-DIF, the GMM-SYS approach uses lags of regressors as instruments for the first-difference transformation of model (1). But, differently from GMM-DIF, the GMM-SYS approach also uses lagged differences of regressors as instruments for model (1) itself. Third, GMM-SYS allows us to estimate the coefficients of time-invariant regressors, while GMM-DIF does not. Fourth, the GMM-SYS estimator has been shown to behave better (in terms of bias and precision) than the GMM-DIF estimator in case of persistent variables (Blundell and Bond, 1998; Blundell et al., 2000), as social capital is likely to be. The choice of instruments is based on Bond et al. (2001), who show how to estimate empirical growth models using the GMM-system approach. In particular, as instruments for the first-difference equation, we use ln yi,t−20, ln sci,t−20, ln popi,t−20 and earlier. As instruments of the level equation, we use Δln yi,t−10, Δln sci,t−10, Δln popi,t−10 and earlier. The variables X1 to XN are taken as exogenous covariates and used as instruments in the standard dynamic GMM practice. The instrument matrix is collapsed (Roodman, 2009). Standard errors are Windmeijer-corrected (Windmeijer, 2005). The goodness of fit for GMM-SYS is based on Bloom et al. (2007). The results are presented in the next section together with a variety of robustness checks. 5. Results and robustness In presenting the results, we focus on the link between social capital and growth. Before commenting on our preferred estimate, it is worth investigating how the social capital coefficient changes in different specifications and estimation methods. 5.1. Main results Controlling for initial income only, the OLS estimate of the social capital coefficient (Table 2, Column 1) is positive and significant. Adding time effects (Column 2), the estimate slightly decreases in terms of magnitude but remains highly significant. When adding the control set (Column 3), the coefficient remains significant but its magnitude decreases. When aggregating the data at provincial and regional level, we find evidence of a MAUP (see Appendix Table A2). In line with Resende (2011), aggregation causes the return to social capital to change and the standard error to get larger. Table 2 Conditional mean estimates on growth: baseline estimates (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.853*** 0.749*** 0.309*** 0.221*** 0.170** 0.595*** (0.0301) (0.034) (0.039) (0.067) (0.068) (0.109) Initial income −0.302*** −0.281*** −0.395*** −0.903*** −0.904*** −0.660*** (0.004) (0.007) (0.008) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.375 0.387 0.441 0.682 0.683 0.407 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.853*** 0.749*** 0.309*** 0.221*** 0.170** 0.595*** (0.0301) (0.034) (0.039) (0.067) (0.068) (0.109) Initial income −0.302*** −0.281*** −0.395*** −0.903*** −0.904*** −0.660*** (0.004) (0.007) (0.008) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.375 0.387 0.441 0.682 0.683 0.407 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 2 Conditional mean estimates on growth: baseline estimates (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.853*** 0.749*** 0.309*** 0.221*** 0.170** 0.595*** (0.0301) (0.034) (0.039) (0.067) (0.068) (0.109) Initial income −0.302*** −0.281*** −0.395*** −0.903*** −0.904*** −0.660*** (0.004) (0.007) (0.008) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.375 0.387 0.441 0.682 0.683 0.407 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.853*** 0.749*** 0.309*** 0.221*** 0.170** 0.595*** (0.0301) (0.034) (0.039) (0.067) (0.068) (0.109) Initial income −0.302*** −0.281*** −0.395*** −0.903*** −0.904*** −0.660*** (0.004) (0.007) (0.008) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.375 0.387 0.441 0.682 0.683 0.407 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. A source of bias may be due to the fact that unobserved time-invariant city characteristics (e.g., the average quality of formal local institutions) are likely to be correlated with social capital. When controlling for such characteristics, the FE estimates (Column 4 and, including the control set, Column 5) are lower in magnitude than the OLS ones, but they are still positive and significant. As highlighted in the previous section, our preferred estimation method is the one that allows for dealing with the endogeneity of all covariates, including the lagged value added, as well as with the unobserved heterogeneity, i.e., the GMM-SYS approach. The GMM-SYS estimates (Column 6) confirm the positive and significant link between social capital and growth. The social capital coefficient estimate is larger than the one obtained using the FE estimator. The identification approach is supported by the fact that both the Hansen and the Difference-in-Hansen tests do not reject the null hypothesis of instruments joint validity (the p-values are equal to 0.276 and 0.131, respectively) as well as by the fact that the Arellano–Bond test does not reject the null hypothesis of no second-order autocorrelation of residuals in first differences (the p-value is equal to 0.819). In contrast, identification is problematic when using data at province level because the instruments for the level equation do not pass the Difference-in-Hansen test (see Column 6 in Appendix Table A3(a) for estimation results). Identification is even more problematic when aggregating the data at regional level because the Arellano–Bond test rejects the null at 10%, the Hansen test rejects the null at 10%, and the Difference-in-Hansen test rejects the null at 5%. Interestingly, we find that aggregation changes the return to social capital, which lowers as aggregation increases (see the GMM-SYS estimates in Table 2, Appendix Table A3(a,b)). This is in line with Westlund and Adam (2010) because the growth return to social capital becomes not statistically significant at 5% level when aggregating at provincial level (see Column 6 in Appendix Table A3(a) for estimation results), and even negative when aggregating at regional level (see Column 6 in Appendix Table A3(b) for estimation results). The OLS results in Appendix Table A2 go in the same direction. An important thing to notice is that the additional moment conditions imposed by the GMM-SYS estimator are valid provided the validity of some initial-condition restrictions. The latter hold under (sufficient but not necessary) assumptions of mean stationarity of the stochastic processes that generate the data involved in the estimation. This assumption also guarantees that the instruments for the first-difference equation are not weak (Blundell and Bond, 1998, 2000; Bond, 2002). In order to test the mean-stationarity hypothesis, we follow Blundell and Bond (2000) and estimate an AR(1) process (with a constant) for the log of real per capita value added, the log of the social capital proxy and the log of population. In the presence of city FEs, the OLS method gives an upward-biased estimate of the lagged dependent variable coefficient. This means that an OLS estimate of the autoregression coefficient below the critical value of 1 would ensure us that we are not working with unit-root series. In the case of the real per capita value added, the estimate is 0.7, i.e., well below the critical value. In the case of the social capital proxy, the estimate is equal to 0.9, i.e., still below the critical value. In the case of population, the estimate increases to 1.0 but the fact that it is upward biased reassures about the mean stationarity of the true data-generating process. Intuitively, there is no reason to believe that growth, social capital and population evolve as random walks. It is more realistic to think that steady-state (i.e., equilibrium) values of these variables do exist, even though persistence is high. The latter is also consistent with standard growth models, including ours. Our first result is that the social capital endowment of a city matters for subsequent economic growth in a positive and significant way. In particular, according to the preferred estimation method (GMM-SYS), the elasticity of social capital endowment with respect to growth is 0.6. The coefficient of initial income is found to be negative and significant. This suggests that municipalities with higher (lower) initial income grow slower (faster), which supports the idea of convergence. However, we do not find evidence of convergence among all Italian municipalities. Since the estimated coefficient of the dummy for location in the South is negative (−0.341) and significant at 1% level,20 the municipalities in the Centre-North of Italy converge to an equilibrium growth rate which is higher than in the South. In the rest of this section, we will provide additional results useful to explain time and spatial growth variability, and the role potentially played by social capital. 5.2. Spatial externalities As a first robustness check, we control for the existence of spatial externalities (Table 3). Specifically, we augment model (1) with the aggregate initial income of all the other municipalities in the same province. The GMM-SYS estimate of the coefficient of this variable is very close to zero from the left (−0.006) and it is not significant. As a result, the social capital impact changes only a little. If we construct an interaction between the initial income of the municipality and the aggregate initial income of all the other municipalities in the same province, the interaction coefficient is still very close to zero (−0.002) and not significant. Again the social capital coefficient does not change much (since the results are very similar to those in Table 3, we omit their presentation in a table). Table 3 Conditional mean estimates on growth controlling for spatial externalities (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.819*** 0.708*** 0.319*** 0.217*** 0.163** 0.562*** (0.030) (0.036) (0.039) (0.068) (0.069) (0.110) Initial income −0.306*** −0.285*** −0.395*** −0.903*** −0.905*** −0.624*** (0.004) (0.006) (0.007) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.382 0.388 0.441 0.682 0.683 0.409 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.819*** 0.708*** 0.319*** 0.217*** 0.163** 0.562*** (0.030) (0.036) (0.039) (0.068) (0.069) (0.110) Initial income −0.306*** −0.285*** −0.395*** −0.903*** −0.905*** −0.624*** (0.004) (0.006) (0.007) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.382 0.388 0.441 0.682 0.683 0.409 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 3 Conditional mean estimates on growth controlling for spatial externalities (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.819*** 0.708*** 0.319*** 0.217*** 0.163** 0.562*** (0.030) (0.036) (0.039) (0.068) (0.069) (0.110) Initial income −0.306*** −0.285*** −0.395*** −0.903*** −0.905*** −0.624*** (0.004) (0.006) (0.007) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.382 0.388 0.441 0.682 0.683 0.409 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.819*** 0.708*** 0.319*** 0.217*** 0.163** 0.562*** (0.030) (0.036) (0.039) (0.068) (0.069) (0.110) Initial income −0.306*** −0.285*** −0.395*** −0.903*** −0.905*** −0.624*** (0.004) (0.006) (0.007) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.382 0.388 0.441 0.682 0.683 0.409 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. We also estimate a spatial panel Arellano–Bond linear dynamic regression using a spatial weight matrix based on inverse distance. In this model, the social capital coefficient is equal to 0.525 with a standard error equal to 0.235, therefore being significant at 5%. This compares favourably with our main estimate of 0.595, which is significant at 1% level. The implication is that, while spatial correlations may in principle play a role, there is no evidence of such a role being crucial in our sample. 5.3. Non-linearities Some explanatory variables may also affect growth in a way that depends on the level of other explanatory variables, such as deeply lagged population or initial income level.21 In Table 4, we replicated Table 2 using both (social capital) and (social capital × population in 1951) as covariates. The evidence suggests that population in 1951 does not generally affect the growth return to social capital. The intuition is that controlling for initial income and population makes the population size in 1951 redundant in the model. Table 4 Conditional mean estimates on growth controlling for interaction between social capital and population size in 1951 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.705*** 0.797*** 0.325*** −0.627 −0.676 0.558*** (0.0335) (0.0376) (0.0427) (0.782) (0.784) (0.102) Initial income −0.248*** −0.294*** −0.411*** −0.821*** −0.820*** −0.654*** (0.00362) (0.00662) (0.00725) (0.00955) (0.00956) (0.0239) Social capital × 0.000249 0.00148 −0.0129*** 0.0775 0.0813 0.00579 Population in 1951 (0.000956) (0.000944) (0.00255) (0.0852) (0.0853) (0.00542) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.335 0.366 0.435 0.616 0.616 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.705*** 0.797*** 0.325*** −0.627 −0.676 0.558*** (0.0335) (0.0376) (0.0427) (0.782) (0.784) (0.102) Initial income −0.248*** −0.294*** −0.411*** −0.821*** −0.820*** −0.654*** (0.00362) (0.00662) (0.00725) (0.00955) (0.00956) (0.0239) Social capital × 0.000249 0.00148 −0.0129*** 0.0775 0.0813 0.00579 Population in 1951 (0.000956) (0.000944) (0.00255) (0.0852) (0.0853) (0.00542) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.335 0.366 0.435 0.616 0.616 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 4 Conditional mean estimates on growth controlling for interaction between social capital and population size in 1951 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.705*** 0.797*** 0.325*** −0.627 −0.676 0.558*** (0.0335) (0.0376) (0.0427) (0.782) (0.784) (0.102) Initial income −0.248*** −0.294*** −0.411*** −0.821*** −0.820*** −0.654*** (0.00362) (0.00662) (0.00725) (0.00955) (0.00956) (0.0239) Social capital × 0.000249 0.00148 −0.0129*** 0.0775 0.0813 0.00579 Population in 1951 (0.000956) (0.000944) (0.00255) (0.0852) (0.0853) (0.00542) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.335 0.366 0.435 0.616 0.616 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.705*** 0.797*** 0.325*** −0.627 −0.676 0.558*** (0.0335) (0.0376) (0.0427) (0.782) (0.784) (0.102) Initial income −0.248*** −0.294*** −0.411*** −0.821*** −0.820*** −0.654*** (0.00362) (0.00662) (0.00725) (0.00955) (0.00956) (0.0239) Social capital × 0.000249 0.00148 −0.0129*** 0.0775 0.0813 0.00579 Population in 1951 (0.000956) (0.000944) (0.00255) (0.0852) (0.0853) (0.00542) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.335 0.366 0.435 0.616 0.616 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. As a second exercise, we augment model (1) with the interaction between social capital and the initial value added (Table 5). The results, however, are not much different from those in Table 2 and the coefficient of the interaction term is very small in magnitude. Furthermore, when we estimate the specification with the GMM-SYS method, the coefficient of the interaction is not significant. Hence, we do not find evidence of significant non-linearities, with respect to the initial-income level. Table 5 Conditional mean estimates on growth controlling for interaction between social capital and initial income (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.753*** 0.451*** 0.096** 0.263* 0.216*** 0.653*** (0.030) (0.030) (0.041) (0.070) (0.071) (0.121) Initial income −0.406*** −0.400*** −0.475*** −0.900*** −0.901*** −0.675*** (0.009) (0.009) (0.009) (0.010) (0.010) (0.028) Social capital × 0.022*** 0.034*** 0.027*** −0.004** −0.005** −0.004 Initial income (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.386 0.409 0.454 0.682 0.683 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.753*** 0.451*** 0.096** 0.263* 0.216*** 0.653*** (0.030) (0.030) (0.041) (0.070) (0.071) (0.121) Initial income −0.406*** −0.400*** −0.475*** −0.900*** −0.901*** −0.675*** (0.009) (0.009) (0.009) (0.010) (0.010) (0.028) Social capital × 0.022*** 0.034*** 0.027*** −0.004** −0.005** −0.004 Initial income (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.386 0.409 0.454 0.682 0.683 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 5 Conditional mean estimates on growth controlling for interaction between social capital and initial income (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.753*** 0.451*** 0.096** 0.263* 0.216*** 0.653*** (0.030) (0.030) (0.041) (0.070) (0.071) (0.121) Initial income −0.406*** −0.400*** −0.475*** −0.900*** −0.901*** −0.675*** (0.009) (0.009) (0.009) (0.010) (0.010) (0.028) Social capital × 0.022*** 0.034*** 0.027*** −0.004** −0.005** −0.004 Initial income (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.386 0.409 0.454 0.682 0.683 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.753*** 0.451*** 0.096** 0.263* 0.216*** 0.653*** (0.030) (0.030) (0.041) (0.070) (0.071) (0.121) Initial income −0.406*** −0.400*** −0.475*** −0.900*** −0.901*** −0.675*** (0.009) (0.009) (0.009) (0.010) (0.010) (0.028) Social capital × 0.022*** 0.034*** 0.027*** −0.004** −0.005** −0.004 Initial income (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.386 0.409 0.454 0.682 0.683 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. In addition, we consider the possibility of differentiated returns between the Italian macro areas. In line with Putnam (1993), the descriptive statistics suggest that the level of social capital is lower in the South, with an average of 85% against an average of 94% in the Centre–North in the 1953–1992 period. Thus, one may expect the productivity of social capital to be lower where the level of social capital is higher (law of decreasing marginal productivity). To test this hypothesis, we replicate the analysis in Table 2 by adding an interaction term between social capital and a dummy for South. In this model, the coefficient of social capital alone represents the impact of social capital on growth in the Centre–North. The impact in the South is given by the sum between the coefficient of social capital alone and the coefficient of the interaction term (social capital × South). The regression analysis is somehow surprising. In our preferred specification, the elasticity of social capital with respect to growth in the South is 0.551 (Table 6), being 79% of the elasticity for the rest of Italy (0.696). This suggests that not only the level but also the return of social capital is lower in the South. Table 6 Conditional mean estimates on growth controlling for interaction between social capital and South (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.463*** 0.420*** 0.324*** 0.341** 0.296** 0.696*** (0.034) (0.036) (0.038) (0.134) (0.134) (0.107) Initial income −0.340*** −0.348*** −0.381*** −0.903*** −0.905*** −0.661*** (0.004) (0.007) (0.007) (0.010) (0.010) (0.024) Social capital × −0.037*** −0.039*** −0.046*** −0.141 −0.149 −0.145*** South (0.001) (0.001) (0.001) (0.137) (0.136) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.408 0.422 0.436 0.682 0.683 0.423 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.463*** 0.420*** 0.324*** 0.341** 0.296** 0.696*** (0.034) (0.036) (0.038) (0.134) (0.134) (0.107) Initial income −0.340*** −0.348*** −0.381*** −0.903*** −0.905*** −0.661*** (0.004) (0.007) (0.007) (0.010) (0.010) (0.024) Social capital × −0.037*** −0.039*** −0.046*** −0.141 −0.149 −0.145*** South (0.001) (0.001) (0.001) (0.137) (0.136) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.408 0.422 0.436 0.682 0.683 0.423 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 6 Conditional mean estimates on growth controlling for interaction between social capital and South (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.463*** 0.420*** 0.324*** 0.341** 0.296** 0.696*** (0.034) (0.036) (0.038) (0.134) (0.134) (0.107) Initial income −0.340*** −0.348*** −0.381*** −0.903*** −0.905*** −0.661*** (0.004) (0.007) (0.007) (0.010) (0.010) (0.024) Social capital × −0.037*** −0.039*** −0.046*** −0.141 −0.149 −0.145*** South (0.001) (0.001) (0.001) (0.137) (0.136) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.408 0.422 0.436 0.682 0.683 0.423 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.463*** 0.420*** 0.324*** 0.341** 0.296** 0.696*** (0.034) (0.036) (0.038) (0.134) (0.134) (0.107) Initial income −0.340*** −0.348*** −0.381*** −0.903*** −0.905*** −0.661*** (0.004) (0.007) (0.007) (0.010) (0.010) (0.024) Social capital × −0.037*** −0.039*** −0.046*** −0.141 −0.149 −0.145*** South (0.001) (0.001) (0.001) (0.137) (0.136) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.408 0.422 0.436 0.682 0.683 0.423 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Finally, we consider the possibility that the link between social capital and growth varies with time. To investigate this issue, we allow for the social capital coefficient to change over time by interacting social capital with time dummies (Table 7). The coefficient of social capital at baseline represents the impact of social capital on the 1951–1961 growth rate. This number is 1.055 in our preferred specification. The interaction coefficients are always significant and negative. This means that social capital matters more for growth in the 1951–1961 period than after.22 The latter is another important result of our analysis. A similar result is presented by Albanese and de Blasio (2014) who find evidence of a positive and time-decreasing correlation between voter turnout and measures of economic development other than growth, at the municipality level of aggregation. Table 7 Conditional mean estimates on growth controlling for interaction between social capital and time effects (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.773*** 1.391*** 0.709*** 0.548*** 0.573*** 1.055*** (0.035) (0.144) (0.141) (0.161) (0.162) (0.195) Social capital × 1961–1971 0.034*** −0.562*** −0.544*** −0.421*** −0.451*** −0.672*** (0.002) (0.167) (0.159) (0.146) (0.148) (0.218) Social capital × 1971–1981 0.049*** −0.416*** −0.326** −0.268* −0.304** −0.515*** (0.002) (0.160) (0.152) (0.145) (0.148) (0.198) Social capital × 1981–1991 0.040*** −0.983*** −0.777*** −0.683*** −0.720*** −0.966*** (0.003) (0.153) (0.146) (0.143) (0.147) (0.195) Social capital × 1991–2001 0.026*** −0.384** −0.218 −0.251* −0.295** −0.488** (0.003) (0.153) (0.146) (0.144) (0.149) (0.196) Initial income −0.291*** −0.293*** −0.410*** −0.819*** −0.819*** −0.597*** (0.006) (0.006) (0.007) (0.009) (0.009) (0.023) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.366 0.371 0.437 0.618 0.618 0.511 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.773*** 1.391*** 0.709*** 0.548*** 0.573*** 1.055*** (0.035) (0.144) (0.141) (0.161) (0.162) (0.195) Social capital × 1961–1971 0.034*** −0.562*** −0.544*** −0.421*** −0.451*** −0.672*** (0.002) (0.167) (0.159) (0.146) (0.148) (0.218) Social capital × 1971–1981 0.049*** −0.416*** −0.326** −0.268* −0.304** −0.515*** (0.002) (0.160) (0.152) (0.145) (0.148) (0.198) Social capital × 1981–1991 0.040*** −0.983*** −0.777*** −0.683*** −0.720*** −0.966*** (0.003) (0.153) (0.146) (0.143) (0.147) (0.195) Social capital × 1991–2001 0.026*** −0.384** −0.218 −0.251* −0.295** −0.488** (0.003) (0.153) (0.146) (0.144) (0.149) (0.196) Initial income −0.291*** −0.293*** −0.410*** −0.819*** −0.819*** −0.597*** (0.006) (0.006) (0.007) (0.009) (0.009) (0.023) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.366 0.371 0.437 0.618 0.618 0.511 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 7 Conditional mean estimates on growth controlling for interaction between social capital and time effects (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.773*** 1.391*** 0.709*** 0.548*** 0.573*** 1.055*** (0.035) (0.144) (0.141) (0.161) (0.162) (0.195) Social capital × 1961–1971 0.034*** −0.562*** −0.544*** −0.421*** −0.451*** −0.672*** (0.002) (0.167) (0.159) (0.146) (0.148) (0.218) Social capital × 1971–1981 0.049*** −0.416*** −0.326** −0.268* −0.304** −0.515*** (0.002) (0.160) (0.152) (0.145) (0.148) (0.198) Social capital × 1981–1991 0.040*** −0.983*** −0.777*** −0.683*** −0.720*** −0.966*** (0.003) (0.153) (0.146) (0.143) (0.147) (0.195) Social capital × 1991–2001 0.026*** −0.384** −0.218 −0.251* −0.295** −0.488** (0.003) (0.153) (0.146) (0.144) (0.149) (0.196) Initial income −0.291*** −0.293*** −0.410*** −0.819*** −0.819*** −0.597*** (0.006) (0.006) (0.007) (0.009) (0.009) (0.023) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.366 0.371 0.437 0.618 0.618 0.511 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.773*** 1.391*** 0.709*** 0.548*** 0.573*** 1.055*** (0.035) (0.144) (0.141) (0.161) (0.162) (0.195) Social capital × 1961–1971 0.034*** −0.562*** −0.544*** −0.421*** −0.451*** −0.672*** (0.002) (0.167) (0.159) (0.146) (0.148) (0.218) Social capital × 1971–1981 0.049*** −0.416*** −0.326** −0.268* −0.304** −0.515*** (0.002) (0.160) (0.152) (0.145) (0.148) (0.198) Social capital × 1981–1991 0.040*** −0.983*** −0.777*** −0.683*** −0.720*** −0.966*** (0.003) (0.153) (0.146) (0.143) (0.147) (0.195) Social capital × 1991–2001 0.026*** −0.384** −0.218 −0.251* −0.295** −0.488** (0.003) (0.153) (0.146) (0.144) (0.149) (0.196) Initial income −0.291*** −0.293*** −0.410*** −0.819*** −0.819*** −0.597*** (0.006) (0.006) (0.007) (0.009) (0.009) (0.023) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.366 0.371 0.437 0.618 0.618 0.511 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Figure 4 provides a clearer picture. The average impact of social capital on growth hides some heterogeneity over time. The role of social capital is bigger immediately after WWII when its level was higher (Figure 2). This result is consistent with the one for the macro areas (South vs. Centre–North). In short, we find that the return to social capital is higher when and where the social capital is higher. Figure 4 View largeDownload slide The impact of social capital on real growth over time. Notes: The horizontal lines are based on Table 2, Column 6. The other lines are based on Table 8, Column 6. Figure 4 View largeDownload slide The impact of social capital on real growth over time. Notes: The horizontal lines are based on Table 2, Column 6. The other lines are based on Table 8, Column 6. The fact that the elasticity of social capital is bigger in the 1950s may have several explanations. Our preferred one is in line with Ahlerup et al. (2009) who suggest that the effect of social capital on economic performance is higher when formal institutions are less developed. Indeed, since the Costituzione della Repubblica italiana was enforced in 1948 and radically changed the Italian formal institutional system, it is reasonable to think that Italy was a country under construction during the 1950s—which explains why social capital may have played a more important role. Of course, additional factors may be at work. For instance, a higher quantity of social capital may imply a higher quality of social capital, and thus higher productivity. Such a mechanism would be in line with what some authors (among others, see Hungerford and Solon, 1987; Budría and Pereira, 2011) have argued to hold for human capital.23 However, further investigation on this hypothesis requires municipality-level measures of social–capital quality which, to our knowledge, are not available. 5.4. Different outcomes As a further robustness check, we also estimate GMM-SYS models with different outcomes using municipality-level data. In particular, Table 8 considers the growth rates of the following variables as additional outcomes24: Plant density (plants over municipality square kilometres) (Column 2). Employees density (employees over municipality square kilometres) (Column 3). Employees per capita (employees over population) (Column 4). Table 8 Conditional mean estimates on different outcomes: changing the dependent variable (measured as growth rate) (1) (2) (3) (4) GMM-SYS GMM-SYS GMM-SYS GMM-SYS Real GDP per capita Plant density Employees density Employees per capita Social capital 0.595*** 0.477** 0.522* 0.721*** (0.109) (0.223) (0.284) (0.230) Initial income −0.660*** −0.112*** −0.0853 −1.530*** (0.0247) (0.0369) (0.0538) (0.0679) Observations 10,145 10,145 10,145 10,145 G. of fit 0.407 0.195 0.243 0.290 (1) (2) (3) (4) GMM-SYS GMM-SYS GMM-SYS GMM-SYS Real GDP per capita Plant density Employees density Employees per capita Social capital 0.595*** 0.477** 0.522* 0.721*** (0.109) (0.223) (0.284) (0.230) Initial income −0.660*** −0.112*** −0.0853 −1.530*** (0.0247) (0.0369) (0.0538) (0.0679) Observations 10,145 10,145 10,145 10,145 G. of fit 0.407 0.195 0.243 0.290 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 8 Conditional mean estimates on different outcomes: changing the dependent variable (measured as growth rate) (1) (2) (3) (4) GMM-SYS GMM-SYS GMM-SYS GMM-SYS Real GDP per capita Plant density Employees density Employees per capita Social capital 0.595*** 0.477** 0.522* 0.721*** (0.109) (0.223) (0.284) (0.230) Initial income −0.660*** −0.112*** −0.0853 −1.530*** (0.0247) (0.0369) (0.0538) (0.0679) Observations 10,145 10,145 10,145 10,145 G. of fit 0.407 0.195 0.243 0.290 (1) (2) (3) (4) GMM-SYS GMM-SYS GMM-SYS GMM-SYS Real GDP per capita Plant density Employees density Employees per capita Social capital 0.595*** 0.477** 0.522* 0.721*** (0.109) (0.223) (0.284) (0.230) Initial income −0.660*** −0.112*** −0.0853 −1.530*** (0.0247) (0.0369) (0.0538) (0.0679) Observations 10,145 10,145 10,145 10,145 G. of fit 0.407 0.195 0.243 0.290 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. The estimates are compared with our main GMM-SYS results (Column 1), which are based on the growth rate of real GDP per capita as outcome. The results suggest that the effect of social capital changes with the outcome chosen, but it is generally positive and significant, as one would reasonably expect. 5.5. Sample selection We work with a sample of municipalities with at least 5000 inhabitants in 1951 because identification is problematic with the full sample. The results obtained with the full sample are provided in the Appendix Table A5. Even though the general picture is very close to the one obtained with the sample of municipalities with at least 5000 inhabitants in 1951, estimates in Column 6 pass neither the Arellano–Bond test nor the Hansen and the Difference-in-Hansen tests. In particular, we find a lower return to social capital than in the restricted sample (both coefficients are significant). There are two possible explanations for this difference in returns. One is that restricted sample estimates are not reliable because of sample selection bias (Hypothesis I). Another one is that full sample estimates are not reliable because of lack of model identification (Hypothesis II). As it is known, sample selection bias occurs when the probability of being in the sample, i.e., the probability of the municipality population being higher than 5000 inhabitants in 1951, is affected by unobservables that are correlated with the unobservables of the growth equation. Here, the role of housing prices is key. Housing prices in 1951 are unobservables in our dataset, and this may create a correlation between the residuals of the growth equation and those of the ‘participation equation’ (i.e., the equation modelling the probability of being in the sample). Housing prices in 1951 are likely to affect population size in 1951 and thus the probability of being in the sample. Thus, if the growth return to social capital is responsive to population size in 1951 in the full sample, this may be an indication of selection bias in the restricted sample. We test the sample-selection bias hypothesis (Hypothesis I) in Table 9 and reject it because the interaction term (social capital × population in 1951) is not significant. This reinforces our view that using the restricted sample is preferable because identification tests are passed in the restricted sample, while they are passed neither in the full sample nor in the sample of municipalities with less than 5000 inhabitants in 1951. Table 9 Conditional mean estimates on growth changing the sample size (1) (2) (3) GMM-SYS GMM-SYS GMM-SYS Full sample Below 5000 Above 5000 Social capital 0.522*** 0.254** 0.558*** (0.0748) (0.104) (0.102) Initial income −0.766*** −0.811*** −0.654*** (0.0184) (0.0197) (0.0239) Social capital × Population in 1951 −0.0127 0.0132 0.00579 (0.00923) (0.0119) (0.00542) Observations 40,185 30,040 10,145 G. of fit 0.398 0.393 0.408 (1) (2) (3) GMM-SYS GMM-SYS GMM-SYS Full sample Below 5000 Above 5000 Social capital 0.522*** 0.254** 0.558*** (0.0748) (0.104) (0.102) Initial income −0.766*** −0.811*** −0.654*** (0.0184) (0.0197) (0.0239) Social capital × Population in 1951 −0.0127 0.0132 0.00579 (0.00923) (0.0119) (0.00542) Observations 40,185 30,040 10,145 G. of fit 0.398 0.393 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 9 Conditional mean estimates on growth changing the sample size (1) (2) (3) GMM-SYS GMM-SYS GMM-SYS Full sample Below 5000 Above 5000 Social capital 0.522*** 0.254** 0.558*** (0.0748) (0.104) (0.102) Initial income −0.766*** −0.811*** −0.654*** (0.0184) (0.0197) (0.0239) Social capital × Population in 1951 −0.0127 0.0132 0.00579 (0.00923) (0.0119) (0.00542) Observations 40,185 30,040 10,145 G. of fit 0.398 0.393 0.408 (1) (2) (3) GMM-SYS GMM-SYS GMM-SYS Full sample Below 5000 Above 5000 Social capital 0.522*** 0.254** 0.558*** (0.0748) (0.104) (0.102) Initial income −0.766*** −0.811*** −0.654*** (0.0184) (0.0197) (0.0239) Social capital × Population in 1951 −0.0127 0.0132 0.00579 (0.00923) (0.0119) (0.00542) Observations 40,185 30,040 10,145 G. of fit 0.398 0.393 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Overall, disregarding municipalities with less than 5000 inhabitants in 1951 does not appear to be a major issue because about 77% of the Italian population is concentrated, on average, in the sample of cities that we use (a minimum of 74% occurs in 1951). In addition, earlier studies, including Guiso et al. (2016), have used municipalities with more than 5000 inhabitants, which makes our restricted sample results more comparable with the previous ones. 5.6. Province-time effects In Table 10, we replicate Table 2 with the only difference being the introduction of additional controls in the form of (province × time) interactions. We find that the coefficient of social capital slightly increases from 0.595 to 0.650 in the main (GMM-SYS) specification. However, the introduction of these additional covariates implies a substantial increase in the number of instruments, which weakens the identification tests, in line with Roodman (2009), even after collapsing the instrument matrix. In addition, the standard error goes from 0.109 to 0.309, with a clear loss of efficiency. Table 10 Conditional mean estimates on growth controlling for province-time effects (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.464*** 0.465*** 0.347*** 0.181** 0.171** 0.650** (0.0480) (0.0478) (0.0484) (0.0758) (0.0770) (0.309) Initial income −0.364*** −0.367*** −0.426*** −0.897*** −0.897*** −0.747*** (0.00739) (0.00729) (0.00802) (0.0107) (0.0107) (0.0502) Province-time effects Yes Yes Yes Yes Yes Yes Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.495 0.499 0.513 0.667 0.667 0.389 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.464*** 0.465*** 0.347*** 0.181** 0.171** 0.650** (0.0480) (0.0478) (0.0484) (0.0758) (0.0770) (0.309) Initial income −0.364*** −0.367*** −0.426*** −0.897*** −0.897*** −0.747*** (0.00739) (0.00729) (0.00802) (0.0107) (0.0107) (0.0502) Province-time effects Yes Yes Yes Yes Yes Yes Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.495 0.499 0.513 0.667 0.667 0.389 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 10 Conditional mean estimates on growth controlling for province-time effects (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.464*** 0.465*** 0.347*** 0.181** 0.171** 0.650** (0.0480) (0.0478) (0.0484) (0.0758) (0.0770) (0.309) Initial income −0.364*** −0.367*** −0.426*** −0.897*** −0.897*** −0.747*** (0.00739) (0.00729) (0.00802) (0.0107) (0.0107) (0.0502) Province-time effects Yes Yes Yes Yes Yes Yes Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.495 0.499 0.513 0.667 0.667 0.389 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.464*** 0.465*** 0.347*** 0.181** 0.171** 0.650** (0.0480) (0.0478) (0.0484) (0.0758) (0.0770) (0.309) Initial income −0.364*** −0.367*** −0.426*** −0.897*** −0.897*** −0.747*** (0.00739) (0.00729) (0.00802) (0.0107) (0.0107) (0.0502) Province-time effects Yes Yes Yes Yes Yes Yes Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.495 0.499 0.513 0.667 0.667 0.389 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. 5.7. Physical and human capital As another check, we control for a proxy of physical capital endowment (Table 11). The latter is measured as the number of plants per inhabitant at the beginning of the period (i.e., at year t−10). Although this variable is a candidate for the main specification, we find the coefficient of physical capital to be close to zero from the left (−0.051) and not significant (p-value = 0.230). As a result, the coefficient of social capital does not change much. One possible explanation is that controlling for the initial income level already captures the effect of the initial physical capital stock. Table 11 Conditional mean estimates on growth controlling for physical capital endowment (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.842*** 0.715*** 0.278*** 0.184*** 0.125* 0.585*** (0.030) (0.035) (0.039) (0.067) (0.068) (0.117) Initial income −0.334*** −0.319*** −0.434*** −0.929*** −0.933*** −0.639*** (0.006) (0.008) (0.008) (0.011) (0.011) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.378 0.392 0.447 0.684 0.685 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.842*** 0.715*** 0.278*** 0.184*** 0.125* 0.585*** (0.030) (0.035) (0.039) (0.067) (0.068) (0.117) Initial income −0.334*** −0.319*** −0.434*** −0.929*** −0.933*** −0.639*** (0.006) (0.008) (0.008) (0.011) (0.011) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.378 0.392 0.447 0.684 0.685 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 11 Conditional mean estimates on growth controlling for physical capital endowment (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.842*** 0.715*** 0.278*** 0.184*** 0.125* 0.585*** (0.030) (0.035) (0.039) (0.067) (0.068) (0.117) Initial income −0.334*** −0.319*** −0.434*** −0.929*** −0.933*** −0.639*** (0.006) (0.008) (0.008) (0.011) (0.011) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.378 0.392 0.447 0.684 0.685 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.842*** 0.715*** 0.278*** 0.184*** 0.125* 0.585*** (0.030) (0.035) (0.039) (0.067) (0.068) (0.117) Initial income −0.334*** −0.319*** −0.434*** −0.929*** −0.933*** −0.639*** (0.006) (0.008) (0.008) (0.011) (0.011) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.378 0.392 0.447 0.684 0.685 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. In addition, we control for the roles that both physical and human capital endowments (at year t−10) as well as the evolution of working age population (from year t−10 to year t) may play in a growth regression (Table 12). As proxies for human capital endowments, we use both the share of population with high-school diploma and the share of population with university degree. The evolution of the working age population is the growth rate of the population aged between 15 and 64 years. These variables should, in principle, be included among the controls in the main specification.25 However, municipality-level data on both human capital and working age population are available in electronic format only since 1971. Hence, this exercise is restricted to a shorter time period. Nevertheless, the positive link between social capital and growth is confirmed and the magnitude of the social capital coefficient decreases. This is consistent with the idea that the effect of social capital on growth is higher in the 1950s. Table 12 Conditional mean estimates on growth controlling for physical and human capital endowments as well as the evolution of the working age population (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.772*** 0.502*** 0.162*** 0.013 −0.023 0.481*** (0.032) (0.036) (0.039) (0.075) (0.076) (0.14) Initial income −0.323*** −0.323*** −0.416*** −1.049*** −1.051*** −0.799*** (0.010) (0.009) (0.010) (0.014) (0.014) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 6087 6087 6087 6087 6087 6087 R-squared/G. of fit 0.332 0.362 0.412 0.716 0.717 0.411 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.772*** 0.502*** 0.162*** 0.013 −0.023 0.481*** (0.032) (0.036) (0.039) (0.075) (0.076) (0.14) Initial income −0.323*** −0.323*** −0.416*** −1.049*** −1.051*** −0.799*** (0.010) (0.009) (0.010) (0.014) (0.014) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 6087 6087 6087 6087 6087 6087 R-squared/G. of fit 0.332 0.362 0.412 0.716 0.717 0.411 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 12 Conditional mean estimates on growth controlling for physical and human capital endowments as well as the evolution of the working age population (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.772*** 0.502*** 0.162*** 0.013 −0.023 0.481*** (0.032) (0.036) (0.039) (0.075) (0.076) (0.14) Initial income −0.323*** −0.323*** −0.416*** −1.049*** −1.051*** −0.799*** (0.010) (0.009) (0.010) (0.014) (0.014) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 6087 6087 6087 6087 6087 6087 R-squared/G. of fit 0.332 0.362 0.412 0.716 0.717 0.411 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.772*** 0.502*** 0.162*** 0.013 −0.023 0.481*** (0.032) (0.036) (0.039) (0.075) (0.076) (0.14) Initial income −0.323*** −0.323*** −0.416*** −1.049*** −1.051*** −0.799*** (0.010) (0.009) (0.010) (0.014) (0.014) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 6087 6087 6087 6087 6087 6087 R-squared/G. of fit 0.332 0.362 0.412 0.716 0.717 0.411 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. 5.8. Summing up Summing up, our analysis suggests the following picture. The link between initial social capital endowment and subsequent growth at municipality level is positive and significant. The result does not change when controlling for spatial externalities, when accounting for non-linearities in initial income and population, when accounting for province-time effects, when accounting for physical capital endowment and when accounting for both physical and human capital endowment of a city as well as for the evolution of the working age population. The results for different outcomes are consistent with those for the main outcome. However, the strength of the link between social capital and growth looks stronger in the Centre–North of Italy and in the 1950s. 6. Local estimates GMM-SYS uses lagged levels and differences of endogenous variables as instruments. These are called internal instruments. Yet, the identification based on internal instruments, while solving a number of statistical problems (including the Nickell bias in dynamic panel-data models, such as our model), is not appealing from a theoretical perspective because the causal chain between the instruments and the endogenous variables is much more opaque than in a 2SLS just-identified model. To address this criticism, this section proposes 2SLS estimates from two just-identified models, which are identified using two different municipality-level external instruments. In defense of our GMM-SYS approach, it should be stressed that 2SLS estimates have low external validity because they refer to the specific group of municipalities whose level of social capital was affected by the instrument. This is why 2SLS estimates are ‘local estimates’ in the spirit of Angrist and Pischke (2009). In contrast, GMM-SYS estimates are more likely to have external validity because all municipalities in the sample are affected by the instruments by construction (the instruments are internal). Even if our GMM-SYS estimates refer to a restricted sample (more than 5000 inhabitants in 1951), they are still of interest for a large, well-defined sub-population. This is generally not the case when estimating local effects with 2SLS. With the above caveats, our first instrument, inspired by Guiso et al. (2016), is a dummy equal to one if a municipality was a free city in 1300.26 The second instrument, inspired by Andini et al. (2017), is a dummy equal to one if the municipality was an incorporating city (i.e., a larger municipality incorporating a smaller one), as a result of a consolidation decision taken by Mussolini during the fascist dictatorship, in the 1920s. An instrument must be valid and relevant. In our 2SLS context, validity means that the instrument must be uncorrelated with the error term of the growth equation. This is like saying that, conditional on other regressors, the instrument must not affect growth directly (exclusion restriction; see also Bürker and Minerva, 2014, 830). To be relevant, the instrument must be (not weakly) correlated with the endogenous variable, i.e., voter turnout. As is well known, while the relevance condition can be theoretically argued and empirically tested, the validity of an instrument can only be theoretically argued. The main argument for validity here is that both historical events, on which our instruments are based, are too distant in the past to have a direct effect on growth from 1951 onwards, once we condition on initial income (see also Tabellini, 2010; Crescenzi et al., 2013; Bürker and Minerva, 2014). The idea behind the first-stage regression in Column (1) of Table 13 is well grounded in the work of Guiso et al. (2016), which highlights the transmission across generations of inhabitants, over long periods of time, of civic-capital characteristics (see also Bürker and Minerva, 2014). Following Guiso et al. (2016), we restrict the sample to the Centre–North due to the Norman domination in the South of Italy. The estimated coefficient is positive (0.042) and significant at 1% level, as expected. The relevance condition is supported by the first-stage F-statistic, which is well above the usual reference values. The latter is robust to the control set (Column 2), though the F-statistic decreases a lot. Table 13 Conditional mean estimates on social capital: first-stage regressions (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.0420*** 0.0185*** 0.0211*** 0.0161*** (0.00444) (0.00558) (0.00255) (0.00255) Initial income 0.0299*** 0.0260*** 0.0297*** 0.0253*** (0.00141) (0.00139) (0.00143) (0.00141) Constant 4.301*** 4.121*** 4.301*** 4.123*** (0.00933) (0.0169) (0.00939) (0.0168) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.050 0.624 0.050 0.621 F-statistic 89.47 10.99 68.57 39.67 (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.0420*** 0.0185*** 0.0211*** 0.0161*** (0.00444) (0.00558) (0.00255) (0.00255) Initial income 0.0299*** 0.0260*** 0.0297*** 0.0253*** (0.00141) (0.00139) (0.00143) (0.00141) Constant 4.301*** 4.121*** 4.301*** 4.123*** (0.00933) (0.0169) (0.00939) (0.0168) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.050 0.624 0.050 0.621 F-statistic 89.47 10.99 68.57 39.67 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 13 Conditional mean estimates on social capital: first-stage regressions (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.0420*** 0.0185*** 0.0211*** 0.0161*** (0.00444) (0.00558) (0.00255) (0.00255) Initial income 0.0299*** 0.0260*** 0.0297*** 0.0253*** (0.00141) (0.00139) (0.00143) (0.00141) Constant 4.301*** 4.121*** 4.301*** 4.123*** (0.00933) (0.0169) (0.00939) (0.0168) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.050 0.624 0.050 0.621 F-statistic 89.47 10.99 68.57 39.67 (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.0420*** 0.0185*** 0.0211*** 0.0161*** (0.00444) (0.00558) (0.00255) (0.00255) Initial income 0.0299*** 0.0260*** 0.0297*** 0.0253*** (0.00141) (0.00139) (0.00143) (0.00141) Constant 4.301*** 4.121*** 4.301*** 4.123*** (0.00933) (0.0169) (0.00939) (0.0168) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.050 0.624 0.050 0.621 F-statistic 89.47 10.99 68.57 39.67 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. The idea behind the first-stage regression in Column (3) of Table 13 requires a few more explanations. Andini et al. (2017) use the fascist consolidation of small municipalities (less than 5000 inhabitants) which were merged each other into a new born municipality to investigate how a substantial enlargement of small-sized cities affects the local welfare of inhabitants. Differently from Andini et al. (2017), we use the fascist consolidation of small municipalities which were merged to an already existing bigger municipality (more than 5000 inhabitants) to investigate how a small enlargement of already existing bigger municipalities changes the level of social capital in bigger cities (this is our first-stage regression in Column 3). Thus, we use the same historical event as Andini et al. (2017) but consider a different group of municipalities, which does not overlap with earlier work. The key argument for the relevance of our second instrument is that the level of social capital in a municipality, measured as voter turnout, is responsive to the fact that the municipality was an incorporating city in the past. This can happen for two reasons. The first one is mechanical: a consolidation is likely to change both the denominator (eligible voters) and the numerator (actual voters) of the turnout rate, both at the time it happens and from that moment onwards. The second one is theoretical: a consolidation is likely to increase the propensity to participate in democratic processes for people (the inhabitants of incorporating cities) who have been exposed to an additional dictator imposition (relative to the non-exposed, i.e., the inhabitants of non-incorporating cities). Following Guiso et al. (2016), such a characteristic is likely to be transmitted across generations of inhabitants. The estimated coefficient is positive (0.021) and significant at 1% level. The F-statistic ensures that the instrument is relevant. Notably, the result is robust to controlling, among other covariates, for population and surface (Column 4), with the implication that the status of incorporating city has a direct positive effect on voter turnout, independent of population and surface. Reduced-form estimates are provided in Table 14, reporting how each of the instruments affects the growth rate of GDP per capita through civic capital. The estimated coefficients are all positive and significant. The results in Columns (1) and (2) extend the work of Guiso et al. (2016) in an interesting direction. The findings in Columns (3) and (4) are in line with the evidence of Andini et al. (2017) that Mussolini’s consolidation has been welfare-enhancing, though our results are not strictly comparable with those of Andini et al. (2017) for the reasons explained above. Table 14 Conditional mean estimates on growth: reduced-form regressions (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.150*** 0.0697*** 0.0617*** 0.0468*** (0.0135) (0.0139) (0.00837) (0.00829) Initial income −0.280*** −0.286*** −0.279*** −0.288*** (0.00563) (0.00574) (0.00564) (0.00579) Constant 2.102*** 1.919*** 2.096*** 1.922*** (0.0363) (0.0496) (0.0362) (0.0492) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.318 0.469 0.317 0.469 (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.150*** 0.0697*** 0.0617*** 0.0468*** (0.0135) (0.0139) (0.00837) (0.00829) Initial income −0.280*** −0.286*** −0.279*** −0.288*** (0.00563) (0.00574) (0.00564) (0.00579) Constant 2.102*** 1.919*** 2.096*** 1.922*** (0.0363) (0.0496) (0.0362) (0.0492) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.318 0.469 0.317 0.469 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 14 Conditional mean estimates on growth: reduced-form regressions (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.150*** 0.0697*** 0.0617*** 0.0468*** (0.0135) (0.0139) (0.00837) (0.00829) Initial income −0.280*** −0.286*** −0.279*** −0.288*** (0.00563) (0.00574) (0.00564) (0.00579) Constant 2.102*** 1.919*** 2.096*** 1.922*** (0.0363) (0.0496) (0.0362) (0.0492) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.318 0.469 0.317 0.469 (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.150*** 0.0697*** 0.0617*** 0.0468*** (0.0135) (0.0139) (0.00837) (0.00829) Initial income −0.280*** −0.286*** −0.279*** −0.288*** (0.00563) (0.00574) (0.00564) (0.00579) Constant 2.102*** 1.919*** 2.096*** 1.922*** (0.0363) (0.0496) (0.0362) (0.0492) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.318 0.469 0.317 0.469 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. As shown in Table 15, the general result that civic capital has a positive and significant effect on growth is confirmed by our 2SLS experiments, even after extending the conditioning set of covariates. Table 15 Conditional mean estimates on growth: second-stage regressions (1) (2) (3) (4) 2SLS 2SLS 2SLS 2SLS Guiso et al. Guiso et al. Andini et al. Andini et al. Social capital 3.562*** 3.765*** 2.924*** 2.911*** (0.432) (1.214) (0.433) (0.570) Initial income −0.386*** −0.384*** −0.366*** −0.362*** (0.0155) (0.0327) (0.0153) (0.0165) Constant −13.22*** −13.60*** −10.48*** −10.08*** (1.851) (4.991) (1.857) (2.343) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 (1) (2) (3) (4) 2SLS 2SLS 2SLS 2SLS Guiso et al. Guiso et al. Andini et al. Andini et al. Social capital 3.562*** 3.765*** 2.924*** 2.911*** (0.432) (1.214) (0.433) (0.570) Initial income −0.386*** −0.384*** −0.366*** −0.362*** (0.0155) (0.0327) (0.0153) (0.0165) Constant −13.22*** −13.60*** −10.48*** −10.08*** (1.851) (4.991) (1.857) (2.343) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 15 Conditional mean estimates on growth: second-stage regressions (1) (2) (3) (4) 2SLS 2SLS 2SLS 2SLS Guiso et al. Guiso et al. Andini et al. Andini et al. Social capital 3.562*** 3.765*** 2.924*** 2.911*** (0.432) (1.214) (0.433) (0.570) Initial income −0.386*** −0.384*** −0.366*** −0.362*** (0.0155) (0.0327) (0.0153) (0.0165) Constant −13.22*** −13.60*** −10.48*** −10.08*** (1.851) (4.991) (1.857) (2.343) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 (1) (2) (3) (4) 2SLS 2SLS 2SLS 2SLS Guiso et al. Guiso et al. Andini et al. Andini et al. Social capital 3.562*** 3.765*** 2.924*** 2.911*** (0.432) (1.214) (0.433) (0.570) Initial income −0.386*** −0.384*** −0.366*** −0.362*** (0.0155) (0.0327) (0.0153) (0.0165) Constant −13.22*** −13.60*** −10.48*** −10.08*** (1.851) (4.991) (1.857) (2.343) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. It is worth noting that 2SLS estimates based on different instruments refer to different local sub-populations and, therefore, these estimates do not need to be similar to each other27 and, most importantly, they do not need to be similar to our main GMM-SYS estimate (0.595), even if exactly the same growth model is estimated. This is in line with our results. 7. Conclusions In this paper, we have investigated the causal link between social capital and economic growth at a very detailed level of aggregation, the municipality. By using data on Italian municipalities over the 1951–2001 period, we have presented evidence of a positive and significant link between social capital endowment and subsequent economic growth. In particular, accounting for the endogeneity of all the explanatory variables as well as for unobserved heterogeneity, we have argued that the elasticity of social capital endowment with respect to growth is equal to 0.6, on average. The result has been shown to be robust to a variety of checks. We have also found that the link between social capital and growth is stronger in the Centre–North of Italy, where the elasticity is 0.7. Considering the time dimension, the link is found to be stronger in the 1950s, with an elasticity of 1.0. For specific sub-populations of municipalities, the elasticity is much bigger, with values ranging between 2.9 and 3.7. One limitation of this paper is that we have used only one measure of social capital and it is an indirect measure. Although the panel-fixed effects approach partly controls for measurement errors, future research could go in the direction of replicating our analysis using different measures of social capital and, in particular, direct measures of social capital at municipality level. Future research may also take advantage of direct measures of per capita GDP growth at municipality level, when they will become available for Italy. Another limitation is that we have implicitly assumed that the effect of social capital on growth is constant along the conditional growth distribution. The use of panel quantile regression techniques may shed light on the existence of quantile parameter heterogeneity (see Deng et al., 2012; Andini and Andini, 2014). This is important because social capital, besides shifting the mean of the conditional growth distribution towards the right, thus being a source of between-groups growth inequality, may be a factor changing the dispersion of the conditional growth distribution, thus being a source of within-groups growth inequality or a brake on it. Due to data limitations, we have not entered the debate about the quality and the quantity of social capital (Eliasson et al., 2013) as well as the issue of the heterogeneity between rural and urban areas (Westlund et al., 2014). Future studies on these topics are warranted. Another interesting topic for future research is that of testing the hypothesis that Italy is a special case (Westlund and Adam, 2010), by replicating our analysis using municipality data from other countries. It is possible that the use of appropriate municipality-level data brings out the underlying link between social capital and growth which is unclear in analyses based on national or regional-level data for countries other than Italy. Nevertheless, if one believes in the argument that the effect of social capital on economic performance is higher when and where the level of social capital is higher, it is likely that differences in the levels of social capital among countries would lead to differences in returns. So, our empirical results cannot be generalised. What can instead be generalised is the econometric approach to the issue that we have presented here. One reason why we look at growth as outcome is that, in our main theoretical frame, social capital affects non-measurable or hardly-measurable ‘intermediate’ outcomes such as trade uncertainty and free-riding behaviour making the production activity more efficient (Routledge and von Amsberg 2003). Yet, in a different theoretical frame such as a model by Akçomak and ter Weel (2009), social capital affects growth by increasing innovation. Future research may test this hypothesis by using specific measures of innovation at municipality level as dependent variables, corroborating the existing evidence on firm-level outcomes (Laursen et al., 2012). Following Sabatini (2008), a further interesting topic for the research agenda is the effect of social capital on broader measures of economic development, capturing the quality of economic development. Whether the link exists at the municipality level, it is an open issue. Footnotes 1 Knack and Keefer (1997) also argue that both trust and civic involvement improve the quality of governance, which in turn raises the quality of public-provided education and the return to human capital. Zak and Knack (2001) show that, in a context of moral hazard where more resources are allocated to inspection and monitoring, both the return to physical capital investment and the rate of investment decrease. Akçomak and ter Weel (2009) argue that the willingness to invest in innovation is higher when social capital is higher. Dasgupta (2011) shows that an increase in trust among members of a group will manifest itself in a total factor productivity growth. Papagapitos and Riley (2009) argue that higher trust leads to higher stocks of physical capital, which increase the productivity of human capital. Carlin et al. (2009) show that, in societies with no complete contracts, no punishment for opportunistic behaviour and/or no frequent interactions, more regulation results in less trust, which depresses aggregate investment and growth. However, Routledge and von Amsberg (2003) show that, in a model when the frequency of trade influences agent decisions to trade cooperatively, a reduction in social capital may have ambiguous effects on welfare. 2 We are aware of only two earlier studies that explore the link between social capital and growth using municipality-level data: Corazzini et al. (2011) and Eliasson et al. (2013). Yet, these works focus on correlation rather than causation and the evidence presented is cross-sectional. 3 We use the words municipality and city as referring to the same territorial unit. 4 The authors attribute these findings, in part, to the fact that the measures of social capital used (typically trust and association membership/density) are hardly able to capture the causal nexus from social capital to economic development when the analysis refers to spatial units characterised by a large number of unrelated actors, such as nations or regions. And, they conclude that it is crucial for future research to develop new and better measures of social capital at national and regional level, for instance for values like creativity, entrepreneurship and tolerance. 5 Laursen et al. (2012) suggest that social capital, measured as social interaction and political participation, increases firm ability to innovate as well as internal investments in R&D. Analogously, Guiso et al. (2004) show that social capital affects the behaviour of households and individuals, in the direction of boosting financial development. 6 Albanese and de Blasio (2014) use municipality-level data to study the impact of social capital on proxies of economic development (employment rate, employment density and plant density) in Italy. The authors estimate a static panel-data model with fixed effects, assuming social capital to be exogenous, and a number of cross-sectional models dealing with the endogeneity of social capital only. They find that social capital is beneficial for development. Westlund et al. (2014) investigate the influence of local entrepreneurial social capital on start-ups per capita, where the former is measured as the firm perception of local public attitudes to entrepreneurship and as the share of small businesses. It is found that both measures positively influence the start-up propensities among Swedish municipalities. The effect is stronger in rural areas. Analogously, Eliasson et al. (2013) report a positive relationship between enterprises’ opinion on the quality of the local business-related social capital (the local business climate) and economic growth at municipality level in Sweden. The authors also suggest that the quality of social capital may be more important than the quantity. A similar approach is taken by Corazzini et al. (2011) who use several measures of social cohesion, social division and attitude towards religion, estimating a growth model with data on Brazilian municipalities. Overall, they suggest a positive influence of social-capital indicators on growth. Finally, Rutten and Gelissen (2010) develop a theory of how social values influence economic development, which is tested in European regions. The evidence highlights that social values matter for subgroups in the population rather than the region as a whole. 7 For instance, the municipality issues the so-called tessera elettorale, which is the document needed for voting; citizens can only vote in the municipality of residence; the municipality chooses the specific infrastructure (typically a secondary school) where each citizen of the municipality is invited to vote; the municipality defines the composition of the committee that supervises the voting operations and enforces the application of the law; this committee certifies the identity of the voters in the day of the elections and provides them with the so-called scheda elettorale, which is the document where the voters are invited to express their preferences; this committee makes the counting of the votes and so on. 8 Some municipalities have been part of different provinces in different decades. We take this issue into account when we construct the value added. 9 To provide an example, suppose a province has two cities. Consider the manufacturing sector. For simplicity, assume the labour supply in each city is inelastic and both cities have the same labour supply. The labour-demand schedule in city 2 is above that of city 1 because of some labour-demand shifter (e.g., a better quality of machineries). The equilibrium wage in city 2 is thus initially bigger than in city 1. In this situation, workers would move from city 1 to city 2 (within a province, labour mobility is likely to be high because people can work in one city and be resident in another one). The decrease in the labour supply in city 1 would increase the wage in city 1. The increase in the labour supply in city 2 would decrease the wage in city 2. The flow of workers will only stop when the manufacturing wage in the two cities will be the same. This would imply the same labour productivity in the two cities. Thus, the productivity in each city will be the same as that in the province (the latter is the average of the productivities in the two cities), no matter whether labour demand conditions differ. A similar argument can be made if labour supply conditions differ. 10 Some attribute the real origin of the term to L.J. Hanifan, a social reformer, in 1916. 11 As suggested by Rutten et al. (2010), there are two main approaches to social capital: the ‘structuralist’ and the ‘interactionist’. The first investigates how network structures affect the performance of networks and of individuals within those networks. The second looks at the connection between association and obligation as well as the impact of this connection on how individuals behave (being associated to other individuals implies the obligation of sharing some form of ‘morality’, i.e., of norms, values, trust, rules, beliefs and codes). 12 As put by Sabatini (2008, 469), ‘repeated interactions among group members foster the diffusion of information and raise reputations’ relevance. The higher opportunity cost of free-riding in prisoners’ dilemma kind of situations makes agents’ behaviour more foreseeable causing an overall reduction of uncertainty. Therefore, an increase in trust-based relations may reduce the average cost of transactions, just as an increase in physical capital reduces the average cost of production’. 13 Of all the referenda occurred in Italy between 1946 and 2000, only two did not have the 50% cutoff because they did not involve the abrogation of an existing law (Monarchy vs. Republic in 1946; European Parliament in 1989). 14 As Keynes would put it, ‘It is not a case of choosing those [politicians] which, to the best of one’s judgment, are really the prettiest, nor even those that average opinion genuinely thinks the prettiest. We have reached the third degree where we devote our intelligences to anticipating what average opinion expects the average opinion to be. And there are some, I believe, who practice the fourth, fifth and higher degrees.’ (Keynes, 1936, 156). 15 Our proxy cannot be intended as an alternative to the ‘ideal’ measures proposed by Westlund and Adam (2010). 16 We measure a type of social capital that is likely to be losing importance (indeed, we will document that its return was higher in the 1950s). In the knowledge economy, other types of social capital (more business related) are likely to be gaining importance. But we cannot test this hypothesis because of lack of data. 17 We must acknowledge that our measure of social capital is not perfect, mainly because it measures ‘one type’ of social capital. 18 On the lines of Audretsch and Keilbach (2004), among others, we look at civic capital, proxied by voter turnout, as an additional component of a production function. The empirical model proposed in this paper is embedded in the standard endogenous growth theory (Eliasson et al., 2013) where cooperation among agents matters for growth. Our main theoretical reference is an article by Routledge and von Amsberg (2003) where social capital affects growth by reducing trade uncertainty and free-riding behaviour, thus making production activity more efficient. Another example is a model by Cozzi (1999) where R&D cooperation among firms is positive for growth under some assumptions about the state of technology. A recent example is a model by Akçomak and ter Weel (2009) where social capital affects growth by fostering innovation. 19 As argued by Sforza (2014), turnout significantly decreases if it is raining in the day of the elections. 20 The estimated coefficients of the additional control variables, not reported in Table 2, generally have the expected sign. 21 For instance, Durlauf et al. (2001) estimate a Solow's (1956) growth model in which the parameters vary according to the initial income level of a country, finding evidence of heterogeneity along that dimension. Contributions on the heterogeneity of the link between social capital and growth are quite few. For instance, Knack and Keefer (1997) include in their regression the interaction term ‘trust multiplied by initial income’ and find a negative coefficient. This implies that the effects of trust on growth might be stronger for poorer economies. Dearmon and Grier (2009) divide their sample in poorest and richest countries (respectively, the 25th and 75th percentiles of the income distribution) and show that there is no differential impact of social capital on growth between the two sub-samples. Ahlerup et al. (2009) find that the effect of social capital on economic performance is positive and decreases as formal institutions develop. On the same lines, Mauro and Pigliaru (2011) show that the strength of social capital as a determinant of long-run growth may depend on the institutional context. 22 Table 7 reports the p-values of the coefficients of all the interactions. Since all these coefficients are different from zero and negative, all the social-capital returns for the periods after 1951–1961 are different from the social-capital return in the period 1951–1961 (the reference period in the regression model) and they are lower. Thus, the return to social capital is higher in the 1950s than in any other decade until 2001. 23 As most of the important issues in economics, also this one is somehow controversial. Indeed, some studies have argued that the return to education may decrease with the education level, in line with the standard relative scarcity theory (see Psacharopoulos, 1985; Psacharopoulos, 1994; Psacharopoulos and Patrinos, 2004). 24 These are the outcomes chosen by Albanese and de Blasio (2014). We do not use population growth as outcome because, though it is positively associated with real GDP growth (0.319), as shown by the correlation matrix in Appendix Table A4, this paper studies the determinants of the growth rate of the real GDP per capita. So, population growth is unlikely to be a good proxy for our outcome variable, as suggested by the correlation coefficient in Appendix Table A4 (−0.012). 25 The importance of human capital as a determinant of growth has been widely documented (Barro, 1991; Mankiw et al., 1992; Temple, 2001). Unfortunately, we cannot directly control for other factors such as ethnic diversity (Platteau, 1994; Knack and Keefer, 1997; Alesina et al., 1999; Florida, 2002; Ottaviano and Peri, 2006), income inequality (Alesina and Rodrik, 1994; Alesina and Perotti, 1996) and the quality of governance (Hall and Jones, 1999; Acemoglu et al., 2001; Acemoglu and Robinson, 2006). However, the average quality of formal local institutions is somehow captured by our city-fixed effects. 26 Note that the estimates by Guiso et al. (2016) refer to municipalities with more than 5000 inhabitants, like ours. 27 A higher return is found in the sub-population of municipalities whose level of social capital has been affected by the free-city experience. Since these municipalities are located in the Centre–North of Italy, the higher-return result is consistent with our GMM-SYS evidence in Table 6. Acknowledgements For useful comments and suggestions, the authors would like to thank the editor Kristian Behrens, two anonymous referees, Guglielmo Barone, Antonio Maria Conti, Guido de Blasio, Steve Fleetwood, Paolo Sestito and the participants to the second workshop on ‘Capitale sociale, comportamenti e istituzioni’ (Bank of Italy, November 2013). The views expressed in this paper are those of the authors and do not necessarily correspond to those of the institutions they are affiliated with. References Abramovitz M. ( 1956 ) Resource and output trends in the United States since 1870 . American Economic Review , 46 : 5 – 23 . Acemoglu D. , Robinson J. ( 2006 ) De facto political power and institutional persistence . American Economic Review: Papers and Proceedings , 96 : 325 – 330 . Google Scholar CrossRef Search ADS Acemoglu D. , Johnson S. , Robinson J. ( 2001 ) The colonial origins of comparative development: an empirical investigation . American Economic Review , 91 : 1369 – 1401 . Google Scholar CrossRef Search ADS Ahlerup P. , Olsson O. , Yanagizawa D. ( 2009 ) Social capital vs institutions in the growth process . European Journal of Political Economy , 25 : 1 – 14 . Google Scholar CrossRef Search ADS Akçomak S. , ter Weel B. ( 2009 ) Social capital, innovation and growth: evidence from Europe . European Economic Review , 53 : 544 – 567 . Google Scholar CrossRef Search ADS Albanese G. , de Blasio G. ( 2014 ) Civic capital and development: Italy, 1951–2001. Quaderni di Storia Economica, No. 32, Banca d’Italia. Alesina A. , Baqir R. , Easterly W. ( 1999 ) Public goods and ethnic divisions . Quarterly Journal of Economics , 114 : 1243 – 1284 . Google Scholar CrossRef Search ADS Alesina A. , Perotti R. ( 1996 ) Income distribution, political instability, and investment . European Economic Review , 40 : 1203 – 1228 . Google Scholar CrossRef Search ADS Alesina A. , Rodrik D. ( 1994 ) Distributive politics and economic growth . Quarterly Journal of Economics , 109 : 465 – 490 . Google Scholar CrossRef Search ADS Algan Y. , Cahuc P. ( 2010 ) Inherited trust and growth . American Economic Review , 100 : 2060 – 2092 . Google Scholar CrossRef Search ADS Almond G. , Verba S. ( 1963 ) The Civic Culture: Political Attitudes and Democracy in Five Nations . Boston : Little, Brown & Co . Google Scholar CrossRef Search ADS Andini M. , Andini C. ( 2014 ) Finance, growth and quantile parameter heterogeneity . Journal of Macroeconomics , 40 : 308 – 322 . Google Scholar CrossRef Search ADS Andini M. , Dalmazzo A. , de Blasio G. ( 2017 ) The size of political jurisdictions: a model with some evidence from a fascist consolidation . Journal of Comparative Economics , 45 : 889 – 909 Google Scholar CrossRef Search ADS Angrist J. D. , Pischke J. S. ( 2009 ) Mostly Harmless Econometrics: An Empiricist’s Companion . Princeton : Princeton University Press . Arellano M. , Bond S. ( 1991 ) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations . Review of Economic Studies , 58 : 277 – 297 . Google Scholar CrossRef Search ADS Arellano M. , Bover O. ( 1995 ) Another look at the instrumental variable estimation of error-components models . Journal of Econometrics , 68 : 29 – 52 . Google Scholar CrossRef Search ADS Arrighetti A. , Lasagni A. , Seravalli G. ( 2003 ) Capitale sociale, associazionismo economico e istituzioni: indicatori sintetici di sintesi . Rivista di Politica Economica , 93 : 47 – 88 . Audretsch D. , Keilbach M. ( 2004 ) Entrepreneurship capital and economic performance . Regional Studies , 38 : 949 – 959 . Google Scholar CrossRef Search ADS Barro R. ( 1991 ) Economic growth in a cross section of countries . Quarterly Journal of Economics , 106 : 407 – 443 . Google Scholar CrossRef Search ADS Bloom N. , Bond S. , Van Reenen J. ( 2007 ) Uncertainty and investment dynamics . Review of Economic Studies , 74 : 391 – 415 . Google Scholar CrossRef Search ADS Blundell R. , Bond S. ( 1998 ) Initial conditions and moment restrictions in dynamic panel data models . Journal of Econometrics , 87 : 115 – 143 . Google Scholar CrossRef Search ADS Blundell R. , Bond S. ( 2000 ) GMM estimation with persistent panel data: an application to production functions . Econometric Reviews , 19 : 321 – 340 . Google Scholar CrossRef Search ADS Blundell R. , Bond S. , Windmeijer F. ( 2000 ) Estimation in dynamic panel data models: improving the performance of the standard GMM estimator. In: Baltagi B. , Fomby T. , Carter Hill R. (eds) Nonstationary Panels, Panel Cointegration and Dynamic Panels , pp. 53 – 91 . Amsterdam : Elsevier JAI . Google Scholar CrossRef Search ADS Bond S. ( 2002 ) Dynamic panel data models: a guide to micro data methods and practice . Portuguese Economic Journal , 1 : 141 – 162 . Google Scholar CrossRef Search ADS Bond S. , Hoeffler A. , Temple J. ( 2001 ) GMM estimation of empirical growth models. CEPR Discussion Papers, No. 3048, Centre for Economic Policy Research. Bourdieu P. ( 1972 ) Outline of a Theory of Practice . Cambridge : Cambridge University Press . Budría S. , Pereira P. ( 2011 ) Educational qualifications and wage inequality: evidence for Europe . Revista de Economia Aplicada , 19 : 5 – 34 . Bürker M. , Minerva G. A. ( 2014 ) Civic capital and the size distribution of plants: short-run dynamics and long-run equilibrium . Journal of Economic Geography , 14 : 797 – 847 . Google Scholar CrossRef Search ADS Carlin B. , Dorobantu F. , Viswanathan S. ( 2009 ) Public trust, the law, and financial investment . Journal of Financial Economics , 92 : 321 – 341 . Google Scholar CrossRef Search ADS Cassese S. , Battini S. , Franchini C. , Perez R. , Vesperini G. ( 2009 ) Manuale Di Diritto Pubblico . Milano : Giuffrè . Corazzini L. , Grazzi M. , Nicolini M. ( 2011 ) Social capital and growth in Brazilian municipalities. In: Nijkamp P. , Siedschlag I. (eds) Innovation, Growth and Competitiveness , pp. 195 – 217 . Heidelberg : Springer-Verlag . Google Scholar CrossRef Search ADS Cozzi G. ( 1999 ) R&D cooperation and growth . Journal of Economic Theory , 86 : 17 – 49 . Google Scholar CrossRef Search ADS Crescenzi R. , Gagliardi L. , Percoco M. ( 2013 ) Social capital and the innovative performance of Italian provinces . Environment and Planning A , 45 : 908 – 929 . Google Scholar CrossRef Search ADS Cuturi V. , Sampugnaro R. , Tomaselli V. ( 2000 ) L’elettore Instabile: Voto/Non Voto . Milano : Franco Angeli . Dasgupta P. ( 2011 ) A matter of trust: social capital and economic development. In: Lin J. , Plescovic B. (eds) Lessons from East Asia and the Global Financial Crisis , pp. 119 – 155 . Washington, DC : World Bank . Dearmon J. , Grier K. ( 2009 ) Trust and development . Journal of Economic Behavior & Organization , 71 : 210 – 220 . Google Scholar CrossRef Search ADS Deng W. , Lin Y. , Gong J. ( 2012 ) A smooth coefficient quantile regression approach to the social capital—economic growth nexus . Economic Modelling , 29 : 185 – 197 . Google Scholar CrossRef Search ADS Durlauf S. , Kourtellos A. , Minkin A. ( 2001 ) The local Solow growth model . European Economic Review , 45 : 928 – 940 . Google Scholar CrossRef Search ADS Eliasson K. , Westlund H. , Fölster S. ( 2013 ) Does social capital contribute to regional economic growth? Swedish experiences. In: Westlund H. , Kobayashi K. (eds) Social Capital and Rural Development in the Knowledge Society , pp. 113 – 126 . Cheltenham : Edward Elgar . Google Scholar CrossRef Search ADS Florida R. ( 2002 ) The Rise of the Creative Class . New York, NY : Basic Books . Fowler J. ( 2006 ) Altruism and turnout . Journal of Politics , 68 : 674 – 683 . Google Scholar CrossRef Search ADS Gallup J. L. , Sachs J. D. , Mellinger A. D. ( 1999 ) Geography and economic development . International Regional Science Review , 22 : 179 – 232 . Google Scholar CrossRef Search ADS Gennaioli N. , La Porta R. , Lopez-de-Silanes F. , Shleifer A. ( 2013 ) Human capital and regional development . Quarterly Journal of Economics , 128 : 105 – 164 . Google Scholar CrossRef Search ADS Glaeser E. L. , Laibson D. , Sacerdote B. ( 2002 ) An economic approach to social capital . Economic Journal , 112 : 437 – 458 . Google Scholar CrossRef Search ADS Guiso L. , Sapienza P. , Zingales L. ( 2004 ) The role of social capital in financial development . American Economic Review , 94 : 526 – 556 . Google Scholar CrossRef Search ADS Guiso L. , Sapienza P. , Zingales L. ( 2011 ) Civic capital as the missing link. In: Benhabib J. , Bisin A. , Jackson M. (eds) Handbook of Social Economics , pp. 417 – 480 . Amsterdam : North Holland . Guiso L. , Sapienza P. , Zingales L. ( 2016 ) Long-term persistence . Journal of the European Economic Association , 14 : 1401 – 1436 . Google Scholar CrossRef Search ADS Hall R. , Jones C. ( 1999 ) Why do some countries produce so much more output per worker than others? Quarterly Journal of Economics , 114 : 83 – 116 . Google Scholar CrossRef Search ADS Hayek F. A. ( 1949 ) Individualism and Economic Order . London : Routledge . Hungerford T. , Solon G. ( 1987 ) Sheepskin effects in the returns to education . Review of Economics and Statistics , 69 : 175 – 177 . Google Scholar CrossRef Search ADS Keynes J. ( 1936 ) The General Theory of Employment, Interest, and Money . New York , NY: Harcourt, Brace & Co . Ketterer T. , Rodríguez-Pose A. ( 2018 ) Institutions vs. “first-nature” geography: what drives economic growth in Europe’s regions? Papers in Regional Science, 95 : S25 – S62 . Knack S. , Keefer P. ( 1997 ) Does social capital have an economic payoff? A cross country investigation . Quarterly Journal of Economics , 112 : 1251 – 1288 . Google Scholar CrossRef Search ADS Laursen K. , Masciarelli F. , Prencipe A. ( 2012 ) Regions matter: how localized social capital affects external knowledge acquisition and innovation . Organization Science , 23 : 177 – 193 . Google Scholar CrossRef Search ADS Levine R. , Loayza N. , Beck T. ( 2000 ) Financial intermediation and growth: causality and causes . Journal of Monetary Economics , 46 : 31 – 77 . Google Scholar CrossRef Search ADS Mankiw G. , Romer D. , Weil D. ( 1992 ) A contribution to the empirics of economic growth . Quarterly Journal of Economics , 107 : 407 – 437 . Google Scholar CrossRef Search ADS Martins R. , Veiga J. F. ( 2013 ) Economic performance and turnout at national and local elections . Public Choice , 157 : 429 – 448 . Google Scholar CrossRef Search ADS Mauro L. , Pigliaru F. ( 2011 ) Social capital, institutions and growth: Further lessons from the Italian regional divide. CRENoS Working Papers, No. 3, Centre for North South Economic Research. Mortara A. ( 1985 ) Le Associazioni Italiane . Milano : Franco Angeli . Myrdal G. ( 1957 ) Economic Theory and Underdeveloped Regions . London : Duckworth . North D. ( 1991 ) Institutions . Journal of Economic Perspectives , 5 : 97 – 112 . Google Scholar CrossRef Search ADS Openshaw S. , Taylor P. J. ( 1979 ) A million or so correlation coefficients: three experiments on the modifiable area unit problem. In: Wrigley N. (ed) Statistical Applications in the Spatial Sciences , pp. 127 – 144 . London : Pion . Ottaviano G. , Peri G. ( 2006 ) The economic value of cultural diversity: evidence from US cities . Journal of Economic Geography , 6 : 9 – 44 . Google Scholar CrossRef Search ADS Papagapitos A. , Riley R. ( 2009 ) Social trust and human capital formation . Economics Letters , 102 : 158 – 160 . Google Scholar CrossRef Search ADS Platteau J. ( 1994 ) Behind the market stage where real societies exist—part II: the role of moral norms . Journal of Development Studies , 30 : 753 – 815 . Google Scholar CrossRef Search ADS Psacharopoulos G. ( 1985 ) Returns to education: a further international update and implications . Journal of Human Resources , 20 : 583 – 604 . Google Scholar CrossRef Search ADS Psacharopoulos G. ( 1994 ) Returns to investment in education: a global update . World Development , 22 : 1325 – 1343 . Google Scholar CrossRef Search ADS Psacharopoulos G. , Patrinos H. ( 2004 ) Returns to investment in education: a further update . Education Economics , 12 : 111 – 134 . Google Scholar CrossRef Search ADS Putnam R. ( 1993 ) Making Democracy Work: Civic Traditions in Modern Italy . Princeton : Princeton University Press . Putnam R. ( 2000 ) Bowling Alone: The Collapse and Revival of American Community . New York, NY : Simon & Schuster . Putnam R. , Feldstein L. ( 2003 ) Better Together: Restoring the American Community . New York, NY : Simon & Schuster . Resende G. M. ( 2011 ) Multiple dimensions of regional economic growth: the Brazilian case, 1991–2000 . Papers in Regional Science , 90 : 629 – 662 . Google Scholar CrossRef Search ADS Roodman D. ( 2009 ) A note on the theme of too many instruments . Oxford Bulletin of Economics and Statistics , 71 : 135 – 158 . Google Scholar CrossRef Search ADS Routledge B. , von Amsberg J. ( 2003 ) Social capital and growth . Journal of Monetary Economics , 50 : 167 – 193 . Google Scholar CrossRef Search ADS Rutten R. , Gelissen J. ( 2010 ) Social values and the economic development of regions . European Planning Studies , 18 : 921 – 939 . Google Scholar CrossRef Search ADS Rutten R. , Westlund H. , Boekema F. ( 2010 ) The spatial dimension of social capital . European Planning Studies , 18 : 863 – 871 . Google Scholar CrossRef Search ADS Sabatini F. ( 2008 ) Social capital and the quality of economic development . Kyklos , 61 : 466 – 499 . Google Scholar CrossRef Search ADS Sforza A. ( 2014 ) The weather effect: Estimating the effect of voter turnout on electoral outcomes in Italy. B.P. Working Papers, No. 5, Bank of Portugal. Solenne D. , Verrilli A. ( 2012 ) Compendio Di Diritto Costituzionale . Santarcangelo di Romagna : Maggioli . Solow R. ( 1956 ) A contribution to the theory of economic growth . Quarterly Journal of Economics , 70 : 65 – 94 . Google Scholar CrossRef Search ADS Solow R. ( 1995 ) But verify . The New Republic , 213 : 36 – 40 . Tabellini G. ( 2010 ) Culture and institutions: economic development in the regions of Europe . Journal of the European Economic Association , 8 : 677 – 716 . Google Scholar CrossRef Search ADS Temple J. ( 2001 ) Generalizations that aren’t? Evidence on education and growth . European Economic Review , 45 : 905 – 918 . Google Scholar CrossRef Search ADS Westlund H. , Adam F. ( 2010 ) Social capital and economic performance: a meta-analysis of 65 studies . European Planning Studies , 18 : 893 – 919 . Google Scholar CrossRef Search ADS Westlund H. , Larsson J. P. , Olsson A. R. ( 2014 ) Start-ups and local entrepreneurial social capital in the municipalities of Sweden . Regional Studies , 48 : 974 – 994 . Google Scholar CrossRef Search ADS Windmeijer F. ( 2005 ) A finite sample correction for the variance of linear efficient two-step GMM estimators . Journal of Econometrics , 126 : 25 – 51 . Google Scholar CrossRef Search ADS Zak P. , Knack S. ( 2001 ) Trust and growth . Economic Journal , 111 : 295 – 321 . Google Scholar CrossRef Search ADS Appendix A Our empirical strategy is based on the idea that social capital is a pre-condition for future growth. Although this hypothesis is supported by a number of theoretical articles (e.g., Cozzi, 1999; Routledge and von Amsberg (2003); Akçomak and ter Weel, 2009), from an empirical point of view it may be questionable whether the direction of causality goes from social capital to growth, and not vice-versa. To address this concern, we provide estimates of a simple panel VAR(1) between social capital and growth, with the following reduced form (the superscripts identify the corresponding equation): gi,t=via+vta+ρ1a gi,t−10+ρ2a  ln sci,t−10+vi,ta. (A.1) ln sci,t=vib+vtb+ρ1bln sci,t−10+ρ2b gi,t−10+vi,tb. (A.2) The evidence suggests that the coefficient ρ^2a is equal to 0.225 with the standard error of 0.075 (being statistically significant at 1% level), while the coefficient ρ^2b is equal to 0.002 with the standard error of 0.013 (being not statistically significant). On the one hand, these results confirm the chosen direction of causality. On the other hand, they provide indirect support to our proxy of social capital. Nevertheless, we are aware that our results cannot be generalised. For instance, Martins and Veiga (2013) find that economic conditions help to predict voter turnout at legislative elections in Portugal. This is an additional reason why we treat social capital as an endogenous explanatory variable in our main regression model, and we use lagged social capital to avoid simultaneity. Table A1 Linear correlation between voter turnout at political elections and alternative proxies of social capital at province level Alternative proxy Source Correlation with Coefficient Blood bags per 1,000,000 inhabitants, 1995 AVIS Turnout at political elections, 1992 0.566 Propensity to collective action, 1965–1974 ISL-UP Average turnout at political elections, 1963–1968 0.552 Volunteers associations per 100,000 inhabitants, 1999 ISTAT Turnout at political elections, 1992 0.493 Volunteers per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.476 Non-profit institutions per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.426 Associations per 100,000 inhabitants, 1982 ISL-UP Turnout at political elections, 1979 0.285 Share of employment in non-profit sector, 2000 ISTAT Turnout at political elections, 1992 0.176 Violent crimes per 100,000 inhabitants, 1998 ISTAT Turnout at political elections, 1992 −0.239 Alternative proxy Source Correlation with Coefficient Blood bags per 1,000,000 inhabitants, 1995 AVIS Turnout at political elections, 1992 0.566 Propensity to collective action, 1965–1974 ISL-UP Average turnout at political elections, 1963–1968 0.552 Volunteers associations per 100,000 inhabitants, 1999 ISTAT Turnout at political elections, 1992 0.493 Volunteers per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.476 Non-profit institutions per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.426 Associations per 100,000 inhabitants, 1982 ISL-UP Turnout at political elections, 1979 0.285 Share of employment in non-profit sector, 2000 ISTAT Turnout at political elections, 1992 0.176 Violent crimes per 100,000 inhabitants, 1998 ISTAT Turnout at political elections, 1992 −0.239 Notes: ISL-UP, Istituzioni e Sviluppo Locale—Università di Parma; ISTAT, Istituto Nazionale di Statistica; AVIS, Associazione Volontari Italiani del Sangue. The propensity to collective action is a principal-component indicator elaborated by Arrighetti et al. (2003) using information for different years between 1965 and 1974. The number of associations per 100,000 inhabitants in 1982 has been elaborated by Mortara (1985). Table A1 Linear correlation between voter turnout at political elections and alternative proxies of social capital at province level Alternative proxy Source Correlation with Coefficient Blood bags per 1,000,000 inhabitants, 1995 AVIS Turnout at political elections, 1992 0.566 Propensity to collective action, 1965–1974 ISL-UP Average turnout at political elections, 1963–1968 0.552 Volunteers associations per 100,000 inhabitants, 1999 ISTAT Turnout at political elections, 1992 0.493 Volunteers per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.476 Non-profit institutions per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.426 Associations per 100,000 inhabitants, 1982 ISL-UP Turnout at political elections, 1979 0.285 Share of employment in non-profit sector, 2000 ISTAT Turnout at political elections, 1992 0.176 Violent crimes per 100,000 inhabitants, 1998 ISTAT Turnout at political elections, 1992 −0.239 Alternative proxy Source Correlation with Coefficient Blood bags per 1,000,000 inhabitants, 1995 AVIS Turnout at political elections, 1992 0.566 Propensity to collective action, 1965–1974 ISL-UP Average turnout at political elections, 1963–1968 0.552 Volunteers associations per 100,000 inhabitants, 1999 ISTAT Turnout at political elections, 1992 0.493 Volunteers per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.476 Non-profit institutions per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.426 Associations per 100,000 inhabitants, 1982 ISL-UP Turnout at political elections, 1979 0.285 Share of employment in non-profit sector, 2000 ISTAT Turnout at political elections, 1992 0.176 Violent crimes per 100,000 inhabitants, 1998 ISTAT Turnout at political elections, 1992 −0.239 Notes: ISL-UP, Istituzioni e Sviluppo Locale—Università di Parma; ISTAT, Istituto Nazionale di Statistica; AVIS, Associazione Volontari Italiani del Sangue. The propensity to collective action is a principal-component indicator elaborated by Arrighetti et al. (2003) using information for different years between 1965 and 1974. The number of associations per 100,000 inhabitants in 1982 has been elaborated by Mortara (1985). Table A2 Conditional mean estimates on growth: MAUP evidence (1) (2) (3) OLS OLS OLS Municipality Province Region Social capital 0.309*** 0.067 −0.237 (0.039) (0.152) (0.218) Initial income −0.395*** −0.560*** −0.471*** (0.008) (0.038) (0.0771) Time effects Yes Yes Yes Control set Yes Yes Yes Observations 10,145 515 100 R-squared 0.441 0.702 0.893 (1) (2) (3) OLS OLS OLS Municipality Province Region Social capital 0.309*** 0.067 −0.237 (0.039) (0.152) (0.218) Initial income −0.395*** −0.560*** −0.471*** (0.008) (0.038) (0.0771) Time effects Yes Yes Yes Control set Yes Yes Yes Observations 10,145 515 100 R-squared 0.441 0.702 0.893 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A2 Conditional mean estimates on growth: MAUP evidence (1) (2) (3) OLS OLS OLS Municipality Province Region Social capital 0.309*** 0.067 −0.237 (0.039) (0.152) (0.218) Initial income −0.395*** −0.560*** −0.471*** (0.008) (0.038) (0.0771) Time effects Yes Yes Yes Control set Yes Yes Yes Observations 10,145 515 100 R-squared 0.441 0.702 0.893 (1) (2) (3) OLS OLS OLS Municipality Province Region Social capital 0.309*** 0.067 −0.237 (0.039) (0.152) (0.218) Initial income −0.395*** −0.560*** −0.471*** (0.008) (0.038) (0.0771) Time effects Yes Yes Yes Control set Yes Yes Yes Observations 10,145 515 100 R-squared 0.441 0.702 0.893 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A3 Conditional mean estimates on growth with aggregation at (a) province level and (b) regional level (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS (a) Social capital 0.878*** 0.854*** 0.067 −0.037 −0.163 0.551* (0.107) (0.145) (0.152) (0.204) (0.221) (0.332) Initial income −0.269*** −0.293*** −0.560*** −0.890*** −0.884*** −1.013*** (0.012) (0.032) (0.038) (0.043) (0.043) (0.219) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 515 515 515 515 515 515 R-squared/G. of fit 0.545 0.592 0.702 0.796 0.797 0.589 (b) Social capital 0.938*** 0.240 −0.237 −0.126 −0.226 −0.104 (0.181) (0.248) (0.218) (0.253) (0.324) (0.785) Initial income −0.250*** −0.0580 −0.471*** −0.681*** −0.671*** −0.863 (0.0189) (0.0665) (0.0771) (0.0781) (0.0806) (0.619) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 100 100 100 100 100 100 R-squared/G. of fit 0.700 0.799 0.893 0.926 0.927 0.760 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS (a) Social capital 0.878*** 0.854*** 0.067 −0.037 −0.163 0.551* (0.107) (0.145) (0.152) (0.204) (0.221) (0.332) Initial income −0.269*** −0.293*** −0.560*** −0.890*** −0.884*** −1.013*** (0.012) (0.032) (0.038) (0.043) (0.043) (0.219) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 515 515 515 515 515 515 R-squared/G. of fit 0.545 0.592 0.702 0.796 0.797 0.589 (b) Social capital 0.938*** 0.240 −0.237 −0.126 −0.226 −0.104 (0.181) (0.248) (0.218) (0.253) (0.324) (0.785) Initial income −0.250*** −0.0580 −0.471*** −0.681*** −0.671*** −0.863 (0.0189) (0.0665) (0.0771) (0.0781) (0.0806) (0.619) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 100 100 100 100 100 100 R-squared/G. of fit 0.700 0.799 0.893 0.926 0.927 0.760 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A3 Conditional mean estimates on growth with aggregation at (a) province level and (b) regional level (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS (a) Social capital 0.878*** 0.854*** 0.067 −0.037 −0.163 0.551* (0.107) (0.145) (0.152) (0.204) (0.221) (0.332) Initial income −0.269*** −0.293*** −0.560*** −0.890*** −0.884*** −1.013*** (0.012) (0.032) (0.038) (0.043) (0.043) (0.219) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 515 515 515 515 515 515 R-squared/G. of fit 0.545 0.592 0.702 0.796 0.797 0.589 (b) Social capital 0.938*** 0.240 −0.237 −0.126 −0.226 −0.104 (0.181) (0.248) (0.218) (0.253) (0.324) (0.785) Initial income −0.250*** −0.0580 −0.471*** −0.681*** −0.671*** −0.863 (0.0189) (0.0665) (0.0771) (0.0781) (0.0806) (0.619) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 100 100 100 100 100 100 R-squared/G. of fit 0.700 0.799 0.893 0.926 0.927 0.760 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS (a) Social capital 0.878*** 0.854*** 0.067 −0.037 −0.163 0.551* (0.107) (0.145) (0.152) (0.204) (0.221) (0.332) Initial income −0.269*** −0.293*** −0.560*** −0.890*** −0.884*** −1.013*** (0.012) (0.032) (0.038) (0.043) (0.043) (0.219) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 515 515 515 515 515 515 R-squared/G. of fit 0.545 0.592 0.702 0.796 0.797 0.589 (b) Social capital 0.938*** 0.240 −0.237 −0.126 −0.226 −0.104 (0.181) (0.248) (0.218) (0.253) (0.324) (0.785) Initial income −0.250*** −0.0580 −0.471*** −0.681*** −0.671*** −0.863 (0.0189) (0.0665) (0.0771) (0.0781) (0.0806) (0.619) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 100 100 100 100 100 100 R-squared/G. of fit 0.700 0.799 0.893 0.926 0.927 0.760 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A4 Correlation matrix for outcome measures (growth rates) Real GDP per capita Real GDP Population Plant density Employees density Employees per capita Real GDP per capita 1.0000 Real GDP 0.8778 1.0000 Population −0.0124 0.3199 1.0000 Plant density 0.3622 0.5963 0.4210 1.0000 Employees density 0.6323 0.7750 0.3187 0.6638 1.0000 Employees per capita 0.6710 0.7139 0.0120 0.5637 0.9516 1.0000 Real GDP per capita Real GDP Population Plant density Employees density Employees per capita Real GDP per capita 1.0000 Real GDP 0.8778 1.0000 Population −0.0124 0.3199 1.0000 Plant density 0.3622 0.5963 0.4210 1.0000 Employees density 0.6323 0.7750 0.3187 0.6638 1.0000 Employees per capita 0.6710 0.7139 0.0120 0.5637 0.9516 1.0000 Table A4 Correlation matrix for outcome measures (growth rates) Real GDP per capita Real GDP Population Plant density Employees density Employees per capita Real GDP per capita 1.0000 Real GDP 0.8778 1.0000 Population −0.0124 0.3199 1.0000 Plant density 0.3622 0.5963 0.4210 1.0000 Employees density 0.6323 0.7750 0.3187 0.6638 1.0000 Employees per capita 0.6710 0.7139 0.0120 0.5637 0.9516 1.0000 Real GDP per capita Real GDP Population Plant density Employees density Employees per capita Real GDP per capita 1.0000 Real GDP 0.8778 1.0000 Population −0.0124 0.3199 1.0000 Plant density 0.3622 0.5963 0.4210 1.0000 Employees density 0.6323 0.7750 0.3187 0.6638 1.0000 Employees per capita 0.6710 0.7139 0.0120 0.5637 0.9516 1.0000 Table A5 Conditional mean estimates on growth with the full sample (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.850*** 0.892*** 0.283*** 0.239*** 0.103*** 0.444*** (0.020) (0.021) (0.0251) (0.038) (0.038) (0.201) Initial income −0.380*** −0.408*** −0.498*** −0.962*** −0.967*** −0.792*** (0.003) (0.004) (0.004) (0.005) (0.005) (0.148) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 40,185 40,185 40,185 40,185 40,185 40,185 R-squared/G. of fit 0.336 0.350 0.408 0.688 0.691 0.398 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.850*** 0.892*** 0.283*** 0.239*** 0.103*** 0.444*** (0.020) (0.021) (0.0251) (0.038) (0.038) (0.201) Initial income −0.380*** −0.408*** −0.498*** −0.962*** −0.967*** −0.792*** (0.003) (0.004) (0.004) (0.005) (0.005) (0.148) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 40,185 40,185 40,185 40,185 40,185 40,185 R-squared/G. of fit 0.336 0.350 0.408 0.688 0.691 0.398 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A5 Conditional mean estimates on growth with the full sample (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.850*** 0.892*** 0.283*** 0.239*** 0.103*** 0.444*** (0.020) (0.021) (0.0251) (0.038) (0.038) (0.201) Initial income −0.380*** −0.408*** −0.498*** −0.962*** −0.967*** −0.792*** (0.003) (0.004) (0.004) (0.005) (0.005) (0.148) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 40,185 40,185 40,185 40,185 40,185 40,185 R-squared/G. of fit 0.336 0.350 0.408 0.688 0.691 0.398 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.850*** 0.892*** 0.283*** 0.239*** 0.103*** 0.444*** (0.020) (0.021) (0.0251) (0.038) (0.038) (0.201) Initial income −0.380*** −0.408*** −0.498*** −0.962*** −0.967*** −0.792*** (0.003) (0.004) (0.004) (0.005) (0.005) (0.148) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 40,185 40,185 40,185 40,185 40,185 40,185 R-squared/G. of fit 0.336 0.350 0.408 0.688 0.691 0.398 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. © The Author(s) (2018). Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Economic Geography Oxford University Press

Social capital and growth: causal evidence from Italian municipalities

Loading next page...
 
/lp/ou_press/social-capital-and-growth-causal-evidence-from-italian-municipalities-E1OVSa7a4C
Publisher
Oxford University Press
Copyright
© The Author(s) (2018). Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com
ISSN
1468-2702
eISSN
1468-2710
D.O.I.
10.1093/jeg/lby024
Publisher site
See Article on Publisher Site

Abstract

Abstract The macroeconomic effects of social capital are typically studied using data at country, region or, at most, province level of aggregation. However, social capital is defined by connections among agents who know each other and its effects, if any, should be detected at a more detailed level of spatial aggregation. To the best of our knowledge, this is the first study using longitudinal municipality-level data to investigate the causal link between social capital and growth. We extend earlier research by accounting for the endogeneity of all the covariates as well as unobserved heterogeneity. The evidence suggests that social capital has been a source of growth inequality in Italy between 1951 and 2001. The causal effect of social capital on growth is positive, on average, and stronger in the Centre-North of Italy. In addition, it was higher in the 1950s. The paper also presents local estimates of the growth return to social capital, which are of interest for specific sub-populations of municipalities. 1. Introduction The role of institutions in shaping economic performance has been emphasised at least since North (1991), if not earlier (Hayek, 1949; Myrdal, 1957). Institutions are seen as humanly devised constraints, which affect political, social and economic interaction. These constraints can be both formal (constitutions, laws, property rights) and informal. Customs, traditions and codes of conduct belong to the latter category. Communities may well be different for their customs, and these differences may well explain their relative stages of economic development. For example, in societies where individuals are used to trust each other, trade is likely to be easier and more developed. Trust, cooperation and participation in the community life are important characteristics of what is usually referred to as ‘social capital’. This paper looks at the link between social capital and economic growth. The specific interest on this link follows the contributions by Putnam (1993, 2000), arguing that civic engagement promotes the formation of trust, which in turn improves the quality of governance and thus economic performance. Analogously, Knack and Keefer (1997) argue that social capital may affect growth in a variety of ways. Economic activities that require some agents to rely on others are accomplished at lower cost in environments with higher social capital. Individuals spend less time to protect themselves from being exploited, thus being more productive. Entrepreneurs devote less time to monitor their employees and suppliers, thus having more time to devote to innovation. Governments are more credible, and people adopt more appropriate horizons in making investment decisions. Publicly provided goods, including education, have higher quality. Hiring decisions are more transparent and based on education credentials.1 The above view implicitly assumes that social capital has a positive effect on growth. The validation of this hypothesis has inspired several empirical studies, which typically rely upon data available at country, region or, at most, province level of aggregation (e.g., Algan and Cahuc, 2010; Tabellini, 2010). However, as argued by Putnam and Feldstein (2003, 9), ‘social capital is necessarily a local phenomenon because it is defined by connections among people who know one another’. In line with Putnam and Feldstein’s (2003) intuition, this paper investigates the link between social capital and growth at a more detailed level of spatial aggregation. In particular, we use municipality-level data for Italy. The research questions that we address are the following: What is the average elasticity of social capital to real growth over a long time span? Is it positive? Is there a causal link from social capital to real growth? Is there time heterogeneity in the link? Is there spatial heterogeneity? The answers to these questions are important because they shed further light on what lies behind the Solow residual, which is ‘a measure of our ignorance’ in standard empirical growth models, as emphasised by Abramovitz (1956). Most economies show a decreasing trend in the growth rate of real GDP per capita over time. Is this pattern related to a decreasing level of social capital? Or to a decreasing return to social capital? Or to both? Some economies show regional imbalances. Are they related to social-capital levels? Or to returns to those levels? Or to both? Although some of these questions are not novel, this paper makes five contributions. First, there is no systematic study of the social capital vs. growth nexus which answers all the above questions using longitudinal municipality-level data.2 These data have several advantages. Being longitudinal, they allow us to separate the growth effects of unobserved time-invariant city characteristics,3 such as the average quality of formal local institutions, from those of social capital attributes. As municipality-level data, they allow us to better approximate the local nature of social capital as well as to increase the number of cross-sectional units, making the empirical analysis more robust to the presence of outliers. Second, we innovate by accounting for the endogeneity of all the explanatory variables, including social capital, by means of internal instruments. In particular, we investigate whether a change in the social capital endowment of a municipality affects the growth rate of its per capita value added over the 1951–2001 period. Third, we investigate the extent to which lower growth in the South of Italy can be, in part, attributed not only to a lower level of social capital but also to a lower return to social capital. Fourth, we consider the possibility of a time-varying relationship between social capital and economic growth. If social capital behaves as a substitute of formal institutions in achieving economic growth (Ahlerup et al., 2009), its growth-enhancing role is likely to decline over time as formal institutions develop. In the specific case of Italy, the formal institutional system of the country was pretty much under construction during the 1950s, after the end of the Monarchy and the enforcement of the Republican Constitution in 1948. It is, therefore, likely that social capital played a more important role in that period than in others. Fifth, we provide novel local estimates of the impact of social capital on growth, using an identification approach based on two municipality-level external instruments (here, the term ‘local’ is used in the spirit of Angrist and Pischke, 2009). The rest of the paper is organised as follows. Section 2 provides reasons to use municipality-level data. Section 3 discusses our measures of social capital and growth. Section 4 describes the dataset and the empirical model. Section 5 presents results and robustness checks. Section 6 deals with local estimates. Section 7 concludes. 2. Municipalities as units of analysis The economic literature on social capital has discussed the interaction between various forms of social capital and economic development. For example, Sabatini (2008) has performed an empirical study using 200 variables on different aspects of economic development and four dimensions of social capital: strong family ties (generally referred to as ‘bonding’ social capital), weak informal ties (‘bridging’ social capital), voluntary organisations (‘linking’ social capital), and political participation. Each dimension includes from 4 to 25 variables. This should immediately alert the reader about the multiplicity of the definitions of social capital and outcome indicators that have been explored in earlier studies so far. One feature of this heterogeneity has to do with the nature of the data used, and specifically with the unit of analysis. With reference to the latter, the existing studies on social capital and economic outcomes can be broadly divided in two main categories: the ‘macro’ studies typically using data at regional or country level; and the ‘micro’ studies using data on individuals and firms. A striking feature of the literature is that the ‘micro’ studies are not necessarily consistent with the ‘macro’ studies, or vice-versa. On the one hand, many ‘macro’ studies report zero or even negative effect of social capital on economic performance, with Italy being a ‘special case’ where the link looks stronger and positive (Westlund and Adam, 2010).4 On the other hand, the ‘micro’ studies find clear evidence that social capital affects firm performance (e.g., Laursen et al., 2012) and people behaviour (e.g., Guiso et al., 2004).5 It follows that the evidence on the nexus between social capital and economic performance or behaviour at the micro (firm or individual) level, while having the expected sign, does not necessarily imply that social capital matters at the macro (GDP growth) level, which is the level investigated in this paper. To the extent that spatial proximity is crucial even in the Internet era (Rutten et al., 2010) and influences the cost of social capital investment (Glaeser et al., 2002), the step from firm or individual outcomes to regional or even national outcomes may be too big. This is one reason why a number of recent studies (Rutten and Gelissen, 2010; Corazzini et al., 2011; Eliasson et al., 2013; Albanese and de Blasio, 2014; Westlund et al., 2014) have focused on smaller spatial units, such as municipalities.6 However, there are four additional reasons for using municipality-level data: Disaggregation allows for the use of primary data on social capital. Our measure of social capital is voter turnout in political elections (Putnam, 1993). In Italy, measures of turnout rates in elections and referenda are produced at municipality level because the law (Decreto del Presidente della Repubblica del 20 marzo 1967) attributes to the municipalities the role of organizing and managing the voting operations, regardless of the type of popular consultation.7 It follows that provincial and regional data on voter turnout, extensively used in earlier studies, are inevitably secondary data, more likely to be measured with error. Aggregation can affect the level of social capital. Since the concept of social capital itself can only be defined within a given community, a change in the administrative definition of the community (e.g., municipality, province, region or nation) may affect the level of social capital. For instance, an inhabitant of a city may be very cooperative with an inhabitant of the same city, but not cooperative at all with an inhabitant of another city in the same province. Thus, cooperation may be high at city level, but low at province level. Aggregation can affect the estimate of the growth return to social capital. After Openshaw and Taylor (1979), who argued that correlations in Iowa between Republican voting and percentage of old people could vary a lot depending on how counties were aggregated, the modifiable area unit problem (MAUP) literature has expanded in several directions, including intersections with the growth literature. An example is Resende (2011) who has recently shown that ‘the determinants of economic growth in Brazil may manifest themselves differently on various spatial scales’ (Resende, 2011, 639), namely, states, micro-regions, spatial clusters and municipalities. Disaggregation provides both reliable generalized method of moments - system (GMM-SYS) estimates and novel two stage least squares (2SLS) estimates of the growth return to social capital. As we will see, the aggregation of data at provincial and regional level does not allow us to obtain reliable GMM-SYS estimates because identification criteria are not passed. In contrast, municipality-level data help us to obtain both reliable GMM-SYS estimates and novel 2SLS estimates of the causal effect of interest. The overall conclusion from the above discussion is that an analysis based on more disaggregated data is important because (dis)aggregation matters, or does it in principle, both from an empirical and a theoretical point of view. 3. Measures of growth and social capital The analysis of the link between social capital and growth with municipality-level data presents at least two empirical challenges. On the one hand, in the absence of a direct measure of economic growth at municipality level, we need to compute a proxy for the growth rate of the real per capita value added at such a detailed level of aggregation. We do this by using the municipality-level sectoral employment shares as weights to spread the value added at province level on all the municipalities of the same province. On the other hand, among the various definitions and related measures of social capital, we need to pick one definition whose measure is available at municipality level. In particular, we use the notion of civic capital (Guiso et al., 2011) proxied by the voter turnout at political elections (Putnam, 1993), i.e., the percentage of eligible voters who actually cast their vote. This measure captures the citizen engagement in the life of the community (Almond and Verba, 1963; Fowler, 2006) even though its use as a proxy of social capital may be controversial. This section discusses our measures of growth and social capital in detail. 3.1. Growth Our measure of economic development cannot be directly observed because, to the best of our knowledge, data on the value added are only available, at most, at the province level of aggregation. However, as province level data are available for each sector j, we can compute a proxy for the per capita value added at municipality level by using the sectoral employment shares—available at municipality level—as weights to spread the value added at province level on all municipalities of the same province. The sectoral employment shares are computed as the number of employees of a given sector in a given municipality over the total number of employees of the same sector in the entire province.8 In practice, we multiply the value added in sector j in province p (say vapj) times the employment share in sector j of municipality i in province p (say emplijemplpj). The result is the value added in sector j of municipality i (say vaij=vapj·emplijemplpj). This approach implicitly assumes that the average labour productivity in a municipality in a given sector is equal to that of the province in the chosen sector ( vaijemplij=vapjemplpj). This assumption is consistent with the idea that labour mobility within a province implies an equalisation of wages and productivities in a sector among municipalities in the same province, which is in turn consistent with the long-run approach taken in this paper.9 Adding up the value added of each sector at municipality level, we get the total value added at municipality level (say vai=∑jvaij). We use the population to compute the per capita value added. The real per capita value added is obtained using the national price index as deflator (base year = 1951). Finally, as we can only compute the value added at Census dates (i.e. every 10 years), its growth rate is obtained as difference between the natural logarithms of real per capita value added at Census dates. 3.2. Social capital But, what do we mean by social capital here? And, how do we measure it? The rest of this section provides an answer to both these questions. 3.2.1 Defining social capital as civic capital After its introduction by Bourdieu (1972) at the beginning of the 1970s,10 the concept of social capital has gained wide acceptance in social sciences, including economics, over the past four decades. Yet, its success has been proportional to the ambiguity of the term.11 In this paper, we follow the literature originated by Solow (1995) who described the specific characteristics that social capital must have ‘to be more than a buzzword’. First, to retain the term ‘capital’, likewise other forms of capital, social capital must have a dynamics of accumulation and depreciation. Second, it must have a non-negative economic payoff. Third, it has to be measurable, even inexactly. Fourth, to retain the term ‘social’, social capital must be clearly distinguishable from other forms of capital accumulated at firm (physical capital) or individual (human capital) level. As convincingly argued by Guiso et al. (2011), a concept of social capital which fits all the Solow criteria is the concept of ‘civic capital’, defined as ‘those persistent and shared beliefs and values that help a group overcome the free rider problem in the pursuit of socially valuable activities’ (Guiso et al., 2011, 419).12Guiso et al. (2011) go one step further suggesting that civic capital is not just one, but it is the only concept of social capital satisfying all the Solow criteria. Despite the latter may be controversial, and a less radical view is advisable, this paper follows the approach by Guiso et al. (2011) using a proxy of civic capital in the empirical analysis. 3.2.2 Measuring civic capital as voter turnout From now on, we will use the expressions ‘social capital’ and ‘civic capital’ interchangeably. Guiso et al. (2011) provide an outstanding overview of all the possible direct and indirect measures of civic capital. Direct measures of civic capital are not currently available at municipality level. So, the analysis must rely on indirect measures. According to Guiso et al. (2011), the most important indirect measures of civic capital are two: anonymous blood donations and participation in referenda. The former is an ideal measure of social capital because an anonymous blood donation is a free individual choice giving no personal returns and supporting the community life. The latter measure, instead, though widely used in the earlier research, is subject to two types of criticism. First, low participation in a referendum does not necessarily mean low participation in the community life—i.e., low social capital in a particular community—because people who are in favour of a particular law may choose not to vote in order to avoid that a referendum reaches the 50% cutoff needed to repeal a law.13 Second, people may have a personal return from a particular law under scrutiny (e.g., married couples and divorce, Catholic women and abort and so on). Both the above two indirect measures of civic capital are not available at municipality level for our period of interest. Thus, we must rely on a different indirect measure of civic capital. In particular, we use voter turnout at political elections as a proxy of civic capital (Putnam, 1993). The usual argument against our proxy is that voting at political elections may be a constrained choice, not a free one as an anonymous blood donation. In addition, it may imply personal returns as politicians may provide benefits to their electorate in exchange for votes. However, both these two arguments do not seem to apply to the case of this paper. The dutifulness of the vote was debated in the Assemblea Costituente. The compromise between those who wanted to qualify the exercise of voting as a legal duty and those who, instead, wanted to qualify it as a moral duty (Solenne and Verrilli, 2012) led to the qualification of voting as a civic duty in the constitutional chart (Costituzione della Repubblica italiana, art. 48). The constitutional norm allowed for the legislator to set only weak penalisations for voter abstention in 1957 (Decreto del Presidente della Repubblica del 30 marzo 1957, art. 115). Abstention should have been justified to the city mayor, who in turn had to publish a list of unjustified non-shows in the municipal albo for 1 month. In addition, the expression ‘he/she did not vote’ had to be recorded for 5 years in the certificato di buona condotta issued by the municipality. However, the application of the 1957 law was progressively disregarded (Cuturi et al., 2000), until the electoral reform of 1993 (Decreto Legislativo del 20 dicembre 1993) fully recognised the abstention as an expression of political choice (Cassese et al., 2009). Hence, even though voting is still intended as a civic duty in the Constitution—something that one should do—it is reasonable to think that voting at political elections has always been a substantially free choice because no punishment for abstention has ever been set, besides a mere register of non-voting. As for the personal returns, the possibility of a benefit-vote link is a real issue and should not be disregarded. However, the argument is more likely to apply to the administrative elections (e.g., city mayor) than to the political ones. Indeed, elected deputies in Italy have the constitutional right to operate without a binding mandate (Costituzione della Repubblica italiana, art. 67), meaning that, once elected, they can do whatever they like, even against the interests of their party or electorate. In addition, despite in the elections covered in our sample people could vote both for parties and specific persons, parties and elected typically had to set coalitions in order to achieve a Parliamentary majority and form a Government. Therefore, compared with a referendum where the potential outcomes are known with certainty (abrogation or not of a specific law), the vote at political elections provided, at best, non-predictable personal returns to the average voter.14 In contrast, the potential troubles could be easily forecasted: the law established—and still does—strong criminal sanctions for both those who promise and those who accept benefits in exchange for votes at political elections (Decreto del Presidente della Repubblica del 30 marzo 1957, art. 96). In sum, voter turnout at political elections may not be the best measure of social capital15 but we use it in this paper because: (i) it has already been used by many authors including Putnam (1993), (ii) it suffers from fewer distortions than referenda turnouts and (iii) to the best of our knowledge, it is the only measure available at municipality level over the long time span that we investigate. The ‘smoking gun’ is the relationship between data on anonymous blood donations (see Guiso et al., 2004), available only for the year of 1995 and at province level, and data on voter turnout at Parliamentary elections (Chamber of Deputies) for the closest year of elections in our dataset, i.e., 1992. Figure 1 suggests that an increase in voter turnout is associated with a more-than-proportional increase in blood donations. The estimated elasticity is, indeed, equal to 7.2 and statistically significant at 1% level. Figure 1 View largeDownload slide Correlation between voter turnout and blood donations. Notes: Data are at province level. Blood donations are measured as the number of blood bags (each bag contains 16 ounces of blood) per million inhabitants in the province. Voter turnout refers to the Parliamentary elections (Chamber of Deputies). Figure 1 View largeDownload slide Correlation between voter turnout and blood donations. Notes: Data are at province level. Blood donations are measured as the number of blood bags (each bag contains 16 ounces of blood) per million inhabitants in the province. Voter turnout refers to the Parliamentary elections (Chamber of Deputies). In addition, the evidence presented in the Appendix Table A1 supports our choice because voter turnout at political elections is, in general, linearly correlated with other proxies of social capital available at province level, with the expected sign (note that imposing a linear association between two variables may be rather restrictive, as documented in Figure 1). 4. Empirical strategy Our main data sources are: (i) the Istituto Tagliacarne for data on the value added at province level; (ii) the Istituto Nazionale di Statistica (ISTAT) for the Census data on surface, population and employment at municipality level; (iii) the Atlante Storico Elettorale d’Italia provided by the Istituto Cattaneo for data on political participation, measured as voter turnout at Parliamentary elections (Chamber of Deputies) at municipality level and (iv) the Associazione Nazionale dei Comuni Italiani (ANCI) for data on geographic characteristics of cities. We use data on about 2100 municipalities with at least 5000 inhabitants in 1951 and we consider the 1951–2001 period, divided in time intervals that correspond to the Census decades. The analysis is not extended further in time because of data constraints. Table 1 provides the summary statistics for the sample. On average, each city has a population of 20,349 inhabitants and extends over 71 km2. The 17% of cities in the sample is located close to the seaside and the 40% is located in the South. The real per capita value added is on average equal to 797,000 constant Italian liras (base year = 1951) and displays a 39% average growth rate over 10 years. The first growth rate refers to the 1951–1961 period. The last is for 1991–2001. Our measure of social capital displays an average of 90%. Ideally, we would like to have data on social capital every 10 years since 1951. In practice, we use data on voter turnout at political elections held in 1953, 1963, 1968, 1979 and 1992. Table 1 Summary statistics Variable Observation Mean Standard deviation Min. Max. Real growth rate (10 years) 10,145 0.39 0.37 −2.18 3.28 Social capital 10,145 0.90 0.07 0.41 0.99 Population 10,145 20,349.42 82,210.04 771 2,546,804 Real income (base year = 1951) 10,145 797.58 525.78 35.56 6613.81 Altitude 10,145 233.42 229.10 0 1211 Difference altitude 10,145 532.60 547.14 1 3282 Coast location 10,145 0.17 0.37 0 1 Surface 10,145 71.10 77.87 1.62 1307.71 Provincial capital 10,145 0.05 0.22 0 1 South 10,145 0.40 0.49 0 1 Variable Observation Mean Standard deviation Min. Max. Real growth rate (10 years) 10,145 0.39 0.37 −2.18 3.28 Social capital 10,145 0.90 0.07 0.41 0.99 Population 10,145 20,349.42 82,210.04 771 2,546,804 Real income (base year = 1951) 10,145 797.58 525.78 35.56 6613.81 Altitude 10,145 233.42 229.10 0 1211 Difference altitude 10,145 532.60 547.14 1 3282 Coast location 10,145 0.17 0.37 0 1 Surface 10,145 71.10 77.87 1.62 1307.71 Provincial capital 10,145 0.05 0.22 0 1 South 10,145 0.40 0.49 0 1 Notes: Data sources are: ANCI, ISTAT, Istituto Cattaneo, Istituto Tagliacarne. Social capital is measured with voter turnout at the Parliamentary elections (Chamber of Deputies). Table 1 Summary statistics Variable Observation Mean Standard deviation Min. Max. Real growth rate (10 years) 10,145 0.39 0.37 −2.18 3.28 Social capital 10,145 0.90 0.07 0.41 0.99 Population 10,145 20,349.42 82,210.04 771 2,546,804 Real income (base year = 1951) 10,145 797.58 525.78 35.56 6613.81 Altitude 10,145 233.42 229.10 0 1211 Difference altitude 10,145 532.60 547.14 1 3282 Coast location 10,145 0.17 0.37 0 1 Surface 10,145 71.10 77.87 1.62 1307.71 Provincial capital 10,145 0.05 0.22 0 1 South 10,145 0.40 0.49 0 1 Variable Observation Mean Standard deviation Min. Max. Real growth rate (10 years) 10,145 0.39 0.37 −2.18 3.28 Social capital 10,145 0.90 0.07 0.41 0.99 Population 10,145 20,349.42 82,210.04 771 2,546,804 Real income (base year = 1951) 10,145 797.58 525.78 35.56 6613.81 Altitude 10,145 233.42 229.10 0 1211 Difference altitude 10,145 532.60 547.14 1 3282 Coast location 10,145 0.17 0.37 0 1 Surface 10,145 71.10 77.87 1.62 1307.71 Provincial capital 10,145 0.05 0.22 0 1 South 10,145 0.40 0.49 0 1 Notes: Data sources are: ANCI, ISTAT, Istituto Cattaneo, Istituto Tagliacarne. Social capital is measured with voter turnout at the Parliamentary elections (Chamber of Deputies). Figure 2 suggests that voter turnout in Italy has decreased over time.16 This pattern is consistent with the one found in Albanese and de Blasio (2014). Figure 2 View largeDownload slide The evolution of voter turnout at political elections. Figure 2 View largeDownload slide The evolution of voter turnout at political elections. Interestingly, Figure 3 depicts a pattern of growth which is similar to the one observed for our proxy of social capital. What is really intriguing, however, is the fact that growth has decreased proportionally more than social capital, which opens the door to the possibility that the growth return to social capital has decreased over time. Figure 3 View largeDownload slide The evolution of real growth. Figure 3 View largeDownload slide The evolution of real growth. As for the empirical strategy, our idea is that social capital may constitute a pre-condition for future growth. Hence, we estimate a model of the following form: gi,t=μi+μt+αln yi,t−10+βln sci,t−10+γln popi,t−10+∑k=1Nδk Xk,i+ei,t, (1) where gi,t is the growth rate of real per capita value added between year t and year t−10 for the unit of interest i (the first observed growth rate refers to the 1951–1961 period); μi is the vector of the city-fixed effects (FEs); μt represents the time effects between year t and year t−10 (1951–1961, 1961–1971 and so on); ln yi,t−10 is the log of real per capita value added at the beginning of the period (the first observation is for 1951); ln sci,t−10 is the log of social capital endowment at the beginning of the period (the first observation should be for 1951, but the closest observation we have is for 1953)17; ln popi,t−10 is the log of population at the beginning of the period (the first observation is for 1951); X1 to XN are observable, mainly geographical, control variables and ei,t is an i.i.d. disturbance. Our model is a ‘social capital and growth’ version of a famous ‘financial intermediation and growth’ model proposed by Levine et al. (2000) in an article in the Journal of Monetary Economics. This article has been very influential in the finance–growth literature and its title has inspired the title of our manuscript. Following Levine et al. (2000), our model can be written as a dynamic panel-data model by noting that gi,t≅ln yi,t−ln yi,t−10. As our model is specified in logarithms, the parameter of interest β should be interpreted as the elasticity of the growth rate of real per capita value added with respect to social capital.18 Our empirical model imposes a given direction of causality, which goes from social capital to GDP growth. This hypothesis is supported by the panel VAR estimates proposed in Appendix A. We control for city-FEs to capture unobserved time-invariant city characteristics, such as the average quality of formal local institutions. We control for the lagged log of population and time effects to capture agglomeration economies and decade heterogeneity, respectively. The variables X1 to XN are as follows: a dummy equal to one if the municipality is located in the South; a dummy equal to one if the municipality is close to the seaside; a dummy equal to one if it is a province capital; the log of surface in square kilometres; the log of altitude from the townhouse in metres and the log of difference in altitudes within the municipality. We control for geographical variables for three main econometric reasons: Turnout is potentially correlated with geographical factors. For example, altitude may be correlated with turnout through weather conditions (the probability of raining in the day of the elections is higher at higher altitudes).19 Hence, not controlling for geographical factors would introduce a new source of social-capital endogeneity in the residuals of the growth regression. Geographical variables are useful for identification as instruments in dynamic panel-data models because they can be treated as exogenous covariates. Geography explains a share of growth variability in regression analyses using country-level data, as stressed by Gallup et al. (1999), among others. Controlling for geographical variables in studies using more disaggregated data is in line with Gennaioli et al. (2013) and Ketterer and Rodríguez-Pose (2018), among others. Our estimation strategy is as follows. We start by disregarding the presence of city-FEs and estimate model (1) by ordinary least squares (OLS). We first control for initial income only; we then add time dummies and finally, we add both time dummies and the control set. Our definition of ‘control set’ includes lagged population and all the variables X1 to XN. Afterwards, we control for the role of unobserved heterogeneity through a FEs approach. Note, however, that we expect the FE estimates to be biased due to the presence of both unobserved heterogeneity and lagged value added among regressors. This source of bias can be only resolved by estimating model (1) with a dynamic GMM approach. In the dynamic GMM framework, a GMM-system (GMM-SYS) approach (Arellano and Bover, 1995; Blundell and Bond, 1998) should be the preferred estimation method for a variety of reasons. First, like the GMM-difference (GMM-DIF) approach (Arellano and Bond, 1991), it allows for dealing with the potential endogeneity of all explanatory variables, rather than only that of social capital. Second, it is more efficient than GMM-DIF as it uses more moment conditions. Indeed, likewise GMM-DIF, the GMM-SYS approach uses lags of regressors as instruments for the first-difference transformation of model (1). But, differently from GMM-DIF, the GMM-SYS approach also uses lagged differences of regressors as instruments for model (1) itself. Third, GMM-SYS allows us to estimate the coefficients of time-invariant regressors, while GMM-DIF does not. Fourth, the GMM-SYS estimator has been shown to behave better (in terms of bias and precision) than the GMM-DIF estimator in case of persistent variables (Blundell and Bond, 1998; Blundell et al., 2000), as social capital is likely to be. The choice of instruments is based on Bond et al. (2001), who show how to estimate empirical growth models using the GMM-system approach. In particular, as instruments for the first-difference equation, we use ln yi,t−20, ln sci,t−20, ln popi,t−20 and earlier. As instruments of the level equation, we use Δln yi,t−10, Δln sci,t−10, Δln popi,t−10 and earlier. The variables X1 to XN are taken as exogenous covariates and used as instruments in the standard dynamic GMM practice. The instrument matrix is collapsed (Roodman, 2009). Standard errors are Windmeijer-corrected (Windmeijer, 2005). The goodness of fit for GMM-SYS is based on Bloom et al. (2007). The results are presented in the next section together with a variety of robustness checks. 5. Results and robustness In presenting the results, we focus on the link between social capital and growth. Before commenting on our preferred estimate, it is worth investigating how the social capital coefficient changes in different specifications and estimation methods. 5.1. Main results Controlling for initial income only, the OLS estimate of the social capital coefficient (Table 2, Column 1) is positive and significant. Adding time effects (Column 2), the estimate slightly decreases in terms of magnitude but remains highly significant. When adding the control set (Column 3), the coefficient remains significant but its magnitude decreases. When aggregating the data at provincial and regional level, we find evidence of a MAUP (see Appendix Table A2). In line with Resende (2011), aggregation causes the return to social capital to change and the standard error to get larger. Table 2 Conditional mean estimates on growth: baseline estimates (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.853*** 0.749*** 0.309*** 0.221*** 0.170** 0.595*** (0.0301) (0.034) (0.039) (0.067) (0.068) (0.109) Initial income −0.302*** −0.281*** −0.395*** −0.903*** −0.904*** −0.660*** (0.004) (0.007) (0.008) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.375 0.387 0.441 0.682 0.683 0.407 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.853*** 0.749*** 0.309*** 0.221*** 0.170** 0.595*** (0.0301) (0.034) (0.039) (0.067) (0.068) (0.109) Initial income −0.302*** −0.281*** −0.395*** −0.903*** −0.904*** −0.660*** (0.004) (0.007) (0.008) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.375 0.387 0.441 0.682 0.683 0.407 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 2 Conditional mean estimates on growth: baseline estimates (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.853*** 0.749*** 0.309*** 0.221*** 0.170** 0.595*** (0.0301) (0.034) (0.039) (0.067) (0.068) (0.109) Initial income −0.302*** −0.281*** −0.395*** −0.903*** −0.904*** −0.660*** (0.004) (0.007) (0.008) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.375 0.387 0.441 0.682 0.683 0.407 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.853*** 0.749*** 0.309*** 0.221*** 0.170** 0.595*** (0.0301) (0.034) (0.039) (0.067) (0.068) (0.109) Initial income −0.302*** −0.281*** −0.395*** −0.903*** −0.904*** −0.660*** (0.004) (0.007) (0.008) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.375 0.387 0.441 0.682 0.683 0.407 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. A source of bias may be due to the fact that unobserved time-invariant city characteristics (e.g., the average quality of formal local institutions) are likely to be correlated with social capital. When controlling for such characteristics, the FE estimates (Column 4 and, including the control set, Column 5) are lower in magnitude than the OLS ones, but they are still positive and significant. As highlighted in the previous section, our preferred estimation method is the one that allows for dealing with the endogeneity of all covariates, including the lagged value added, as well as with the unobserved heterogeneity, i.e., the GMM-SYS approach. The GMM-SYS estimates (Column 6) confirm the positive and significant link between social capital and growth. The social capital coefficient estimate is larger than the one obtained using the FE estimator. The identification approach is supported by the fact that both the Hansen and the Difference-in-Hansen tests do not reject the null hypothesis of instruments joint validity (the p-values are equal to 0.276 and 0.131, respectively) as well as by the fact that the Arellano–Bond test does not reject the null hypothesis of no second-order autocorrelation of residuals in first differences (the p-value is equal to 0.819). In contrast, identification is problematic when using data at province level because the instruments for the level equation do not pass the Difference-in-Hansen test (see Column 6 in Appendix Table A3(a) for estimation results). Identification is even more problematic when aggregating the data at regional level because the Arellano–Bond test rejects the null at 10%, the Hansen test rejects the null at 10%, and the Difference-in-Hansen test rejects the null at 5%. Interestingly, we find that aggregation changes the return to social capital, which lowers as aggregation increases (see the GMM-SYS estimates in Table 2, Appendix Table A3(a,b)). This is in line with Westlund and Adam (2010) because the growth return to social capital becomes not statistically significant at 5% level when aggregating at provincial level (see Column 6 in Appendix Table A3(a) for estimation results), and even negative when aggregating at regional level (see Column 6 in Appendix Table A3(b) for estimation results). The OLS results in Appendix Table A2 go in the same direction. An important thing to notice is that the additional moment conditions imposed by the GMM-SYS estimator are valid provided the validity of some initial-condition restrictions. The latter hold under (sufficient but not necessary) assumptions of mean stationarity of the stochastic processes that generate the data involved in the estimation. This assumption also guarantees that the instruments for the first-difference equation are not weak (Blundell and Bond, 1998, 2000; Bond, 2002). In order to test the mean-stationarity hypothesis, we follow Blundell and Bond (2000) and estimate an AR(1) process (with a constant) for the log of real per capita value added, the log of the social capital proxy and the log of population. In the presence of city FEs, the OLS method gives an upward-biased estimate of the lagged dependent variable coefficient. This means that an OLS estimate of the autoregression coefficient below the critical value of 1 would ensure us that we are not working with unit-root series. In the case of the real per capita value added, the estimate is 0.7, i.e., well below the critical value. In the case of the social capital proxy, the estimate is equal to 0.9, i.e., still below the critical value. In the case of population, the estimate increases to 1.0 but the fact that it is upward biased reassures about the mean stationarity of the true data-generating process. Intuitively, there is no reason to believe that growth, social capital and population evolve as random walks. It is more realistic to think that steady-state (i.e., equilibrium) values of these variables do exist, even though persistence is high. The latter is also consistent with standard growth models, including ours. Our first result is that the social capital endowment of a city matters for subsequent economic growth in a positive and significant way. In particular, according to the preferred estimation method (GMM-SYS), the elasticity of social capital endowment with respect to growth is 0.6. The coefficient of initial income is found to be negative and significant. This suggests that municipalities with higher (lower) initial income grow slower (faster), which supports the idea of convergence. However, we do not find evidence of convergence among all Italian municipalities. Since the estimated coefficient of the dummy for location in the South is negative (−0.341) and significant at 1% level,20 the municipalities in the Centre-North of Italy converge to an equilibrium growth rate which is higher than in the South. In the rest of this section, we will provide additional results useful to explain time and spatial growth variability, and the role potentially played by social capital. 5.2. Spatial externalities As a first robustness check, we control for the existence of spatial externalities (Table 3). Specifically, we augment model (1) with the aggregate initial income of all the other municipalities in the same province. The GMM-SYS estimate of the coefficient of this variable is very close to zero from the left (−0.006) and it is not significant. As a result, the social capital impact changes only a little. If we construct an interaction between the initial income of the municipality and the aggregate initial income of all the other municipalities in the same province, the interaction coefficient is still very close to zero (−0.002) and not significant. Again the social capital coefficient does not change much (since the results are very similar to those in Table 3, we omit their presentation in a table). Table 3 Conditional mean estimates on growth controlling for spatial externalities (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.819*** 0.708*** 0.319*** 0.217*** 0.163** 0.562*** (0.030) (0.036) (0.039) (0.068) (0.069) (0.110) Initial income −0.306*** −0.285*** −0.395*** −0.903*** −0.905*** −0.624*** (0.004) (0.006) (0.007) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.382 0.388 0.441 0.682 0.683 0.409 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.819*** 0.708*** 0.319*** 0.217*** 0.163** 0.562*** (0.030) (0.036) (0.039) (0.068) (0.069) (0.110) Initial income −0.306*** −0.285*** −0.395*** −0.903*** −0.905*** −0.624*** (0.004) (0.006) (0.007) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.382 0.388 0.441 0.682 0.683 0.409 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 3 Conditional mean estimates on growth controlling for spatial externalities (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.819*** 0.708*** 0.319*** 0.217*** 0.163** 0.562*** (0.030) (0.036) (0.039) (0.068) (0.069) (0.110) Initial income −0.306*** −0.285*** −0.395*** −0.903*** −0.905*** −0.624*** (0.004) (0.006) (0.007) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.382 0.388 0.441 0.682 0.683 0.409 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.819*** 0.708*** 0.319*** 0.217*** 0.163** 0.562*** (0.030) (0.036) (0.039) (0.068) (0.069) (0.110) Initial income −0.306*** −0.285*** −0.395*** −0.903*** −0.905*** −0.624*** (0.004) (0.006) (0.007) (0.010) (0.010) (0.024) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared 0.382 0.388 0.441 0.682 0.683 0.409 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. We also estimate a spatial panel Arellano–Bond linear dynamic regression using a spatial weight matrix based on inverse distance. In this model, the social capital coefficient is equal to 0.525 with a standard error equal to 0.235, therefore being significant at 5%. This compares favourably with our main estimate of 0.595, which is significant at 1% level. The implication is that, while spatial correlations may in principle play a role, there is no evidence of such a role being crucial in our sample. 5.3. Non-linearities Some explanatory variables may also affect growth in a way that depends on the level of other explanatory variables, such as deeply lagged population or initial income level.21 In Table 4, we replicated Table 2 using both (social capital) and (social capital × population in 1951) as covariates. The evidence suggests that population in 1951 does not generally affect the growth return to social capital. The intuition is that controlling for initial income and population makes the population size in 1951 redundant in the model. Table 4 Conditional mean estimates on growth controlling for interaction between social capital and population size in 1951 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.705*** 0.797*** 0.325*** −0.627 −0.676 0.558*** (0.0335) (0.0376) (0.0427) (0.782) (0.784) (0.102) Initial income −0.248*** −0.294*** −0.411*** −0.821*** −0.820*** −0.654*** (0.00362) (0.00662) (0.00725) (0.00955) (0.00956) (0.0239) Social capital × 0.000249 0.00148 −0.0129*** 0.0775 0.0813 0.00579 Population in 1951 (0.000956) (0.000944) (0.00255) (0.0852) (0.0853) (0.00542) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.335 0.366 0.435 0.616 0.616 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.705*** 0.797*** 0.325*** −0.627 −0.676 0.558*** (0.0335) (0.0376) (0.0427) (0.782) (0.784) (0.102) Initial income −0.248*** −0.294*** −0.411*** −0.821*** −0.820*** −0.654*** (0.00362) (0.00662) (0.00725) (0.00955) (0.00956) (0.0239) Social capital × 0.000249 0.00148 −0.0129*** 0.0775 0.0813 0.00579 Population in 1951 (0.000956) (0.000944) (0.00255) (0.0852) (0.0853) (0.00542) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.335 0.366 0.435 0.616 0.616 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 4 Conditional mean estimates on growth controlling for interaction between social capital and population size in 1951 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.705*** 0.797*** 0.325*** −0.627 −0.676 0.558*** (0.0335) (0.0376) (0.0427) (0.782) (0.784) (0.102) Initial income −0.248*** −0.294*** −0.411*** −0.821*** −0.820*** −0.654*** (0.00362) (0.00662) (0.00725) (0.00955) (0.00956) (0.0239) Social capital × 0.000249 0.00148 −0.0129*** 0.0775 0.0813 0.00579 Population in 1951 (0.000956) (0.000944) (0.00255) (0.0852) (0.0853) (0.00542) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.335 0.366 0.435 0.616 0.616 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.705*** 0.797*** 0.325*** −0.627 −0.676 0.558*** (0.0335) (0.0376) (0.0427) (0.782) (0.784) (0.102) Initial income −0.248*** −0.294*** −0.411*** −0.821*** −0.820*** −0.654*** (0.00362) (0.00662) (0.00725) (0.00955) (0.00956) (0.0239) Social capital × 0.000249 0.00148 −0.0129*** 0.0775 0.0813 0.00579 Population in 1951 (0.000956) (0.000944) (0.00255) (0.0852) (0.0853) (0.00542) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.335 0.366 0.435 0.616 0.616 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. As a second exercise, we augment model (1) with the interaction between social capital and the initial value added (Table 5). The results, however, are not much different from those in Table 2 and the coefficient of the interaction term is very small in magnitude. Furthermore, when we estimate the specification with the GMM-SYS method, the coefficient of the interaction is not significant. Hence, we do not find evidence of significant non-linearities, with respect to the initial-income level. Table 5 Conditional mean estimates on growth controlling for interaction between social capital and initial income (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.753*** 0.451*** 0.096** 0.263* 0.216*** 0.653*** (0.030) (0.030) (0.041) (0.070) (0.071) (0.121) Initial income −0.406*** −0.400*** −0.475*** −0.900*** −0.901*** −0.675*** (0.009) (0.009) (0.009) (0.010) (0.010) (0.028) Social capital × 0.022*** 0.034*** 0.027*** −0.004** −0.005** −0.004 Initial income (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.386 0.409 0.454 0.682 0.683 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.753*** 0.451*** 0.096** 0.263* 0.216*** 0.653*** (0.030) (0.030) (0.041) (0.070) (0.071) (0.121) Initial income −0.406*** −0.400*** −0.475*** −0.900*** −0.901*** −0.675*** (0.009) (0.009) (0.009) (0.010) (0.010) (0.028) Social capital × 0.022*** 0.034*** 0.027*** −0.004** −0.005** −0.004 Initial income (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.386 0.409 0.454 0.682 0.683 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 5 Conditional mean estimates on growth controlling for interaction between social capital and initial income (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.753*** 0.451*** 0.096** 0.263* 0.216*** 0.653*** (0.030) (0.030) (0.041) (0.070) (0.071) (0.121) Initial income −0.406*** −0.400*** −0.475*** −0.900*** −0.901*** −0.675*** (0.009) (0.009) (0.009) (0.010) (0.010) (0.028) Social capital × 0.022*** 0.034*** 0.027*** −0.004** −0.005** −0.004 Initial income (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.386 0.409 0.454 0.682 0.683 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.753*** 0.451*** 0.096** 0.263* 0.216*** 0.653*** (0.030) (0.030) (0.041) (0.070) (0.071) (0.121) Initial income −0.406*** −0.400*** −0.475*** −0.900*** −0.901*** −0.675*** (0.009) (0.009) (0.009) (0.010) (0.010) (0.028) Social capital × 0.022*** 0.034*** 0.027*** −0.004** −0.005** −0.004 Initial income (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.386 0.409 0.454 0.682 0.683 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. In addition, we consider the possibility of differentiated returns between the Italian macro areas. In line with Putnam (1993), the descriptive statistics suggest that the level of social capital is lower in the South, with an average of 85% against an average of 94% in the Centre–North in the 1953–1992 period. Thus, one may expect the productivity of social capital to be lower where the level of social capital is higher (law of decreasing marginal productivity). To test this hypothesis, we replicate the analysis in Table 2 by adding an interaction term between social capital and a dummy for South. In this model, the coefficient of social capital alone represents the impact of social capital on growth in the Centre–North. The impact in the South is given by the sum between the coefficient of social capital alone and the coefficient of the interaction term (social capital × South). The regression analysis is somehow surprising. In our preferred specification, the elasticity of social capital with respect to growth in the South is 0.551 (Table 6), being 79% of the elasticity for the rest of Italy (0.696). This suggests that not only the level but also the return of social capital is lower in the South. Table 6 Conditional mean estimates on growth controlling for interaction between social capital and South (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.463*** 0.420*** 0.324*** 0.341** 0.296** 0.696*** (0.034) (0.036) (0.038) (0.134) (0.134) (0.107) Initial income −0.340*** −0.348*** −0.381*** −0.903*** −0.905*** −0.661*** (0.004) (0.007) (0.007) (0.010) (0.010) (0.024) Social capital × −0.037*** −0.039*** −0.046*** −0.141 −0.149 −0.145*** South (0.001) (0.001) (0.001) (0.137) (0.136) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.408 0.422 0.436 0.682 0.683 0.423 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.463*** 0.420*** 0.324*** 0.341** 0.296** 0.696*** (0.034) (0.036) (0.038) (0.134) (0.134) (0.107) Initial income −0.340*** −0.348*** −0.381*** −0.903*** −0.905*** −0.661*** (0.004) (0.007) (0.007) (0.010) (0.010) (0.024) Social capital × −0.037*** −0.039*** −0.046*** −0.141 −0.149 −0.145*** South (0.001) (0.001) (0.001) (0.137) (0.136) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.408 0.422 0.436 0.682 0.683 0.423 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 6 Conditional mean estimates on growth controlling for interaction between social capital and South (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.463*** 0.420*** 0.324*** 0.341** 0.296** 0.696*** (0.034) (0.036) (0.038) (0.134) (0.134) (0.107) Initial income −0.340*** −0.348*** −0.381*** −0.903*** −0.905*** −0.661*** (0.004) (0.007) (0.007) (0.010) (0.010) (0.024) Social capital × −0.037*** −0.039*** −0.046*** −0.141 −0.149 −0.145*** South (0.001) (0.001) (0.001) (0.137) (0.136) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.408 0.422 0.436 0.682 0.683 0.423 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.463*** 0.420*** 0.324*** 0.341** 0.296** 0.696*** (0.034) (0.036) (0.038) (0.134) (0.134) (0.107) Initial income −0.340*** −0.348*** −0.381*** −0.903*** −0.905*** −0.661*** (0.004) (0.007) (0.007) (0.010) (0.010) (0.024) Social capital × −0.037*** −0.039*** −0.046*** −0.141 −0.149 −0.145*** South (0.001) (0.001) (0.001) (0.137) (0.136) (0.004) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.408 0.422 0.436 0.682 0.683 0.423 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Finally, we consider the possibility that the link between social capital and growth varies with time. To investigate this issue, we allow for the social capital coefficient to change over time by interacting social capital with time dummies (Table 7). The coefficient of social capital at baseline represents the impact of social capital on the 1951–1961 growth rate. This number is 1.055 in our preferred specification. The interaction coefficients are always significant and negative. This means that social capital matters more for growth in the 1951–1961 period than after.22 The latter is another important result of our analysis. A similar result is presented by Albanese and de Blasio (2014) who find evidence of a positive and time-decreasing correlation between voter turnout and measures of economic development other than growth, at the municipality level of aggregation. Table 7 Conditional mean estimates on growth controlling for interaction between social capital and time effects (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.773*** 1.391*** 0.709*** 0.548*** 0.573*** 1.055*** (0.035) (0.144) (0.141) (0.161) (0.162) (0.195) Social capital × 1961–1971 0.034*** −0.562*** −0.544*** −0.421*** −0.451*** −0.672*** (0.002) (0.167) (0.159) (0.146) (0.148) (0.218) Social capital × 1971–1981 0.049*** −0.416*** −0.326** −0.268* −0.304** −0.515*** (0.002) (0.160) (0.152) (0.145) (0.148) (0.198) Social capital × 1981–1991 0.040*** −0.983*** −0.777*** −0.683*** −0.720*** −0.966*** (0.003) (0.153) (0.146) (0.143) (0.147) (0.195) Social capital × 1991–2001 0.026*** −0.384** −0.218 −0.251* −0.295** −0.488** (0.003) (0.153) (0.146) (0.144) (0.149) (0.196) Initial income −0.291*** −0.293*** −0.410*** −0.819*** −0.819*** −0.597*** (0.006) (0.006) (0.007) (0.009) (0.009) (0.023) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.366 0.371 0.437 0.618 0.618 0.511 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.773*** 1.391*** 0.709*** 0.548*** 0.573*** 1.055*** (0.035) (0.144) (0.141) (0.161) (0.162) (0.195) Social capital × 1961–1971 0.034*** −0.562*** −0.544*** −0.421*** −0.451*** −0.672*** (0.002) (0.167) (0.159) (0.146) (0.148) (0.218) Social capital × 1971–1981 0.049*** −0.416*** −0.326** −0.268* −0.304** −0.515*** (0.002) (0.160) (0.152) (0.145) (0.148) (0.198) Social capital × 1981–1991 0.040*** −0.983*** −0.777*** −0.683*** −0.720*** −0.966*** (0.003) (0.153) (0.146) (0.143) (0.147) (0.195) Social capital × 1991–2001 0.026*** −0.384** −0.218 −0.251* −0.295** −0.488** (0.003) (0.153) (0.146) (0.144) (0.149) (0.196) Initial income −0.291*** −0.293*** −0.410*** −0.819*** −0.819*** −0.597*** (0.006) (0.006) (0.007) (0.009) (0.009) (0.023) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.366 0.371 0.437 0.618 0.618 0.511 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 7 Conditional mean estimates on growth controlling for interaction between social capital and time effects (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.773*** 1.391*** 0.709*** 0.548*** 0.573*** 1.055*** (0.035) (0.144) (0.141) (0.161) (0.162) (0.195) Social capital × 1961–1971 0.034*** −0.562*** −0.544*** −0.421*** −0.451*** −0.672*** (0.002) (0.167) (0.159) (0.146) (0.148) (0.218) Social capital × 1971–1981 0.049*** −0.416*** −0.326** −0.268* −0.304** −0.515*** (0.002) (0.160) (0.152) (0.145) (0.148) (0.198) Social capital × 1981–1991 0.040*** −0.983*** −0.777*** −0.683*** −0.720*** −0.966*** (0.003) (0.153) (0.146) (0.143) (0.147) (0.195) Social capital × 1991–2001 0.026*** −0.384** −0.218 −0.251* −0.295** −0.488** (0.003) (0.153) (0.146) (0.144) (0.149) (0.196) Initial income −0.291*** −0.293*** −0.410*** −0.819*** −0.819*** −0.597*** (0.006) (0.006) (0.007) (0.009) (0.009) (0.023) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.366 0.371 0.437 0.618 0.618 0.511 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.773*** 1.391*** 0.709*** 0.548*** 0.573*** 1.055*** (0.035) (0.144) (0.141) (0.161) (0.162) (0.195) Social capital × 1961–1971 0.034*** −0.562*** −0.544*** −0.421*** −0.451*** −0.672*** (0.002) (0.167) (0.159) (0.146) (0.148) (0.218) Social capital × 1971–1981 0.049*** −0.416*** −0.326** −0.268* −0.304** −0.515*** (0.002) (0.160) (0.152) (0.145) (0.148) (0.198) Social capital × 1981–1991 0.040*** −0.983*** −0.777*** −0.683*** −0.720*** −0.966*** (0.003) (0.153) (0.146) (0.143) (0.147) (0.195) Social capital × 1991–2001 0.026*** −0.384** −0.218 −0.251* −0.295** −0.488** (0.003) (0.153) (0.146) (0.144) (0.149) (0.196) Initial income −0.291*** −0.293*** −0.410*** −0.819*** −0.819*** −0.597*** (0.006) (0.006) (0.007) (0.009) (0.009) (0.023) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.366 0.371 0.437 0.618 0.618 0.511 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Figure 4 provides a clearer picture. The average impact of social capital on growth hides some heterogeneity over time. The role of social capital is bigger immediately after WWII when its level was higher (Figure 2). This result is consistent with the one for the macro areas (South vs. Centre–North). In short, we find that the return to social capital is higher when and where the social capital is higher. Figure 4 View largeDownload slide The impact of social capital on real growth over time. Notes: The horizontal lines are based on Table 2, Column 6. The other lines are based on Table 8, Column 6. Figure 4 View largeDownload slide The impact of social capital on real growth over time. Notes: The horizontal lines are based on Table 2, Column 6. The other lines are based on Table 8, Column 6. The fact that the elasticity of social capital is bigger in the 1950s may have several explanations. Our preferred one is in line with Ahlerup et al. (2009) who suggest that the effect of social capital on economic performance is higher when formal institutions are less developed. Indeed, since the Costituzione della Repubblica italiana was enforced in 1948 and radically changed the Italian formal institutional system, it is reasonable to think that Italy was a country under construction during the 1950s—which explains why social capital may have played a more important role. Of course, additional factors may be at work. For instance, a higher quantity of social capital may imply a higher quality of social capital, and thus higher productivity. Such a mechanism would be in line with what some authors (among others, see Hungerford and Solon, 1987; Budría and Pereira, 2011) have argued to hold for human capital.23 However, further investigation on this hypothesis requires municipality-level measures of social–capital quality which, to our knowledge, are not available. 5.4. Different outcomes As a further robustness check, we also estimate GMM-SYS models with different outcomes using municipality-level data. In particular, Table 8 considers the growth rates of the following variables as additional outcomes24: Plant density (plants over municipality square kilometres) (Column 2). Employees density (employees over municipality square kilometres) (Column 3). Employees per capita (employees over population) (Column 4). Table 8 Conditional mean estimates on different outcomes: changing the dependent variable (measured as growth rate) (1) (2) (3) (4) GMM-SYS GMM-SYS GMM-SYS GMM-SYS Real GDP per capita Plant density Employees density Employees per capita Social capital 0.595*** 0.477** 0.522* 0.721*** (0.109) (0.223) (0.284) (0.230) Initial income −0.660*** −0.112*** −0.0853 −1.530*** (0.0247) (0.0369) (0.0538) (0.0679) Observations 10,145 10,145 10,145 10,145 G. of fit 0.407 0.195 0.243 0.290 (1) (2) (3) (4) GMM-SYS GMM-SYS GMM-SYS GMM-SYS Real GDP per capita Plant density Employees density Employees per capita Social capital 0.595*** 0.477** 0.522* 0.721*** (0.109) (0.223) (0.284) (0.230) Initial income −0.660*** −0.112*** −0.0853 −1.530*** (0.0247) (0.0369) (0.0538) (0.0679) Observations 10,145 10,145 10,145 10,145 G. of fit 0.407 0.195 0.243 0.290 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 8 Conditional mean estimates on different outcomes: changing the dependent variable (measured as growth rate) (1) (2) (3) (4) GMM-SYS GMM-SYS GMM-SYS GMM-SYS Real GDP per capita Plant density Employees density Employees per capita Social capital 0.595*** 0.477** 0.522* 0.721*** (0.109) (0.223) (0.284) (0.230) Initial income −0.660*** −0.112*** −0.0853 −1.530*** (0.0247) (0.0369) (0.0538) (0.0679) Observations 10,145 10,145 10,145 10,145 G. of fit 0.407 0.195 0.243 0.290 (1) (2) (3) (4) GMM-SYS GMM-SYS GMM-SYS GMM-SYS Real GDP per capita Plant density Employees density Employees per capita Social capital 0.595*** 0.477** 0.522* 0.721*** (0.109) (0.223) (0.284) (0.230) Initial income −0.660*** −0.112*** −0.0853 −1.530*** (0.0247) (0.0369) (0.0538) (0.0679) Observations 10,145 10,145 10,145 10,145 G. of fit 0.407 0.195 0.243 0.290 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. The estimates are compared with our main GMM-SYS results (Column 1), which are based on the growth rate of real GDP per capita as outcome. The results suggest that the effect of social capital changes with the outcome chosen, but it is generally positive and significant, as one would reasonably expect. 5.5. Sample selection We work with a sample of municipalities with at least 5000 inhabitants in 1951 because identification is problematic with the full sample. The results obtained with the full sample are provided in the Appendix Table A5. Even though the general picture is very close to the one obtained with the sample of municipalities with at least 5000 inhabitants in 1951, estimates in Column 6 pass neither the Arellano–Bond test nor the Hansen and the Difference-in-Hansen tests. In particular, we find a lower return to social capital than in the restricted sample (both coefficients are significant). There are two possible explanations for this difference in returns. One is that restricted sample estimates are not reliable because of sample selection bias (Hypothesis I). Another one is that full sample estimates are not reliable because of lack of model identification (Hypothesis II). As it is known, sample selection bias occurs when the probability of being in the sample, i.e., the probability of the municipality population being higher than 5000 inhabitants in 1951, is affected by unobservables that are correlated with the unobservables of the growth equation. Here, the role of housing prices is key. Housing prices in 1951 are unobservables in our dataset, and this may create a correlation between the residuals of the growth equation and those of the ‘participation equation’ (i.e., the equation modelling the probability of being in the sample). Housing prices in 1951 are likely to affect population size in 1951 and thus the probability of being in the sample. Thus, if the growth return to social capital is responsive to population size in 1951 in the full sample, this may be an indication of selection bias in the restricted sample. We test the sample-selection bias hypothesis (Hypothesis I) in Table 9 and reject it because the interaction term (social capital × population in 1951) is not significant. This reinforces our view that using the restricted sample is preferable because identification tests are passed in the restricted sample, while they are passed neither in the full sample nor in the sample of municipalities with less than 5000 inhabitants in 1951. Table 9 Conditional mean estimates on growth changing the sample size (1) (2) (3) GMM-SYS GMM-SYS GMM-SYS Full sample Below 5000 Above 5000 Social capital 0.522*** 0.254** 0.558*** (0.0748) (0.104) (0.102) Initial income −0.766*** −0.811*** −0.654*** (0.0184) (0.0197) (0.0239) Social capital × Population in 1951 −0.0127 0.0132 0.00579 (0.00923) (0.0119) (0.00542) Observations 40,185 30,040 10,145 G. of fit 0.398 0.393 0.408 (1) (2) (3) GMM-SYS GMM-SYS GMM-SYS Full sample Below 5000 Above 5000 Social capital 0.522*** 0.254** 0.558*** (0.0748) (0.104) (0.102) Initial income −0.766*** −0.811*** −0.654*** (0.0184) (0.0197) (0.0239) Social capital × Population in 1951 −0.0127 0.0132 0.00579 (0.00923) (0.0119) (0.00542) Observations 40,185 30,040 10,145 G. of fit 0.398 0.393 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 9 Conditional mean estimates on growth changing the sample size (1) (2) (3) GMM-SYS GMM-SYS GMM-SYS Full sample Below 5000 Above 5000 Social capital 0.522*** 0.254** 0.558*** (0.0748) (0.104) (0.102) Initial income −0.766*** −0.811*** −0.654*** (0.0184) (0.0197) (0.0239) Social capital × Population in 1951 −0.0127 0.0132 0.00579 (0.00923) (0.0119) (0.00542) Observations 40,185 30,040 10,145 G. of fit 0.398 0.393 0.408 (1) (2) (3) GMM-SYS GMM-SYS GMM-SYS Full sample Below 5000 Above 5000 Social capital 0.522*** 0.254** 0.558*** (0.0748) (0.104) (0.102) Initial income −0.766*** −0.811*** −0.654*** (0.0184) (0.0197) (0.0239) Social capital × Population in 1951 −0.0127 0.0132 0.00579 (0.00923) (0.0119) (0.00542) Observations 40,185 30,040 10,145 G. of fit 0.398 0.393 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Overall, disregarding municipalities with less than 5000 inhabitants in 1951 does not appear to be a major issue because about 77% of the Italian population is concentrated, on average, in the sample of cities that we use (a minimum of 74% occurs in 1951). In addition, earlier studies, including Guiso et al. (2016), have used municipalities with more than 5000 inhabitants, which makes our restricted sample results more comparable with the previous ones. 5.6. Province-time effects In Table 10, we replicate Table 2 with the only difference being the introduction of additional controls in the form of (province × time) interactions. We find that the coefficient of social capital slightly increases from 0.595 to 0.650 in the main (GMM-SYS) specification. However, the introduction of these additional covariates implies a substantial increase in the number of instruments, which weakens the identification tests, in line with Roodman (2009), even after collapsing the instrument matrix. In addition, the standard error goes from 0.109 to 0.309, with a clear loss of efficiency. Table 10 Conditional mean estimates on growth controlling for province-time effects (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.464*** 0.465*** 0.347*** 0.181** 0.171** 0.650** (0.0480) (0.0478) (0.0484) (0.0758) (0.0770) (0.309) Initial income −0.364*** −0.367*** −0.426*** −0.897*** −0.897*** −0.747*** (0.00739) (0.00729) (0.00802) (0.0107) (0.0107) (0.0502) Province-time effects Yes Yes Yes Yes Yes Yes Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.495 0.499 0.513 0.667 0.667 0.389 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.464*** 0.465*** 0.347*** 0.181** 0.171** 0.650** (0.0480) (0.0478) (0.0484) (0.0758) (0.0770) (0.309) Initial income −0.364*** −0.367*** −0.426*** −0.897*** −0.897*** −0.747*** (0.00739) (0.00729) (0.00802) (0.0107) (0.0107) (0.0502) Province-time effects Yes Yes Yes Yes Yes Yes Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.495 0.499 0.513 0.667 0.667 0.389 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 10 Conditional mean estimates on growth controlling for province-time effects (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.464*** 0.465*** 0.347*** 0.181** 0.171** 0.650** (0.0480) (0.0478) (0.0484) (0.0758) (0.0770) (0.309) Initial income −0.364*** −0.367*** −0.426*** −0.897*** −0.897*** −0.747*** (0.00739) (0.00729) (0.00802) (0.0107) (0.0107) (0.0502) Province-time effects Yes Yes Yes Yes Yes Yes Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.495 0.499 0.513 0.667 0.667 0.389 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.464*** 0.465*** 0.347*** 0.181** 0.171** 0.650** (0.0480) (0.0478) (0.0484) (0.0758) (0.0770) (0.309) Initial income −0.364*** −0.367*** −0.426*** −0.897*** −0.897*** −0.747*** (0.00739) (0.00729) (0.00802) (0.0107) (0.0107) (0.0502) Province-time effects Yes Yes Yes Yes Yes Yes Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.495 0.499 0.513 0.667 0.667 0.389 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. 5.7. Physical and human capital As another check, we control for a proxy of physical capital endowment (Table 11). The latter is measured as the number of plants per inhabitant at the beginning of the period (i.e., at year t−10). Although this variable is a candidate for the main specification, we find the coefficient of physical capital to be close to zero from the left (−0.051) and not significant (p-value = 0.230). As a result, the coefficient of social capital does not change much. One possible explanation is that controlling for the initial income level already captures the effect of the initial physical capital stock. Table 11 Conditional mean estimates on growth controlling for physical capital endowment (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.842*** 0.715*** 0.278*** 0.184*** 0.125* 0.585*** (0.030) (0.035) (0.039) (0.067) (0.068) (0.117) Initial income −0.334*** −0.319*** −0.434*** −0.929*** −0.933*** −0.639*** (0.006) (0.008) (0.008) (0.011) (0.011) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.378 0.392 0.447 0.684 0.685 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.842*** 0.715*** 0.278*** 0.184*** 0.125* 0.585*** (0.030) (0.035) (0.039) (0.067) (0.068) (0.117) Initial income −0.334*** −0.319*** −0.434*** −0.929*** −0.933*** −0.639*** (0.006) (0.008) (0.008) (0.011) (0.011) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.378 0.392 0.447 0.684 0.685 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 11 Conditional mean estimates on growth controlling for physical capital endowment (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.842*** 0.715*** 0.278*** 0.184*** 0.125* 0.585*** (0.030) (0.035) (0.039) (0.067) (0.068) (0.117) Initial income −0.334*** −0.319*** −0.434*** −0.929*** −0.933*** −0.639*** (0.006) (0.008) (0.008) (0.011) (0.011) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.378 0.392 0.447 0.684 0.685 0.408 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.842*** 0.715*** 0.278*** 0.184*** 0.125* 0.585*** (0.030) (0.035) (0.039) (0.067) (0.068) (0.117) Initial income −0.334*** −0.319*** −0.434*** −0.929*** −0.933*** −0.639*** (0.006) (0.008) (0.008) (0.011) (0.011) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 10,145 10,145 10,145 10,145 10,145 10,145 R-squared/G. of fit 0.378 0.392 0.447 0.684 0.685 0.408 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. In addition, we control for the roles that both physical and human capital endowments (at year t−10) as well as the evolution of working age population (from year t−10 to year t) may play in a growth regression (Table 12). As proxies for human capital endowments, we use both the share of population with high-school diploma and the share of population with university degree. The evolution of the working age population is the growth rate of the population aged between 15 and 64 years. These variables should, in principle, be included among the controls in the main specification.25 However, municipality-level data on both human capital and working age population are available in electronic format only since 1971. Hence, this exercise is restricted to a shorter time period. Nevertheless, the positive link between social capital and growth is confirmed and the magnitude of the social capital coefficient decreases. This is consistent with the idea that the effect of social capital on growth is higher in the 1950s. Table 12 Conditional mean estimates on growth controlling for physical and human capital endowments as well as the evolution of the working age population (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.772*** 0.502*** 0.162*** 0.013 −0.023 0.481*** (0.032) (0.036) (0.039) (0.075) (0.076) (0.14) Initial income −0.323*** −0.323*** −0.416*** −1.049*** −1.051*** −0.799*** (0.010) (0.009) (0.010) (0.014) (0.014) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 6087 6087 6087 6087 6087 6087 R-squared/G. of fit 0.332 0.362 0.412 0.716 0.717 0.411 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.772*** 0.502*** 0.162*** 0.013 −0.023 0.481*** (0.032) (0.036) (0.039) (0.075) (0.076) (0.14) Initial income −0.323*** −0.323*** −0.416*** −1.049*** −1.051*** −0.799*** (0.010) (0.009) (0.010) (0.014) (0.014) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 6087 6087 6087 6087 6087 6087 R-squared/G. of fit 0.332 0.362 0.412 0.716 0.717 0.411 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 12 Conditional mean estimates on growth controlling for physical and human capital endowments as well as the evolution of the working age population (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.772*** 0.502*** 0.162*** 0.013 −0.023 0.481*** (0.032) (0.036) (0.039) (0.075) (0.076) (0.14) Initial income −0.323*** −0.323*** −0.416*** −1.049*** −1.051*** −0.799*** (0.010) (0.009) (0.010) (0.014) (0.014) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 6087 6087 6087 6087 6087 6087 R-squared/G. of fit 0.332 0.362 0.412 0.716 0.717 0.411 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.772*** 0.502*** 0.162*** 0.013 −0.023 0.481*** (0.032) (0.036) (0.039) (0.075) (0.076) (0.14) Initial income −0.323*** −0.323*** −0.416*** −1.049*** −1.051*** −0.799*** (0.010) (0.009) (0.010) (0.014) (0.014) (0.028) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 6087 6087 6087 6087 6087 6087 R-squared/G. of fit 0.332 0.362 0.412 0.716 0.717 0.411 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. 5.8. Summing up Summing up, our analysis suggests the following picture. The link between initial social capital endowment and subsequent growth at municipality level is positive and significant. The result does not change when controlling for spatial externalities, when accounting for non-linearities in initial income and population, when accounting for province-time effects, when accounting for physical capital endowment and when accounting for both physical and human capital endowment of a city as well as for the evolution of the working age population. The results for different outcomes are consistent with those for the main outcome. However, the strength of the link between social capital and growth looks stronger in the Centre–North of Italy and in the 1950s. 6. Local estimates GMM-SYS uses lagged levels and differences of endogenous variables as instruments. These are called internal instruments. Yet, the identification based on internal instruments, while solving a number of statistical problems (including the Nickell bias in dynamic panel-data models, such as our model), is not appealing from a theoretical perspective because the causal chain between the instruments and the endogenous variables is much more opaque than in a 2SLS just-identified model. To address this criticism, this section proposes 2SLS estimates from two just-identified models, which are identified using two different municipality-level external instruments. In defense of our GMM-SYS approach, it should be stressed that 2SLS estimates have low external validity because they refer to the specific group of municipalities whose level of social capital was affected by the instrument. This is why 2SLS estimates are ‘local estimates’ in the spirit of Angrist and Pischke (2009). In contrast, GMM-SYS estimates are more likely to have external validity because all municipalities in the sample are affected by the instruments by construction (the instruments are internal). Even if our GMM-SYS estimates refer to a restricted sample (more than 5000 inhabitants in 1951), they are still of interest for a large, well-defined sub-population. This is generally not the case when estimating local effects with 2SLS. With the above caveats, our first instrument, inspired by Guiso et al. (2016), is a dummy equal to one if a municipality was a free city in 1300.26 The second instrument, inspired by Andini et al. (2017), is a dummy equal to one if the municipality was an incorporating city (i.e., a larger municipality incorporating a smaller one), as a result of a consolidation decision taken by Mussolini during the fascist dictatorship, in the 1920s. An instrument must be valid and relevant. In our 2SLS context, validity means that the instrument must be uncorrelated with the error term of the growth equation. This is like saying that, conditional on other regressors, the instrument must not affect growth directly (exclusion restriction; see also Bürker and Minerva, 2014, 830). To be relevant, the instrument must be (not weakly) correlated with the endogenous variable, i.e., voter turnout. As is well known, while the relevance condition can be theoretically argued and empirically tested, the validity of an instrument can only be theoretically argued. The main argument for validity here is that both historical events, on which our instruments are based, are too distant in the past to have a direct effect on growth from 1951 onwards, once we condition on initial income (see also Tabellini, 2010; Crescenzi et al., 2013; Bürker and Minerva, 2014). The idea behind the first-stage regression in Column (1) of Table 13 is well grounded in the work of Guiso et al. (2016), which highlights the transmission across generations of inhabitants, over long periods of time, of civic-capital characteristics (see also Bürker and Minerva, 2014). Following Guiso et al. (2016), we restrict the sample to the Centre–North due to the Norman domination in the South of Italy. The estimated coefficient is positive (0.042) and significant at 1% level, as expected. The relevance condition is supported by the first-stage F-statistic, which is well above the usual reference values. The latter is robust to the control set (Column 2), though the F-statistic decreases a lot. Table 13 Conditional mean estimates on social capital: first-stage regressions (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.0420*** 0.0185*** 0.0211*** 0.0161*** (0.00444) (0.00558) (0.00255) (0.00255) Initial income 0.0299*** 0.0260*** 0.0297*** 0.0253*** (0.00141) (0.00139) (0.00143) (0.00141) Constant 4.301*** 4.121*** 4.301*** 4.123*** (0.00933) (0.0169) (0.00939) (0.0168) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.050 0.624 0.050 0.621 F-statistic 89.47 10.99 68.57 39.67 (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.0420*** 0.0185*** 0.0211*** 0.0161*** (0.00444) (0.00558) (0.00255) (0.00255) Initial income 0.0299*** 0.0260*** 0.0297*** 0.0253*** (0.00141) (0.00139) (0.00143) (0.00141) Constant 4.301*** 4.121*** 4.301*** 4.123*** (0.00933) (0.0169) (0.00939) (0.0168) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.050 0.624 0.050 0.621 F-statistic 89.47 10.99 68.57 39.67 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 13 Conditional mean estimates on social capital: first-stage regressions (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.0420*** 0.0185*** 0.0211*** 0.0161*** (0.00444) (0.00558) (0.00255) (0.00255) Initial income 0.0299*** 0.0260*** 0.0297*** 0.0253*** (0.00141) (0.00139) (0.00143) (0.00141) Constant 4.301*** 4.121*** 4.301*** 4.123*** (0.00933) (0.0169) (0.00939) (0.0168) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.050 0.624 0.050 0.621 F-statistic 89.47 10.99 68.57 39.67 (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.0420*** 0.0185*** 0.0211*** 0.0161*** (0.00444) (0.00558) (0.00255) (0.00255) Initial income 0.0299*** 0.0260*** 0.0297*** 0.0253*** (0.00141) (0.00139) (0.00143) (0.00141) Constant 4.301*** 4.121*** 4.301*** 4.123*** (0.00933) (0.0169) (0.00939) (0.0168) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.050 0.624 0.050 0.621 F-statistic 89.47 10.99 68.57 39.67 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. The idea behind the first-stage regression in Column (3) of Table 13 requires a few more explanations. Andini et al. (2017) use the fascist consolidation of small municipalities (less than 5000 inhabitants) which were merged each other into a new born municipality to investigate how a substantial enlargement of small-sized cities affects the local welfare of inhabitants. Differently from Andini et al. (2017), we use the fascist consolidation of small municipalities which were merged to an already existing bigger municipality (more than 5000 inhabitants) to investigate how a small enlargement of already existing bigger municipalities changes the level of social capital in bigger cities (this is our first-stage regression in Column 3). Thus, we use the same historical event as Andini et al. (2017) but consider a different group of municipalities, which does not overlap with earlier work. The key argument for the relevance of our second instrument is that the level of social capital in a municipality, measured as voter turnout, is responsive to the fact that the municipality was an incorporating city in the past. This can happen for two reasons. The first one is mechanical: a consolidation is likely to change both the denominator (eligible voters) and the numerator (actual voters) of the turnout rate, both at the time it happens and from that moment onwards. The second one is theoretical: a consolidation is likely to increase the propensity to participate in democratic processes for people (the inhabitants of incorporating cities) who have been exposed to an additional dictator imposition (relative to the non-exposed, i.e., the inhabitants of non-incorporating cities). Following Guiso et al. (2016), such a characteristic is likely to be transmitted across generations of inhabitants. The estimated coefficient is positive (0.021) and significant at 1% level. The F-statistic ensures that the instrument is relevant. Notably, the result is robust to controlling, among other covariates, for population and surface (Column 4), with the implication that the status of incorporating city has a direct positive effect on voter turnout, independent of population and surface. Reduced-form estimates are provided in Table 14, reporting how each of the instruments affects the growth rate of GDP per capita through civic capital. The estimated coefficients are all positive and significant. The results in Columns (1) and (2) extend the work of Guiso et al. (2016) in an interesting direction. The findings in Columns (3) and (4) are in line with the evidence of Andini et al. (2017) that Mussolini’s consolidation has been welfare-enhancing, though our results are not strictly comparable with those of Andini et al. (2017) for the reasons explained above. Table 14 Conditional mean estimates on growth: reduced-form regressions (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.150*** 0.0697*** 0.0617*** 0.0468*** (0.0135) (0.0139) (0.00837) (0.00829) Initial income −0.280*** −0.286*** −0.279*** −0.288*** (0.00563) (0.00574) (0.00564) (0.00579) Constant 2.102*** 1.919*** 2.096*** 1.922*** (0.0363) (0.0496) (0.0362) (0.0492) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.318 0.469 0.317 0.469 (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.150*** 0.0697*** 0.0617*** 0.0468*** (0.0135) (0.0139) (0.00837) (0.00829) Initial income −0.280*** −0.286*** −0.279*** −0.288*** (0.00563) (0.00574) (0.00564) (0.00579) Constant 2.102*** 1.919*** 2.096*** 1.922*** (0.0363) (0.0496) (0.0362) (0.0492) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.318 0.469 0.317 0.469 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 14 Conditional mean estimates on growth: reduced-form regressions (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.150*** 0.0697*** 0.0617*** 0.0468*** (0.0135) (0.0139) (0.00837) (0.00829) Initial income −0.280*** −0.286*** −0.279*** −0.288*** (0.00563) (0.00574) (0.00564) (0.00579) Constant 2.102*** 1.919*** 2.096*** 1.922*** (0.0363) (0.0496) (0.0362) (0.0492) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.318 0.469 0.317 0.469 (1) (2) (3) (4) OLS OLS OLS OLS Guiso et al. Guiso et al. Andini et al. Andini et al. Instrument 0.150*** 0.0697*** 0.0617*** 0.0468*** (0.0135) (0.0139) (0.00837) (0.00829) Initial income −0.280*** −0.286*** −0.279*** −0.288*** (0.00563) (0.00574) (0.00564) (0.00579) Constant 2.102*** 1.919*** 2.096*** 1.922*** (0.0363) (0.0496) (0.0362) (0.0492) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 R-squared 0.318 0.469 0.317 0.469 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. As shown in Table 15, the general result that civic capital has a positive and significant effect on growth is confirmed by our 2SLS experiments, even after extending the conditioning set of covariates. Table 15 Conditional mean estimates on growth: second-stage regressions (1) (2) (3) (4) 2SLS 2SLS 2SLS 2SLS Guiso et al. Guiso et al. Andini et al. Andini et al. Social capital 3.562*** 3.765*** 2.924*** 2.911*** (0.432) (1.214) (0.433) (0.570) Initial income −0.386*** −0.384*** −0.366*** −0.362*** (0.0155) (0.0327) (0.0153) (0.0165) Constant −13.22*** −13.60*** −10.48*** −10.08*** (1.851) (4.991) (1.857) (2.343) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 (1) (2) (3) (4) 2SLS 2SLS 2SLS 2SLS Guiso et al. Guiso et al. Andini et al. Andini et al. Social capital 3.562*** 3.765*** 2.924*** 2.911*** (0.432) (1.214) (0.433) (0.570) Initial income −0.386*** −0.384*** −0.366*** −0.362*** (0.0155) (0.0327) (0.0153) (0.0165) Constant −13.22*** −13.60*** −10.48*** −10.08*** (1.851) (4.991) (1.857) (2.343) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table 15 Conditional mean estimates on growth: second-stage regressions (1) (2) (3) (4) 2SLS 2SLS 2SLS 2SLS Guiso et al. Guiso et al. Andini et al. Andini et al. Social capital 3.562*** 3.765*** 2.924*** 2.911*** (0.432) (1.214) (0.433) (0.570) Initial income −0.386*** −0.384*** −0.366*** −0.362*** (0.0155) (0.0327) (0.0153) (0.0165) Constant −13.22*** −13.60*** −10.48*** −10.08*** (1.851) (4.991) (1.857) (2.343) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 (1) (2) (3) (4) 2SLS 2SLS 2SLS 2SLS Guiso et al. Guiso et al. Andini et al. Andini et al. Social capital 3.562*** 3.765*** 2.924*** 2.911*** (0.432) (1.214) (0.433) (0.570) Initial income −0.386*** −0.384*** −0.366*** −0.362*** (0.0155) (0.0327) (0.0153) (0.0165) Constant −13.22*** −13.60*** −10.48*** −10.08*** (1.851) (4.991) (1.857) (2.343) Control set No Yes No Yes Time effects No Yes No Yes Province effects No Yes No Yes Observations 6006 6006 10,145 10,145 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. It is worth noting that 2SLS estimates based on different instruments refer to different local sub-populations and, therefore, these estimates do not need to be similar to each other27 and, most importantly, they do not need to be similar to our main GMM-SYS estimate (0.595), even if exactly the same growth model is estimated. This is in line with our results. 7. Conclusions In this paper, we have investigated the causal link between social capital and economic growth at a very detailed level of aggregation, the municipality. By using data on Italian municipalities over the 1951–2001 period, we have presented evidence of a positive and significant link between social capital endowment and subsequent economic growth. In particular, accounting for the endogeneity of all the explanatory variables as well as for unobserved heterogeneity, we have argued that the elasticity of social capital endowment with respect to growth is equal to 0.6, on average. The result has been shown to be robust to a variety of checks. We have also found that the link between social capital and growth is stronger in the Centre–North of Italy, where the elasticity is 0.7. Considering the time dimension, the link is found to be stronger in the 1950s, with an elasticity of 1.0. For specific sub-populations of municipalities, the elasticity is much bigger, with values ranging between 2.9 and 3.7. One limitation of this paper is that we have used only one measure of social capital and it is an indirect measure. Although the panel-fixed effects approach partly controls for measurement errors, future research could go in the direction of replicating our analysis using different measures of social capital and, in particular, direct measures of social capital at municipality level. Future research may also take advantage of direct measures of per capita GDP growth at municipality level, when they will become available for Italy. Another limitation is that we have implicitly assumed that the effect of social capital on growth is constant along the conditional growth distribution. The use of panel quantile regression techniques may shed light on the existence of quantile parameter heterogeneity (see Deng et al., 2012; Andini and Andini, 2014). This is important because social capital, besides shifting the mean of the conditional growth distribution towards the right, thus being a source of between-groups growth inequality, may be a factor changing the dispersion of the conditional growth distribution, thus being a source of within-groups growth inequality or a brake on it. Due to data limitations, we have not entered the debate about the quality and the quantity of social capital (Eliasson et al., 2013) as well as the issue of the heterogeneity between rural and urban areas (Westlund et al., 2014). Future studies on these topics are warranted. Another interesting topic for future research is that of testing the hypothesis that Italy is a special case (Westlund and Adam, 2010), by replicating our analysis using municipality data from other countries. It is possible that the use of appropriate municipality-level data brings out the underlying link between social capital and growth which is unclear in analyses based on national or regional-level data for countries other than Italy. Nevertheless, if one believes in the argument that the effect of social capital on economic performance is higher when and where the level of social capital is higher, it is likely that differences in the levels of social capital among countries would lead to differences in returns. So, our empirical results cannot be generalised. What can instead be generalised is the econometric approach to the issue that we have presented here. One reason why we look at growth as outcome is that, in our main theoretical frame, social capital affects non-measurable or hardly-measurable ‘intermediate’ outcomes such as trade uncertainty and free-riding behaviour making the production activity more efficient (Routledge and von Amsberg 2003). Yet, in a different theoretical frame such as a model by Akçomak and ter Weel (2009), social capital affects growth by increasing innovation. Future research may test this hypothesis by using specific measures of innovation at municipality level as dependent variables, corroborating the existing evidence on firm-level outcomes (Laursen et al., 2012). Following Sabatini (2008), a further interesting topic for the research agenda is the effect of social capital on broader measures of economic development, capturing the quality of economic development. Whether the link exists at the municipality level, it is an open issue. Footnotes 1 Knack and Keefer (1997) also argue that both trust and civic involvement improve the quality of governance, which in turn raises the quality of public-provided education and the return to human capital. Zak and Knack (2001) show that, in a context of moral hazard where more resources are allocated to inspection and monitoring, both the return to physical capital investment and the rate of investment decrease. Akçomak and ter Weel (2009) argue that the willingness to invest in innovation is higher when social capital is higher. Dasgupta (2011) shows that an increase in trust among members of a group will manifest itself in a total factor productivity growth. Papagapitos and Riley (2009) argue that higher trust leads to higher stocks of physical capital, which increase the productivity of human capital. Carlin et al. (2009) show that, in societies with no complete contracts, no punishment for opportunistic behaviour and/or no frequent interactions, more regulation results in less trust, which depresses aggregate investment and growth. However, Routledge and von Amsberg (2003) show that, in a model when the frequency of trade influences agent decisions to trade cooperatively, a reduction in social capital may have ambiguous effects on welfare. 2 We are aware of only two earlier studies that explore the link between social capital and growth using municipality-level data: Corazzini et al. (2011) and Eliasson et al. (2013). Yet, these works focus on correlation rather than causation and the evidence presented is cross-sectional. 3 We use the words municipality and city as referring to the same territorial unit. 4 The authors attribute these findings, in part, to the fact that the measures of social capital used (typically trust and association membership/density) are hardly able to capture the causal nexus from social capital to economic development when the analysis refers to spatial units characterised by a large number of unrelated actors, such as nations or regions. And, they conclude that it is crucial for future research to develop new and better measures of social capital at national and regional level, for instance for values like creativity, entrepreneurship and tolerance. 5 Laursen et al. (2012) suggest that social capital, measured as social interaction and political participation, increases firm ability to innovate as well as internal investments in R&D. Analogously, Guiso et al. (2004) show that social capital affects the behaviour of households and individuals, in the direction of boosting financial development. 6 Albanese and de Blasio (2014) use municipality-level data to study the impact of social capital on proxies of economic development (employment rate, employment density and plant density) in Italy. The authors estimate a static panel-data model with fixed effects, assuming social capital to be exogenous, and a number of cross-sectional models dealing with the endogeneity of social capital only. They find that social capital is beneficial for development. Westlund et al. (2014) investigate the influence of local entrepreneurial social capital on start-ups per capita, where the former is measured as the firm perception of local public attitudes to entrepreneurship and as the share of small businesses. It is found that both measures positively influence the start-up propensities among Swedish municipalities. The effect is stronger in rural areas. Analogously, Eliasson et al. (2013) report a positive relationship between enterprises’ opinion on the quality of the local business-related social capital (the local business climate) and economic growth at municipality level in Sweden. The authors also suggest that the quality of social capital may be more important than the quantity. A similar approach is taken by Corazzini et al. (2011) who use several measures of social cohesion, social division and attitude towards religion, estimating a growth model with data on Brazilian municipalities. Overall, they suggest a positive influence of social-capital indicators on growth. Finally, Rutten and Gelissen (2010) develop a theory of how social values influence economic development, which is tested in European regions. The evidence highlights that social values matter for subgroups in the population rather than the region as a whole. 7 For instance, the municipality issues the so-called tessera elettorale, which is the document needed for voting; citizens can only vote in the municipality of residence; the municipality chooses the specific infrastructure (typically a secondary school) where each citizen of the municipality is invited to vote; the municipality defines the composition of the committee that supervises the voting operations and enforces the application of the law; this committee certifies the identity of the voters in the day of the elections and provides them with the so-called scheda elettorale, which is the document where the voters are invited to express their preferences; this committee makes the counting of the votes and so on. 8 Some municipalities have been part of different provinces in different decades. We take this issue into account when we construct the value added. 9 To provide an example, suppose a province has two cities. Consider the manufacturing sector. For simplicity, assume the labour supply in each city is inelastic and both cities have the same labour supply. The labour-demand schedule in city 2 is above that of city 1 because of some labour-demand shifter (e.g., a better quality of machineries). The equilibrium wage in city 2 is thus initially bigger than in city 1. In this situation, workers would move from city 1 to city 2 (within a province, labour mobility is likely to be high because people can work in one city and be resident in another one). The decrease in the labour supply in city 1 would increase the wage in city 1. The increase in the labour supply in city 2 would decrease the wage in city 2. The flow of workers will only stop when the manufacturing wage in the two cities will be the same. This would imply the same labour productivity in the two cities. Thus, the productivity in each city will be the same as that in the province (the latter is the average of the productivities in the two cities), no matter whether labour demand conditions differ. A similar argument can be made if labour supply conditions differ. 10 Some attribute the real origin of the term to L.J. Hanifan, a social reformer, in 1916. 11 As suggested by Rutten et al. (2010), there are two main approaches to social capital: the ‘structuralist’ and the ‘interactionist’. The first investigates how network structures affect the performance of networks and of individuals within those networks. The second looks at the connection between association and obligation as well as the impact of this connection on how individuals behave (being associated to other individuals implies the obligation of sharing some form of ‘morality’, i.e., of norms, values, trust, rules, beliefs and codes). 12 As put by Sabatini (2008, 469), ‘repeated interactions among group members foster the diffusion of information and raise reputations’ relevance. The higher opportunity cost of free-riding in prisoners’ dilemma kind of situations makes agents’ behaviour more foreseeable causing an overall reduction of uncertainty. Therefore, an increase in trust-based relations may reduce the average cost of transactions, just as an increase in physical capital reduces the average cost of production’. 13 Of all the referenda occurred in Italy between 1946 and 2000, only two did not have the 50% cutoff because they did not involve the abrogation of an existing law (Monarchy vs. Republic in 1946; European Parliament in 1989). 14 As Keynes would put it, ‘It is not a case of choosing those [politicians] which, to the best of one’s judgment, are really the prettiest, nor even those that average opinion genuinely thinks the prettiest. We have reached the third degree where we devote our intelligences to anticipating what average opinion expects the average opinion to be. And there are some, I believe, who practice the fourth, fifth and higher degrees.’ (Keynes, 1936, 156). 15 Our proxy cannot be intended as an alternative to the ‘ideal’ measures proposed by Westlund and Adam (2010). 16 We measure a type of social capital that is likely to be losing importance (indeed, we will document that its return was higher in the 1950s). In the knowledge economy, other types of social capital (more business related) are likely to be gaining importance. But we cannot test this hypothesis because of lack of data. 17 We must acknowledge that our measure of social capital is not perfect, mainly because it measures ‘one type’ of social capital. 18 On the lines of Audretsch and Keilbach (2004), among others, we look at civic capital, proxied by voter turnout, as an additional component of a production function. The empirical model proposed in this paper is embedded in the standard endogenous growth theory (Eliasson et al., 2013) where cooperation among agents matters for growth. Our main theoretical reference is an article by Routledge and von Amsberg (2003) where social capital affects growth by reducing trade uncertainty and free-riding behaviour, thus making production activity more efficient. Another example is a model by Cozzi (1999) where R&D cooperation among firms is positive for growth under some assumptions about the state of technology. A recent example is a model by Akçomak and ter Weel (2009) where social capital affects growth by fostering innovation. 19 As argued by Sforza (2014), turnout significantly decreases if it is raining in the day of the elections. 20 The estimated coefficients of the additional control variables, not reported in Table 2, generally have the expected sign. 21 For instance, Durlauf et al. (2001) estimate a Solow's (1956) growth model in which the parameters vary according to the initial income level of a country, finding evidence of heterogeneity along that dimension. Contributions on the heterogeneity of the link between social capital and growth are quite few. For instance, Knack and Keefer (1997) include in their regression the interaction term ‘trust multiplied by initial income’ and find a negative coefficient. This implies that the effects of trust on growth might be stronger for poorer economies. Dearmon and Grier (2009) divide their sample in poorest and richest countries (respectively, the 25th and 75th percentiles of the income distribution) and show that there is no differential impact of social capital on growth between the two sub-samples. Ahlerup et al. (2009) find that the effect of social capital on economic performance is positive and decreases as formal institutions develop. On the same lines, Mauro and Pigliaru (2011) show that the strength of social capital as a determinant of long-run growth may depend on the institutional context. 22 Table 7 reports the p-values of the coefficients of all the interactions. Since all these coefficients are different from zero and negative, all the social-capital returns for the periods after 1951–1961 are different from the social-capital return in the period 1951–1961 (the reference period in the regression model) and they are lower. Thus, the return to social capital is higher in the 1950s than in any other decade until 2001. 23 As most of the important issues in economics, also this one is somehow controversial. Indeed, some studies have argued that the return to education may decrease with the education level, in line with the standard relative scarcity theory (see Psacharopoulos, 1985; Psacharopoulos, 1994; Psacharopoulos and Patrinos, 2004). 24 These are the outcomes chosen by Albanese and de Blasio (2014). We do not use population growth as outcome because, though it is positively associated with real GDP growth (0.319), as shown by the correlation matrix in Appendix Table A4, this paper studies the determinants of the growth rate of the real GDP per capita. So, population growth is unlikely to be a good proxy for our outcome variable, as suggested by the correlation coefficient in Appendix Table A4 (−0.012). 25 The importance of human capital as a determinant of growth has been widely documented (Barro, 1991; Mankiw et al., 1992; Temple, 2001). Unfortunately, we cannot directly control for other factors such as ethnic diversity (Platteau, 1994; Knack and Keefer, 1997; Alesina et al., 1999; Florida, 2002; Ottaviano and Peri, 2006), income inequality (Alesina and Rodrik, 1994; Alesina and Perotti, 1996) and the quality of governance (Hall and Jones, 1999; Acemoglu et al., 2001; Acemoglu and Robinson, 2006). However, the average quality of formal local institutions is somehow captured by our city-fixed effects. 26 Note that the estimates by Guiso et al. (2016) refer to municipalities with more than 5000 inhabitants, like ours. 27 A higher return is found in the sub-population of municipalities whose level of social capital has been affected by the free-city experience. Since these municipalities are located in the Centre–North of Italy, the higher-return result is consistent with our GMM-SYS evidence in Table 6. Acknowledgements For useful comments and suggestions, the authors would like to thank the editor Kristian Behrens, two anonymous referees, Guglielmo Barone, Antonio Maria Conti, Guido de Blasio, Steve Fleetwood, Paolo Sestito and the participants to the second workshop on ‘Capitale sociale, comportamenti e istituzioni’ (Bank of Italy, November 2013). The views expressed in this paper are those of the authors and do not necessarily correspond to those of the institutions they are affiliated with. References Abramovitz M. ( 1956 ) Resource and output trends in the United States since 1870 . American Economic Review , 46 : 5 – 23 . Acemoglu D. , Robinson J. ( 2006 ) De facto political power and institutional persistence . American Economic Review: Papers and Proceedings , 96 : 325 – 330 . Google Scholar CrossRef Search ADS Acemoglu D. , Johnson S. , Robinson J. ( 2001 ) The colonial origins of comparative development: an empirical investigation . American Economic Review , 91 : 1369 – 1401 . Google Scholar CrossRef Search ADS Ahlerup P. , Olsson O. , Yanagizawa D. ( 2009 ) Social capital vs institutions in the growth process . European Journal of Political Economy , 25 : 1 – 14 . Google Scholar CrossRef Search ADS Akçomak S. , ter Weel B. ( 2009 ) Social capital, innovation and growth: evidence from Europe . European Economic Review , 53 : 544 – 567 . Google Scholar CrossRef Search ADS Albanese G. , de Blasio G. ( 2014 ) Civic capital and development: Italy, 1951–2001. Quaderni di Storia Economica, No. 32, Banca d’Italia. Alesina A. , Baqir R. , Easterly W. ( 1999 ) Public goods and ethnic divisions . Quarterly Journal of Economics , 114 : 1243 – 1284 . Google Scholar CrossRef Search ADS Alesina A. , Perotti R. ( 1996 ) Income distribution, political instability, and investment . European Economic Review , 40 : 1203 – 1228 . Google Scholar CrossRef Search ADS Alesina A. , Rodrik D. ( 1994 ) Distributive politics and economic growth . Quarterly Journal of Economics , 109 : 465 – 490 . Google Scholar CrossRef Search ADS Algan Y. , Cahuc P. ( 2010 ) Inherited trust and growth . American Economic Review , 100 : 2060 – 2092 . Google Scholar CrossRef Search ADS Almond G. , Verba S. ( 1963 ) The Civic Culture: Political Attitudes and Democracy in Five Nations . Boston : Little, Brown & Co . Google Scholar CrossRef Search ADS Andini M. , Andini C. ( 2014 ) Finance, growth and quantile parameter heterogeneity . Journal of Macroeconomics , 40 : 308 – 322 . Google Scholar CrossRef Search ADS Andini M. , Dalmazzo A. , de Blasio G. ( 2017 ) The size of political jurisdictions: a model with some evidence from a fascist consolidation . Journal of Comparative Economics , 45 : 889 – 909 Google Scholar CrossRef Search ADS Angrist J. D. , Pischke J. S. ( 2009 ) Mostly Harmless Econometrics: An Empiricist’s Companion . Princeton : Princeton University Press . Arellano M. , Bond S. ( 1991 ) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations . Review of Economic Studies , 58 : 277 – 297 . Google Scholar CrossRef Search ADS Arellano M. , Bover O. ( 1995 ) Another look at the instrumental variable estimation of error-components models . Journal of Econometrics , 68 : 29 – 52 . Google Scholar CrossRef Search ADS Arrighetti A. , Lasagni A. , Seravalli G. ( 2003 ) Capitale sociale, associazionismo economico e istituzioni: indicatori sintetici di sintesi . Rivista di Politica Economica , 93 : 47 – 88 . Audretsch D. , Keilbach M. ( 2004 ) Entrepreneurship capital and economic performance . Regional Studies , 38 : 949 – 959 . Google Scholar CrossRef Search ADS Barro R. ( 1991 ) Economic growth in a cross section of countries . Quarterly Journal of Economics , 106 : 407 – 443 . Google Scholar CrossRef Search ADS Bloom N. , Bond S. , Van Reenen J. ( 2007 ) Uncertainty and investment dynamics . Review of Economic Studies , 74 : 391 – 415 . Google Scholar CrossRef Search ADS Blundell R. , Bond S. ( 1998 ) Initial conditions and moment restrictions in dynamic panel data models . Journal of Econometrics , 87 : 115 – 143 . Google Scholar CrossRef Search ADS Blundell R. , Bond S. ( 2000 ) GMM estimation with persistent panel data: an application to production functions . Econometric Reviews , 19 : 321 – 340 . Google Scholar CrossRef Search ADS Blundell R. , Bond S. , Windmeijer F. ( 2000 ) Estimation in dynamic panel data models: improving the performance of the standard GMM estimator. In: Baltagi B. , Fomby T. , Carter Hill R. (eds) Nonstationary Panels, Panel Cointegration and Dynamic Panels , pp. 53 – 91 . Amsterdam : Elsevier JAI . Google Scholar CrossRef Search ADS Bond S. ( 2002 ) Dynamic panel data models: a guide to micro data methods and practice . Portuguese Economic Journal , 1 : 141 – 162 . Google Scholar CrossRef Search ADS Bond S. , Hoeffler A. , Temple J. ( 2001 ) GMM estimation of empirical growth models. CEPR Discussion Papers, No. 3048, Centre for Economic Policy Research. Bourdieu P. ( 1972 ) Outline of a Theory of Practice . Cambridge : Cambridge University Press . Budría S. , Pereira P. ( 2011 ) Educational qualifications and wage inequality: evidence for Europe . Revista de Economia Aplicada , 19 : 5 – 34 . Bürker M. , Minerva G. A. ( 2014 ) Civic capital and the size distribution of plants: short-run dynamics and long-run equilibrium . Journal of Economic Geography , 14 : 797 – 847 . Google Scholar CrossRef Search ADS Carlin B. , Dorobantu F. , Viswanathan S. ( 2009 ) Public trust, the law, and financial investment . Journal of Financial Economics , 92 : 321 – 341 . Google Scholar CrossRef Search ADS Cassese S. , Battini S. , Franchini C. , Perez R. , Vesperini G. ( 2009 ) Manuale Di Diritto Pubblico . Milano : Giuffrè . Corazzini L. , Grazzi M. , Nicolini M. ( 2011 ) Social capital and growth in Brazilian municipalities. In: Nijkamp P. , Siedschlag I. (eds) Innovation, Growth and Competitiveness , pp. 195 – 217 . Heidelberg : Springer-Verlag . Google Scholar CrossRef Search ADS Cozzi G. ( 1999 ) R&D cooperation and growth . Journal of Economic Theory , 86 : 17 – 49 . Google Scholar CrossRef Search ADS Crescenzi R. , Gagliardi L. , Percoco M. ( 2013 ) Social capital and the innovative performance of Italian provinces . Environment and Planning A , 45 : 908 – 929 . Google Scholar CrossRef Search ADS Cuturi V. , Sampugnaro R. , Tomaselli V. ( 2000 ) L’elettore Instabile: Voto/Non Voto . Milano : Franco Angeli . Dasgupta P. ( 2011 ) A matter of trust: social capital and economic development. In: Lin J. , Plescovic B. (eds) Lessons from East Asia and the Global Financial Crisis , pp. 119 – 155 . Washington, DC : World Bank . Dearmon J. , Grier K. ( 2009 ) Trust and development . Journal of Economic Behavior & Organization , 71 : 210 – 220 . Google Scholar CrossRef Search ADS Deng W. , Lin Y. , Gong J. ( 2012 ) A smooth coefficient quantile regression approach to the social capital—economic growth nexus . Economic Modelling , 29 : 185 – 197 . Google Scholar CrossRef Search ADS Durlauf S. , Kourtellos A. , Minkin A. ( 2001 ) The local Solow growth model . European Economic Review , 45 : 928 – 940 . Google Scholar CrossRef Search ADS Eliasson K. , Westlund H. , Fölster S. ( 2013 ) Does social capital contribute to regional economic growth? Swedish experiences. In: Westlund H. , Kobayashi K. (eds) Social Capital and Rural Development in the Knowledge Society , pp. 113 – 126 . Cheltenham : Edward Elgar . Google Scholar CrossRef Search ADS Florida R. ( 2002 ) The Rise of the Creative Class . New York, NY : Basic Books . Fowler J. ( 2006 ) Altruism and turnout . Journal of Politics , 68 : 674 – 683 . Google Scholar CrossRef Search ADS Gallup J. L. , Sachs J. D. , Mellinger A. D. ( 1999 ) Geography and economic development . International Regional Science Review , 22 : 179 – 232 . Google Scholar CrossRef Search ADS Gennaioli N. , La Porta R. , Lopez-de-Silanes F. , Shleifer A. ( 2013 ) Human capital and regional development . Quarterly Journal of Economics , 128 : 105 – 164 . Google Scholar CrossRef Search ADS Glaeser E. L. , Laibson D. , Sacerdote B. ( 2002 ) An economic approach to social capital . Economic Journal , 112 : 437 – 458 . Google Scholar CrossRef Search ADS Guiso L. , Sapienza P. , Zingales L. ( 2004 ) The role of social capital in financial development . American Economic Review , 94 : 526 – 556 . Google Scholar CrossRef Search ADS Guiso L. , Sapienza P. , Zingales L. ( 2011 ) Civic capital as the missing link. In: Benhabib J. , Bisin A. , Jackson M. (eds) Handbook of Social Economics , pp. 417 – 480 . Amsterdam : North Holland . Guiso L. , Sapienza P. , Zingales L. ( 2016 ) Long-term persistence . Journal of the European Economic Association , 14 : 1401 – 1436 . Google Scholar CrossRef Search ADS Hall R. , Jones C. ( 1999 ) Why do some countries produce so much more output per worker than others? Quarterly Journal of Economics , 114 : 83 – 116 . Google Scholar CrossRef Search ADS Hayek F. A. ( 1949 ) Individualism and Economic Order . London : Routledge . Hungerford T. , Solon G. ( 1987 ) Sheepskin effects in the returns to education . Review of Economics and Statistics , 69 : 175 – 177 . Google Scholar CrossRef Search ADS Keynes J. ( 1936 ) The General Theory of Employment, Interest, and Money . New York , NY: Harcourt, Brace & Co . Ketterer T. , Rodríguez-Pose A. ( 2018 ) Institutions vs. “first-nature” geography: what drives economic growth in Europe’s regions? Papers in Regional Science, 95 : S25 – S62 . Knack S. , Keefer P. ( 1997 ) Does social capital have an economic payoff? A cross country investigation . Quarterly Journal of Economics , 112 : 1251 – 1288 . Google Scholar CrossRef Search ADS Laursen K. , Masciarelli F. , Prencipe A. ( 2012 ) Regions matter: how localized social capital affects external knowledge acquisition and innovation . Organization Science , 23 : 177 – 193 . Google Scholar CrossRef Search ADS Levine R. , Loayza N. , Beck T. ( 2000 ) Financial intermediation and growth: causality and causes . Journal of Monetary Economics , 46 : 31 – 77 . Google Scholar CrossRef Search ADS Mankiw G. , Romer D. , Weil D. ( 1992 ) A contribution to the empirics of economic growth . Quarterly Journal of Economics , 107 : 407 – 437 . Google Scholar CrossRef Search ADS Martins R. , Veiga J. F. ( 2013 ) Economic performance and turnout at national and local elections . Public Choice , 157 : 429 – 448 . Google Scholar CrossRef Search ADS Mauro L. , Pigliaru F. ( 2011 ) Social capital, institutions and growth: Further lessons from the Italian regional divide. CRENoS Working Papers, No. 3, Centre for North South Economic Research. Mortara A. ( 1985 ) Le Associazioni Italiane . Milano : Franco Angeli . Myrdal G. ( 1957 ) Economic Theory and Underdeveloped Regions . London : Duckworth . North D. ( 1991 ) Institutions . Journal of Economic Perspectives , 5 : 97 – 112 . Google Scholar CrossRef Search ADS Openshaw S. , Taylor P. J. ( 1979 ) A million or so correlation coefficients: three experiments on the modifiable area unit problem. In: Wrigley N. (ed) Statistical Applications in the Spatial Sciences , pp. 127 – 144 . London : Pion . Ottaviano G. , Peri G. ( 2006 ) The economic value of cultural diversity: evidence from US cities . Journal of Economic Geography , 6 : 9 – 44 . Google Scholar CrossRef Search ADS Papagapitos A. , Riley R. ( 2009 ) Social trust and human capital formation . Economics Letters , 102 : 158 – 160 . Google Scholar CrossRef Search ADS Platteau J. ( 1994 ) Behind the market stage where real societies exist—part II: the role of moral norms . Journal of Development Studies , 30 : 753 – 815 . Google Scholar CrossRef Search ADS Psacharopoulos G. ( 1985 ) Returns to education: a further international update and implications . Journal of Human Resources , 20 : 583 – 604 . Google Scholar CrossRef Search ADS Psacharopoulos G. ( 1994 ) Returns to investment in education: a global update . World Development , 22 : 1325 – 1343 . Google Scholar CrossRef Search ADS Psacharopoulos G. , Patrinos H. ( 2004 ) Returns to investment in education: a further update . Education Economics , 12 : 111 – 134 . Google Scholar CrossRef Search ADS Putnam R. ( 1993 ) Making Democracy Work: Civic Traditions in Modern Italy . Princeton : Princeton University Press . Putnam R. ( 2000 ) Bowling Alone: The Collapse and Revival of American Community . New York, NY : Simon & Schuster . Putnam R. , Feldstein L. ( 2003 ) Better Together: Restoring the American Community . New York, NY : Simon & Schuster . Resende G. M. ( 2011 ) Multiple dimensions of regional economic growth: the Brazilian case, 1991–2000 . Papers in Regional Science , 90 : 629 – 662 . Google Scholar CrossRef Search ADS Roodman D. ( 2009 ) A note on the theme of too many instruments . Oxford Bulletin of Economics and Statistics , 71 : 135 – 158 . Google Scholar CrossRef Search ADS Routledge B. , von Amsberg J. ( 2003 ) Social capital and growth . Journal of Monetary Economics , 50 : 167 – 193 . Google Scholar CrossRef Search ADS Rutten R. , Gelissen J. ( 2010 ) Social values and the economic development of regions . European Planning Studies , 18 : 921 – 939 . Google Scholar CrossRef Search ADS Rutten R. , Westlund H. , Boekema F. ( 2010 ) The spatial dimension of social capital . European Planning Studies , 18 : 863 – 871 . Google Scholar CrossRef Search ADS Sabatini F. ( 2008 ) Social capital and the quality of economic development . Kyklos , 61 : 466 – 499 . Google Scholar CrossRef Search ADS Sforza A. ( 2014 ) The weather effect: Estimating the effect of voter turnout on electoral outcomes in Italy. B.P. Working Papers, No. 5, Bank of Portugal. Solenne D. , Verrilli A. ( 2012 ) Compendio Di Diritto Costituzionale . Santarcangelo di Romagna : Maggioli . Solow R. ( 1956 ) A contribution to the theory of economic growth . Quarterly Journal of Economics , 70 : 65 – 94 . Google Scholar CrossRef Search ADS Solow R. ( 1995 ) But verify . The New Republic , 213 : 36 – 40 . Tabellini G. ( 2010 ) Culture and institutions: economic development in the regions of Europe . Journal of the European Economic Association , 8 : 677 – 716 . Google Scholar CrossRef Search ADS Temple J. ( 2001 ) Generalizations that aren’t? Evidence on education and growth . European Economic Review , 45 : 905 – 918 . Google Scholar CrossRef Search ADS Westlund H. , Adam F. ( 2010 ) Social capital and economic performance: a meta-analysis of 65 studies . European Planning Studies , 18 : 893 – 919 . Google Scholar CrossRef Search ADS Westlund H. , Larsson J. P. , Olsson A. R. ( 2014 ) Start-ups and local entrepreneurial social capital in the municipalities of Sweden . Regional Studies , 48 : 974 – 994 . Google Scholar CrossRef Search ADS Windmeijer F. ( 2005 ) A finite sample correction for the variance of linear efficient two-step GMM estimators . Journal of Econometrics , 126 : 25 – 51 . Google Scholar CrossRef Search ADS Zak P. , Knack S. ( 2001 ) Trust and growth . Economic Journal , 111 : 295 – 321 . Google Scholar CrossRef Search ADS Appendix A Our empirical strategy is based on the idea that social capital is a pre-condition for future growth. Although this hypothesis is supported by a number of theoretical articles (e.g., Cozzi, 1999; Routledge and von Amsberg (2003); Akçomak and ter Weel, 2009), from an empirical point of view it may be questionable whether the direction of causality goes from social capital to growth, and not vice-versa. To address this concern, we provide estimates of a simple panel VAR(1) between social capital and growth, with the following reduced form (the superscripts identify the corresponding equation): gi,t=via+vta+ρ1a gi,t−10+ρ2a  ln sci,t−10+vi,ta. (A.1) ln sci,t=vib+vtb+ρ1bln sci,t−10+ρ2b gi,t−10+vi,tb. (A.2) The evidence suggests that the coefficient ρ^2a is equal to 0.225 with the standard error of 0.075 (being statistically significant at 1% level), while the coefficient ρ^2b is equal to 0.002 with the standard error of 0.013 (being not statistically significant). On the one hand, these results confirm the chosen direction of causality. On the other hand, they provide indirect support to our proxy of social capital. Nevertheless, we are aware that our results cannot be generalised. For instance, Martins and Veiga (2013) find that economic conditions help to predict voter turnout at legislative elections in Portugal. This is an additional reason why we treat social capital as an endogenous explanatory variable in our main regression model, and we use lagged social capital to avoid simultaneity. Table A1 Linear correlation between voter turnout at political elections and alternative proxies of social capital at province level Alternative proxy Source Correlation with Coefficient Blood bags per 1,000,000 inhabitants, 1995 AVIS Turnout at political elections, 1992 0.566 Propensity to collective action, 1965–1974 ISL-UP Average turnout at political elections, 1963–1968 0.552 Volunteers associations per 100,000 inhabitants, 1999 ISTAT Turnout at political elections, 1992 0.493 Volunteers per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.476 Non-profit institutions per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.426 Associations per 100,000 inhabitants, 1982 ISL-UP Turnout at political elections, 1979 0.285 Share of employment in non-profit sector, 2000 ISTAT Turnout at political elections, 1992 0.176 Violent crimes per 100,000 inhabitants, 1998 ISTAT Turnout at political elections, 1992 −0.239 Alternative proxy Source Correlation with Coefficient Blood bags per 1,000,000 inhabitants, 1995 AVIS Turnout at political elections, 1992 0.566 Propensity to collective action, 1965–1974 ISL-UP Average turnout at political elections, 1963–1968 0.552 Volunteers associations per 100,000 inhabitants, 1999 ISTAT Turnout at political elections, 1992 0.493 Volunteers per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.476 Non-profit institutions per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.426 Associations per 100,000 inhabitants, 1982 ISL-UP Turnout at political elections, 1979 0.285 Share of employment in non-profit sector, 2000 ISTAT Turnout at political elections, 1992 0.176 Violent crimes per 100,000 inhabitants, 1998 ISTAT Turnout at political elections, 1992 −0.239 Notes: ISL-UP, Istituzioni e Sviluppo Locale—Università di Parma; ISTAT, Istituto Nazionale di Statistica; AVIS, Associazione Volontari Italiani del Sangue. The propensity to collective action is a principal-component indicator elaborated by Arrighetti et al. (2003) using information for different years between 1965 and 1974. The number of associations per 100,000 inhabitants in 1982 has been elaborated by Mortara (1985). Table A1 Linear correlation between voter turnout at political elections and alternative proxies of social capital at province level Alternative proxy Source Correlation with Coefficient Blood bags per 1,000,000 inhabitants, 1995 AVIS Turnout at political elections, 1992 0.566 Propensity to collective action, 1965–1974 ISL-UP Average turnout at political elections, 1963–1968 0.552 Volunteers associations per 100,000 inhabitants, 1999 ISTAT Turnout at political elections, 1992 0.493 Volunteers per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.476 Non-profit institutions per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.426 Associations per 100,000 inhabitants, 1982 ISL-UP Turnout at political elections, 1979 0.285 Share of employment in non-profit sector, 2000 ISTAT Turnout at political elections, 1992 0.176 Violent crimes per 100,000 inhabitants, 1998 ISTAT Turnout at political elections, 1992 −0.239 Alternative proxy Source Correlation with Coefficient Blood bags per 1,000,000 inhabitants, 1995 AVIS Turnout at political elections, 1992 0.566 Propensity to collective action, 1965–1974 ISL-UP Average turnout at political elections, 1963–1968 0.552 Volunteers associations per 100,000 inhabitants, 1999 ISTAT Turnout at political elections, 1992 0.493 Volunteers per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.476 Non-profit institutions per 100,000 inhabitants, 2000 ISTAT Turnout at political elections, 1992 0.426 Associations per 100,000 inhabitants, 1982 ISL-UP Turnout at political elections, 1979 0.285 Share of employment in non-profit sector, 2000 ISTAT Turnout at political elections, 1992 0.176 Violent crimes per 100,000 inhabitants, 1998 ISTAT Turnout at political elections, 1992 −0.239 Notes: ISL-UP, Istituzioni e Sviluppo Locale—Università di Parma; ISTAT, Istituto Nazionale di Statistica; AVIS, Associazione Volontari Italiani del Sangue. The propensity to collective action is a principal-component indicator elaborated by Arrighetti et al. (2003) using information for different years between 1965 and 1974. The number of associations per 100,000 inhabitants in 1982 has been elaborated by Mortara (1985). Table A2 Conditional mean estimates on growth: MAUP evidence (1) (2) (3) OLS OLS OLS Municipality Province Region Social capital 0.309*** 0.067 −0.237 (0.039) (0.152) (0.218) Initial income −0.395*** −0.560*** −0.471*** (0.008) (0.038) (0.0771) Time effects Yes Yes Yes Control set Yes Yes Yes Observations 10,145 515 100 R-squared 0.441 0.702 0.893 (1) (2) (3) OLS OLS OLS Municipality Province Region Social capital 0.309*** 0.067 −0.237 (0.039) (0.152) (0.218) Initial income −0.395*** −0.560*** −0.471*** (0.008) (0.038) (0.0771) Time effects Yes Yes Yes Control set Yes Yes Yes Observations 10,145 515 100 R-squared 0.441 0.702 0.893 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A2 Conditional mean estimates on growth: MAUP evidence (1) (2) (3) OLS OLS OLS Municipality Province Region Social capital 0.309*** 0.067 −0.237 (0.039) (0.152) (0.218) Initial income −0.395*** −0.560*** −0.471*** (0.008) (0.038) (0.0771) Time effects Yes Yes Yes Control set Yes Yes Yes Observations 10,145 515 100 R-squared 0.441 0.702 0.893 (1) (2) (3) OLS OLS OLS Municipality Province Region Social capital 0.309*** 0.067 −0.237 (0.039) (0.152) (0.218) Initial income −0.395*** −0.560*** −0.471*** (0.008) (0.038) (0.0771) Time effects Yes Yes Yes Control set Yes Yes Yes Observations 10,145 515 100 R-squared 0.441 0.702 0.893 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A3 Conditional mean estimates on growth with aggregation at (a) province level and (b) regional level (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS (a) Social capital 0.878*** 0.854*** 0.067 −0.037 −0.163 0.551* (0.107) (0.145) (0.152) (0.204) (0.221) (0.332) Initial income −0.269*** −0.293*** −0.560*** −0.890*** −0.884*** −1.013*** (0.012) (0.032) (0.038) (0.043) (0.043) (0.219) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 515 515 515 515 515 515 R-squared/G. of fit 0.545 0.592 0.702 0.796 0.797 0.589 (b) Social capital 0.938*** 0.240 −0.237 −0.126 −0.226 −0.104 (0.181) (0.248) (0.218) (0.253) (0.324) (0.785) Initial income −0.250*** −0.0580 −0.471*** −0.681*** −0.671*** −0.863 (0.0189) (0.0665) (0.0771) (0.0781) (0.0806) (0.619) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 100 100 100 100 100 100 R-squared/G. of fit 0.700 0.799 0.893 0.926 0.927 0.760 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS (a) Social capital 0.878*** 0.854*** 0.067 −0.037 −0.163 0.551* (0.107) (0.145) (0.152) (0.204) (0.221) (0.332) Initial income −0.269*** −0.293*** −0.560*** −0.890*** −0.884*** −1.013*** (0.012) (0.032) (0.038) (0.043) (0.043) (0.219) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 515 515 515 515 515 515 R-squared/G. of fit 0.545 0.592 0.702 0.796 0.797 0.589 (b) Social capital 0.938*** 0.240 −0.237 −0.126 −0.226 −0.104 (0.181) (0.248) (0.218) (0.253) (0.324) (0.785) Initial income −0.250*** −0.0580 −0.471*** −0.681*** −0.671*** −0.863 (0.0189) (0.0665) (0.0771) (0.0781) (0.0806) (0.619) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 100 100 100 100 100 100 R-squared/G. of fit 0.700 0.799 0.893 0.926 0.927 0.760 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A3 Conditional mean estimates on growth with aggregation at (a) province level and (b) regional level (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS (a) Social capital 0.878*** 0.854*** 0.067 −0.037 −0.163 0.551* (0.107) (0.145) (0.152) (0.204) (0.221) (0.332) Initial income −0.269*** −0.293*** −0.560*** −0.890*** −0.884*** −1.013*** (0.012) (0.032) (0.038) (0.043) (0.043) (0.219) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 515 515 515 515 515 515 R-squared/G. of fit 0.545 0.592 0.702 0.796 0.797 0.589 (b) Social capital 0.938*** 0.240 −0.237 −0.126 −0.226 −0.104 (0.181) (0.248) (0.218) (0.253) (0.324) (0.785) Initial income −0.250*** −0.0580 −0.471*** −0.681*** −0.671*** −0.863 (0.0189) (0.0665) (0.0771) (0.0781) (0.0806) (0.619) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 100 100 100 100 100 100 R-squared/G. of fit 0.700 0.799 0.893 0.926 0.927 0.760 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS (a) Social capital 0.878*** 0.854*** 0.067 −0.037 −0.163 0.551* (0.107) (0.145) (0.152) (0.204) (0.221) (0.332) Initial income −0.269*** −0.293*** −0.560*** −0.890*** −0.884*** −1.013*** (0.012) (0.032) (0.038) (0.043) (0.043) (0.219) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 515 515 515 515 515 515 R-squared/G. of fit 0.545 0.592 0.702 0.796 0.797 0.589 (b) Social capital 0.938*** 0.240 −0.237 −0.126 −0.226 −0.104 (0.181) (0.248) (0.218) (0.253) (0.324) (0.785) Initial income −0.250*** −0.0580 −0.471*** −0.681*** −0.671*** −0.863 (0.0189) (0.0665) (0.0771) (0.0781) (0.0806) (0.619) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 100 100 100 100 100 100 R-squared/G. of fit 0.700 0.799 0.893 0.926 0.927 0.760 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A4 Correlation matrix for outcome measures (growth rates) Real GDP per capita Real GDP Population Plant density Employees density Employees per capita Real GDP per capita 1.0000 Real GDP 0.8778 1.0000 Population −0.0124 0.3199 1.0000 Plant density 0.3622 0.5963 0.4210 1.0000 Employees density 0.6323 0.7750 0.3187 0.6638 1.0000 Employees per capita 0.6710 0.7139 0.0120 0.5637 0.9516 1.0000 Real GDP per capita Real GDP Population Plant density Employees density Employees per capita Real GDP per capita 1.0000 Real GDP 0.8778 1.0000 Population −0.0124 0.3199 1.0000 Plant density 0.3622 0.5963 0.4210 1.0000 Employees density 0.6323 0.7750 0.3187 0.6638 1.0000 Employees per capita 0.6710 0.7139 0.0120 0.5637 0.9516 1.0000 Table A4 Correlation matrix for outcome measures (growth rates) Real GDP per capita Real GDP Population Plant density Employees density Employees per capita Real GDP per capita 1.0000 Real GDP 0.8778 1.0000 Population −0.0124 0.3199 1.0000 Plant density 0.3622 0.5963 0.4210 1.0000 Employees density 0.6323 0.7750 0.3187 0.6638 1.0000 Employees per capita 0.6710 0.7139 0.0120 0.5637 0.9516 1.0000 Real GDP per capita Real GDP Population Plant density Employees density Employees per capita Real GDP per capita 1.0000 Real GDP 0.8778 1.0000 Population −0.0124 0.3199 1.0000 Plant density 0.3622 0.5963 0.4210 1.0000 Employees density 0.6323 0.7750 0.3187 0.6638 1.0000 Employees per capita 0.6710 0.7139 0.0120 0.5637 0.9516 1.0000 Table A5 Conditional mean estimates on growth with the full sample (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.850*** 0.892*** 0.283*** 0.239*** 0.103*** 0.444*** (0.020) (0.021) (0.0251) (0.038) (0.038) (0.201) Initial income −0.380*** −0.408*** −0.498*** −0.962*** −0.967*** −0.792*** (0.003) (0.004) (0.004) (0.005) (0.005) (0.148) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 40,185 40,185 40,185 40,185 40,185 40,185 R-squared/G. of fit 0.336 0.350 0.408 0.688 0.691 0.398 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.850*** 0.892*** 0.283*** 0.239*** 0.103*** 0.444*** (0.020) (0.021) (0.0251) (0.038) (0.038) (0.201) Initial income −0.380*** −0.408*** −0.498*** −0.962*** −0.967*** −0.792*** (0.003) (0.004) (0.004) (0.005) (0.005) (0.148) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 40,185 40,185 40,185 40,185 40,185 40,185 R-squared/G. of fit 0.336 0.350 0.408 0.688 0.691 0.398 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. Table A5 Conditional mean estimates on growth with the full sample (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.850*** 0.892*** 0.283*** 0.239*** 0.103*** 0.444*** (0.020) (0.021) (0.0251) (0.038) (0.038) (0.201) Initial income −0.380*** −0.408*** −0.498*** −0.962*** −0.967*** −0.792*** (0.003) (0.004) (0.004) (0.005) (0.005) (0.148) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 40,185 40,185 40,185 40,185 40,185 40,185 R-squared/G. of fit 0.336 0.350 0.408 0.688 0.691 0.398 (1) (2) (3) (4) (5) (6) OLS OLS OLS FE FE GMM-SYS Social capital 0.850*** 0.892*** 0.283*** 0.239*** 0.103*** 0.444*** (0.020) (0.021) (0.0251) (0.038) (0.038) (0.201) Initial income −0.380*** −0.408*** −0.498*** −0.962*** −0.967*** −0.792*** (0.003) (0.004) (0.004) (0.005) (0.005) (0.148) Time effects No Yes Yes Yes Yes Yes Fixed effects No No No Yes Yes Yes Control set No No Yes No Yes Yes Instruments No No No No No Yes Observations 40,185 40,185 40,185 40,185 40,185 40,185 R-squared/G. of fit 0.336 0.350 0.408 0.688 0.691 0.398 Notes: *,**,***Statistical significance at the 10%, 5% and 1% level. Standard errors in brackets. © The Author(s) (2018). Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Journal of Economic GeographyOxford University Press

Published: Jun 2, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off