Methodology of regression analysis

We contribute to prior research by examining a total of 32 European countries beyond the typical set of Western European countries. Specifically, we introduce countries of Central East Europe and the former Soviet Union, some of which have recently become members of the EU or stand as candidate countries for future accession. We are able to enrich our sample, because of social capital data made available by the European Values Study (2003) (EVS, hereafter), which is part of the World Values Study employed by most empirical work. We formally test our hypothesis by estimating cross-country regressions, where human development is regressed on social capital, while simultaneously controlling for a series of economic, social, political, and regional development factors. The econometric specification employed draws from the work of Ranis et al. (Reference Ranis, Stewart and Ramirez2000), and particularly from their emphasis on factors such as civil society, NGOs, and social capital, which are considered a major source of human development enhancement (pp. 198–201).Footnote ¹ Our empirical analysis differs in that we focus on a set of developed rather than developing countries. Also, instead of life expectancy and adult literacy rates, we use the UN's composite Human Development Indicator (HDI, hereafter) (UNDP, 2004) to measure human development and examine its association to social capital. Finally, we expand our set of controls to include development factors that capture the quality of governance, social exclusion, market operations, and political stability, in addition to public provisions, economic growth, income inequality, and regional dummies used by Ranis et al. (Reference Ranis, Stewart and Ramirez2000). For a detailed description of data and sources, see Table A of the Appendix.

The HDI is the average of three measurable dimensions of human development corresponding to health, education, and income (UNDP, 2004: 127–128, 259). From Table 1 we observe that countries of Scandinavia and Northern Europe achieve the highest scores of HDI – around 0.9. Countries with the lowest scores include Turkey and countries of Central East and Eastern Europe. We must make note of the rather small variation of HDI, due to the fact that the vast majority of countries in our sample belongs to the high category of HDI rankings, which implies small differences in the underlying indicators (UNDP, 2004: 138). However, this can also be viewed as a benefit for our analysis, in the sense that it can provide a more stringent test of our hypothesis on the role of social capital.

Table 1.

Human development and social capital across European countries

Note. Human development is measured by the UN's Human Development Indicator, averaged for the period 1975–2002 (UNDP, 2004). Group membership, social trust, and civic norms are derived from the 1999/2000 wave of the EVS (2003).

Social capital in each country is captured by three standard indicators: group membership, as the percentage of respondents belonging to at least one social organization (Beugelsdijk and Schaik, Reference Beugelsdijk and van Schaik2005); social trust, as the percentage of respondents claiming to trust most people (Knack and Keefer, Reference Knack and Keefer1997; Zak and Knack, Reference Zak and Knack2001); and civic norms, as the extent to which respondents justify, on average, a series of unsocial acts (Oorschot and Arts, Reference Oorschot and Arts2005). From Table 1 it is evident that group membership in Scandinavian countries is above 80%, whereas in the Mediterranean South, with the exception of Greece, group membership is below the average (47%), which is also true for a number of Central East and Eastern European countries. The distribution is similar for generalized trust, which is positively correlated with group membership (r = 0.76). However, the civic norms index appears to be uncorrelated with group membership (r = 0.001) and social trust (r = –0.04). It is perhaps the case that it captures different cultural or institutional elements of trust (Clercq and Dakhli, Reference Clercq and Dakhli2003); nevertheless, we keep this variable for reasons of comparison with prior research.

Regression results

Regression results are presented in Table 2. For the purposes of our cross-sectional analysis, variables for each country are averaged over the period for which data are available from 1960 to 2002, with the exception of our social capital variables for which data were available for the survey year 1999/2000. Due to limited sample size and a wide array of explanatory variables, we estimate a series of specifications in which each group of economic, social, and political variables enters individually (equations (1), (3), (5)). Variables representing public expenditure are included in all equations, because they have statistically significant coefficients across different specifications. We then introduce social capital variables to test our main hypothesis with regard to their contribution in human development (equations (2), (4), (6)). Finally, we evaluate the combined performance of economic, social, and political variables (equation (7)), as well as the influence of alternative social capital variables that distinguish between different types of groups (equation (8)). Equations are estimated with OLS procedures, except for equation (9), where we employ 2SLS methods after testing for endogeneity. Further tests to ensure the validity of the assumptions of the classical normal linear model are summarized in the final rows of Table 2.

Table 2.

Cross-country regressions of human development and social capital (N = 32)

Note: Standard errors are in parentheses. *** 1% level, **5% level, and *10% level of statistical significance. Probabilities are in brackets.

First of all, we observe that public provisions in health and education play an important role in the improvement of human development.Footnote ² In all equations depicted in Table 2, coefficients for public expenditure on health and education are positive and statistically significant. Their interaction term also has a statistically significant coefficient with a negative sign. This perhaps reveals the substitutability between these two types of public expenditure in satisfying development objectives. What it implies is that there is a certain threshold beyond which a rise in public expenditure in one area would not contribute to human development, given the level of expenditure in the other area. For instance, coefficients in equations (1) and (2) of Table 2 imply that a 1 percentage point increase in public expenditure on education as a share of GDP would produce a rise in HDI only in countries with levels of public expenditure on health as a share of GDP that are roughly below 5% (= 0.055/0.011), ceteris paribus. This would include all countries but the Netherlands, Estonia, Denmark, and Sweden. The variable for subsidies and other transfers performs less well, because its coefficient, albeit positive, is smaller and statistically insignificant. It might have been useful to distinguish between particular types of subsidies and transfers, but data were not available.

Another variable that appears to be equally important in the development process is the quality of governance, introduced in equations (5) and (6), as part of a set of political indicators employed in our analysis, including political stability and corruption. We observe that amongst these variables governance has a statistically significant coefficient with the expected positive sign. In particular, an increase in the scale of governance by one unit, which corresponds to an improvement in the accountability and effectiveness of government authorities, will increase the value of HDI by 0.03, ceteris paribus. However, the variable capturing conditions of political stability has a coefficient that is statistically insignificant. It is likely that our variable is not capturing events of political unrest and conflict that took place in the post-war period, such as the dictatorships in South Europe and the collapse of the communist regime in East Europe, because it covers the period from 1996 onwards. Finally, with regard to corruption, the coefficient becomes statistically significant when we introduce variables of social capital in equation (6). However, since higher values of the corruption index indicate less corruption, the negative sign of the coefficient implies that less corruption will contribute less to human development. An explanation could be that, in less developed countries, where the quantity and quality of public and private provisions are more limited, citizens may have to rely on unofficial and often corrupt dealings with the state in order to secure a substantive level of income and welfare.

Moreover, it is worth noting that the positive and rather strong influence of public expenditure and governance variables could be indicative of the contribution of synergistic relations between public and private sectors, when combined with the positive and statistically significant effect of group membership in equations (6)–(9). This would imply that human development depends on strong social participation in third sector organizations coupled with strong state institutions that respond to social demands by ensuring government effectiveness and accountability in welfare expenditure and public goods provision.

Economic growth is introduced in equations (1) and (2), along with economic variables of international trade and financial development. Indeed, we observe that GDP per capita growth has a positive and statistically significant coefficient such that a rise in economic growth by 1 percentage point would increase the value of HDI by 0.007 units, ceteris paribus. The other variables become statistically significant in equation (2), where social capital variables are introduced. The coefficient for financial development, measured as the share of money supply M1 in GDP,Footnote ³ is positive, which is consistent with arguments that the creation of financial institutions for ensuring credit could improve the development prospects of a region. International trade, measured as exports plus imports over GDP, has a negative coefficient, which is counter to arguments emphasizing the benefits from the international realloction of resources towards more productive and efficient uses. Outcomes here might be influenced by the less developed countries of Eastern Europe that exhibit a relatively high degree of openness, similar to more developed countries such as Belgium and the Netherlands. It is the case that either the benefits of international trade have not yet been realized, or opening markets at such early stages of transition, without the appropriate domestic institutions for policy, finance, and research, has had adverse effects on development in these countries. Finally, the coefficient for economic growth loses statistical significance when the variable for governance is introduced in the equation, which is also true for variables of money supply and technological innovation (research and development expenditure) (results not presented here).Footnote ⁴ This perhaps reveals that economic growth, financial conditions, and technology do have an impact on human development; but this depends on political institutions to ensure the effective allocation of output and resources. Therefore, we omit these variables from our final specification in equation (7).

Variables of social exclusion – namely inequality, poverty, and unemployment – are introduced in equations (3) and (4). In both specifications, with and without social capital, coefficients are statistically insignificant, and only that of income inequality has the expected – negative – sign. One explanation for such findings is lack of long-run data to capture the actual magnitude of such problems, since, for inequality and poverty, data are available only for the period 1994–2002. Alternatively, one might think that aggregate measures are usually unable to capture often dispersed, localized, or simply undetected instances of low income and high poverty in more developed countries. However, as Ranis et al. observe in their study of developing countries, income distribution variables run counter to their expectations; that is, a more equal distribution does not seem to advance human development (Reference Ranis, Stewart and Ramirez2000: 204–205). Also, it can be argued that the consequences of social exclusion cease to have overall adverse effects on development, because we have controlled for the contribution of the welfare system, based on the provision of public goods by the state and the third sector. Indeed, by dropping all other regressors, the coefficient for income inequality becomes statistically significant at the 1% level; it loses statistical significance once group membership and governance are introduced in the regression (results not presented here).

We investigate the effect of further regional characteristics of European countries by introducing a set of regional dummies, which represent geographical groupings with similar historical, cultural, and institutional features: Scandinavia, Anglo-Saxon countries, Continental Europe, the Mediterranean South, Central East European countries, former Soviet countries, and the Balkans. We discover that only the coefficient for the regional dummy of Central East European countries displays relatively stronger statistical significance across alternative specifications (results not presented here). This dummy variable appears in all equations of Table 2 to account for regional-specific effects, which become statistically significant at the 10% level in equation (7). However, the coefficient has a negative sign and thus implies that factors specific to this region have had an adverse effect on human development in these countries, perhaps as a result of times of tribulation during the communist era.

We now come to test our hypothesis on the contribution of social capital to human development. From Table 2 we observe that group membership has a coefficient that is positive and statistically significant across specifications. This is not in general agreement with findings in the growth literature, where the weak relationship between group membership and growth performance is attributed to negative externalities produced by special-interest groups as a result of rent-seeking and lobbying. Again counter to empirical work on growth, social trust appears to have a statistically insignificant, albeit positive, coefficient. This outcome maintains even if we drop the group membership variable (result not presented here). One possible explanation could be that human development depends less on attitudes and sentiments of generalized trust and more on the capacity of members of social groups to organize, co-operate, and set goals to fulfill development objectives. Finally, with regard to civic norms, the coefficient becomes statistically significant in equation (6), when we introduce political variables of governance; but it is positive, that is, an increase in the index, which corresponds to more uncivic behaviour, appears to improve rather than inhibit development. Our rationale here is similar to that discussed for corruption: in countries that are less endowed with effective and accountable public institutions, improvements in development will depend on the ability of civilians to pursue their personal welfare individually by participating in clientelistic networks of state patronage. We therefore omit civic norms and corruption variables from our final specification in equation (7), because they are difficult to interpret. In sum, we find that participation in social organizations of the third sector of the economy has a direct impact on human development through the voluntary provision of public goods, and not only indirectly through the resolution of state and market inefficiencies. Indeed, in equation (7), which controls for a compact set of variables with a relatively stronger association to human development,Footnote ⁵ group membership is amongst the variables that appear to have a positive and statistically significant influence on human development: its coefficient implies that a 1 percentage point increase in membership would produce an increase in HDI of approximately 0.0008 units, ceteris paribus.Footnote ⁶

Of course, the simple act of membership in a variety of social organizations ranging from human rights associations to choir groups, and relating to more or less active or benign forms of participation, would not suffice to claim its beneficial impact on development. To consider the influence of different types of social organizations, we replace the group membership index with two distinct variables to distinguish between Olsonian groups (trade unions, political parties, and professional associations) and Putnamian groups (religious/church organizations, cultural groups, and youth work associations) (Knack and Keefer, Reference Knack and Keefer1997; Beugelsdijk and Schaik, Reference Beugelsdijk and van Schaik2005). In equation (8) of Table 2, we observe that, counter to expectations, Olsonian groups have a positive and statistically significant coefficient, whereas Putnamian groups have a negative, albeit statistically insignificant one. A possible explanation is that Olsonian groups include organizations, such as trade unions, that had a positive contribution to reform and development for regions such as Scandinavia, whereas Putnamian groups incorporate religious groups that for some countries pose obstacles to development. We examine qualitative aspects of social organizations in the next section.

Further tests presented in Table 2 might be indicative of the fact that there are qualitative aspects of social capital that are not fully captured by our group membership indices. Tests for spatial autocorrelation point to the existence of a spatial autoregressive process in the error term, which is attributed to data problems when the scale of observation (e.g. administrative units) does not match the spatial scale of the process under study and thus does not take into account spill-over effects between multiple units.Footnote ⁷ Nevertheless, when we re-estimate equation (7) with ML procedures to account for spatial dependence, we discover that coefficients for all regressors, including the autoregressive parameter for spatial errors, become statistically significant at the 1% level (results not presented here). Also, even though equation (7) appears to have one of the lowest information criterion values, equation (6), which includes variables on corruption and civic norms, has the lowest information criterion value amongst alternative specifications. Perhaps these variables capture aspects of bonding ties within groups without bridging ties across groups.

Finally, one might argue that in our final specification, the coefficient for group membership is capturing the effect of development on social capital, as improved conditions of health, education, and income could expand the ability for social involvement (Fafchamps, Reference Fafchamps2006). To solve for potential endogeneity of the group membership variable, we apply 2SLS methods. Since the instruments we use are relatively large in number and perhaps interrelated, we conduct a principal components analysis to obtain instruments that are uncorrelated, valid, and more efficient (Bai and Ng, Reference Bai and Ng2006) (Table B of the Appendix). Formal tests for instrument relevance and validity reveal that there are no severe inconsistencies in 2SLS.Footnote ⁸ In equation (9), we see that the coefficient for group membership falls, but remains positive and statistically significant. Alternatively, we can re-estimate the regression using values of HDI from the year 2002 as the dependent variable, and maintain values of group membership prior to that year. The coefficient for group membership is positive, statistically significant, and with a value equivalent to that obtained with 2SLS (results not presented here).

Overall, results draw our attention to the importance of social organizations in more developed countries, where they contribute directly to broader welfare aspects of development as part of the third sector of the economy in synergy with state institutions of welfare expenditure and governance. However, historical and cultural differences across regions lead to conceptual differences of what the third sector is and does in the economy, which cannot be captured by survey questions when they address queries concerning respondents’ membership in a specified list of voluntary associations. For instance, an approach borne out of the experience of Northern and Western regions of Europe tends to give more emphasis on the participation of individuals in a set of formal organizations, possibly underestimating the richness of informal social groups and networks present in South and East Europe. However, informal networks established to combat pressing social problems can, under certain conditions, lead to a highly localized, dispersed and fragmented third sector, which is not only incapable of scaling up activities to encompass more groups and influence policy, but is also susceptible to control from formal institutions. Furthermore, synergy of social organizations with state institutions can only be implicitly deduced from our empirical results if we accept that the existence of more encompassing organizations is associated with a high proportion of group membership and a high quantity and quality of public provisions by the state. Hence, further case study analyses across regions are necessary to obtain a more complete account of the conditions that determine participation in and effectiveness of social organizations.