Social Capital and Social Inequality Corporate Networks in the United States and Germany (1896-1938)

Paul Windolf, 1946, professor of Sociology at the University of Trier, Germany. Areas of research: network analyses, economic sociology, historical sociology. From 1987-92 professor of sociology at Heidelberg. Visiting research fellowships at the London School of Economics (1980/81), Stanford/department of sociology (1984), European University Institute, Florence (1986/87), Haas School of Business, Berkeley (1996/97), Center for European Studies, Harvard (1999), fellow of the Wissenschaftskolleg Berlin (2005/06). Publications: Expansion and Structural Change: Higher Education in Germany, the United States, and Japan 1870-1990, Westview Press 1997; Corporate Networks in Europe and the United States, Oxford University Press 2002; Corruption, Fraud, and Corporate Governance: A Report on Enron, in: Corporate Governance and Firm Organization, ed. A. Grandori, Oxford University Press 2004.

Paul Windolf Department of Sociology [email protected] Universit of Trier Phone: ++49-651-2012703 54286 Trier Fax: ++49-651-2013933 Germany

1. Social capital: an unspecific resource 2. Diffuse expectations 3. Corporate networks 4. Quantifying social capital 5. The unequal distribution of social capital 6. The regional distribution of social capital 7. Regression analyses: Who has the most social capital? 8. Stability over time: Multilevel regression analysis 9. Summary Appendix References

Abstract Social capital is an unspecific resource that can improve the market opportunities of individuals and the chances of survival for organizations. In this historical study it will be shown that social capital was divided up among big companies during the early twentieth century as unequally as were income and wealth in Western societies. The comparative analyses concentrate on the distributional structure of social capital available to large corporations in the United States and Germany. A great deal of social capital could be accumulated especially by firms that were located in the metropoles of New York or Berlin, had a relatively high number of bankers sitting on their boards of directors, and were among the largest existing enterprises. A multilevel regression shows that these causal relations remained relatively stable throughout the period from 1896 to 1938.

Words: 9.987

2 1. Social capital: an unspecific resource Bourdieu has proposed a definition for the term “capital” that goes beyond a strictly economic meaning and enables us to consider various resources that are important for the reproduction of social classes. Among such resources are economic, cultural, and social capital. He expands Marx’s use of the term ‘capital’ and thereby approa- ches Weber’s concept of ‘market opportunities’ (1964: 123). Bourdieu analyzes various forms of accumulated labor, not only wealth in form of the means of production, but also in the form of socialization or investments in social capital. “Capital is accumulated labor, either in a material form or an internalized, incorporated form” (Bourdieu 1983: 183). Both cultural and social capital are inheritable and influence the market opportunities of the following generation in the competition for social status.

The term ‘social capital’ has been applied in many areas of social science in the past decades and has thereby often lost its sharpness (Kadushin 2004). It is defined basi- cally in two different ways. On the one hand, social capital is considered to be a type of public good. This definition makes it nearly synonymous with the cultural and nor- mative integration of a group or with trust and reliability found in business relations.1 On the other hand, social capital is defined as an individual resource that creates competitive advantages. Burt (2001: 33) reduces this meaning to a concise statement: “Better connected people enjoy higher returns.”

The imprecise application of the term can be attributed primarily to the lack of a clear differentiation between what social capital actually “is” – meaning the substance of the term – and the instrumental character of social capital – meaning that which one can do with social capital. In order to clarify this, let us first return to the definition offered by Bourdieu: “Le capital social est l’ensemble des ressources actuelles ou potentielles qui sont liées à la possession d’un réseau durable de relations plus ou moins institutionnalisées d’interconnaissance et d’interreconnaissance” (Bourdieu 1980 : 2). Bourdieu defines social capital as the sum of the resources, actual or potential, that are linked with the possession of a durable network of more or less institutionalized relations based on mutual acquaintanceship (interconnaisssance) and recognition (interreconnaissance). The French word reconnaissance has three

1 On this point, see Coleman (1988), Putnam (1995), and Uzzi (1990). 3 meanings: recognition (of a person known), acknowledgment and legitimation in the sense of genuine belonging, and gratitude in the sense of mutual obligation and commitment. Thus, the main characteristics of social capital are based on mutual recognition and recall (interconnaissance), on institutionalization and stability (durable), and on legitimation and mutual commitment (interreconnaissance). Bourdieu does not name any specific purpose that can be achieved with social capital, but limits his definition to the characterization of a specific form of social relations (inter(re)con- naissance). The creation and continuation of these networks requires work on relations, which can be inherited as social capital and improve the market opportunities of successive generations.

In his work The Philosophy of Money, Simmel defines money as an “absolute means.”2 In doing so, he emphasizes the unspecific character that money has as a tool. With money, we can buy all goods and services that have taken on the form of commodities. Money is a general and abstract means, which itself is not based on an inherent value. In a similar sense, social capital is also an abstract and general instrument that can be applied for various purposes. It can strengthen social cohe- sion, create trust, be used to obtain information or to tap the resources of others for one’s own purposes. Often we can achieve something with social capital that cannot be purchased with money. Therefore, if we differentiate between the substance of social capital (réseau durable de relations) and the various functions that social capital can fulfill, it becomes clear that social capital is usually defined in the literature in terms of the aims pursued with this resource. However, it would be reasonable to differentiate between the instrument and the aims that can be achieved with it.

2. Diffuse expectations In traditional kinship systems, exchange was based on diffuse expectations with regard to the scope of the obligation and the time period in which the obligation needed to be reciprocated.3 In modern societies the market is coordinated through specific contractual relations. The transition from social relations based on diffuse expectations to modern exchange relations coordinated through specific contracts is a central element of the modernization process.

2 Simmel (1987: 219). Elsewhere he refers to money as “being solely a means and tool“ (p. 206) or “abstract means” (p. 209). 3 See the term “generalized exchange” in Lévi-Strauss (2002: Chapter XV). 4

Various studies have shown that networks are not coordinated through contracts, but rather through unspecific expectations of reciprocity.4 Therefore, they are considered premodern, and critics have argued that networks could only be productive – if at all – during the early phase of economic development. In developed economies, the market emancipates itself from networks, and market exchange is based exclusively on specific contracts (Stiglitz 2000: 64; Katz 1998: 3).

However, a more exact analysis shows that the market and networks exist parallel to one another. Networks do not replace rational market exchange but fulfill a comple- mentary function. They reduce transaction costs, particularly when uncertainty either prevents a contract from being concluded or compels the parties involved to exert considerable effort to do so (Williamson 1985: 50-51). Contracting parties have to resort to substitutes in order to fill the ensuing gaps, and one of these substitutes is the trust that develops within networks.

3. Corporate networks This study will analyze a specific form of social capital, namely, the “réseau durable de relations” that developed in the late nineteenth century between big companies and became one of the most important institutions of managerial capitalism (Mizruchi 1982). Like other forms of social capital, this specific form can be utilized to pursue a number of aims. Banks can use interlocking directorates to control companies to which they have given credit or have introduced on the stock market (Minz and Schwartz 1985). Such ties can help reduce resource dependency between firms (Pennings 1980). Companies can appoint directors of reputable banks to their supervisory boards in order to enhance their reputation (Fohlin 2006). Various studies have proven that companies that hold a key position in the network perform better than do isolated companies (DeLong 1991). Firms that are closely interlocked also create a political lobby group and support the same political parties (Mizruchi 1992). Useem (1984) argues that the dense interlocking between companies socially

4 Cf. "relational contracting" in Macneil (2000) and the symposium in honor of Ian Macneil in: Northwestern University Law Review 94 (2000). 5 integrates members of the economic elite and obliges them to champion the interests of capital above and beyond the immediate needs of their own companies.5

The cited studies prove that such aims were indeed pursued within the corporate networks. However, the statistic relationships often show only a low level of significance, which indicates that networks perform different functions and that a variety of diverging interests are pursued within them. The evidence shows that banks often try to use the board of directors to control their debtors; yet many bank interlocks had another function or were mobilized only on occasion for control purposes (Wellhöner 1989). This is also the case for further studies seeking to prove other very specific functions that the corporate network is thought to serve. Sometimes interlocking directorates are nothing more than dead ends - ties that do exist at some specific time but are not mobilized in the pursuit of interests by any of the participants.6

Therefore, it is constructive to view a corporate network as an opportunity structure. The “réseau durable de relations” spun by the interlocking between firms can be utilized by the members to pursue different interests. The social capital incorporated in such networks is an abstract and generalized means – a master key that opens many doors.

In economics, the term ‘asset specificity’ is used to determine the possible alternative uses for an investment (Williamson 1985: 54-55). The single-purpose machine tools that Ford installed in his factory prior to the First World War had a very high degree of asset specificity: They could only produce one certain part for the Model T. An industrial robot can fulfill a (nearly) unlimited number of tasks and therefore has a low degree of asset specificity. When we apply this term to money and social capital, we can say that money has the lowest asset specificity because money can buy (almost) anything. Social capital has a higher level of asset specificity because we cannot use it as a resource in all situations. Still, social capital is not linked to a specific function, but can be mobilized for a number of different interests and used in various social contexts.

5 Big linkers who hold many positions in intercorporate networks further the "generalized interest of the capitalist class" (Useem 1978: 225). 6 Undirected (secondary) interlocks that are often created accidentally within a network are frequently "dead ends". 6

If social capital – much like money – is a resource applicable to many purposes, then two questions arise: How can this resource be measured? How is this resource distributed among actors? In the subsequent sections, we will address the following arguments. 1. In the first half of the twentieth century, social capital is as unequally divided among big companies in the United States and Germany as income and wealth are among individuals or families within each society. Social capital has an extremely unequal distribution structure, which can be approximated by a power law distribution. 2. Social capital is unequally distributed among geographic regions. There are – especially in the United States – only a few metropoles where social capital is highly concentrated. 3. Social capital is concentrated in only a few economic sectors. The “power of the banks” is justified by the fact that a high degree of social capital is concentrated in the financial sector. 4. The unequal distribution among firms, regions, and economic sectors remains considerably stable during the years 1900 to 1938.

4. Quantifying social capital The economic capital at a person’s disposal can be quantified fairly precisely in an asset and liability statement. In a similar way, we can also measure a person’s educational capital by the number of school years completed, the academic degrees earned, and the prestige of the educational institutions attended. Likewise, we can also quantify social capital.7 The number of interlocks maintained by a firm and the prestige of the companies with which interlocks are kept are measurable indicators. In this way, we can determine the amount of social capital available in the network of big companies.

Table 1 offers an overview of several indicators that facilitate the quantification of social capital. In line 1, we find the number of firms included in the study for each year. All members of the executive management and the board of directors (BoD) of each firm were included in a databank. From this dataset, an adjacency matrix can

7 Cf. Fukuyama (1999, Sect. III); Borgattti et al. (1998). 7 be produced for each firm that shows the other firms with which it is linked. The sum of the interlocks in the adjacency matrix is defined here as the amount of social capital. The number of interlocks is listed on line 4 (italics). Line 5 contains the number of the directed (primary) interlocks.8 These two lines contain the basic figures that serve as the basis for calculating the different distribution structures in the following sections.

Table 1: Amount of social capital

Germany United States 1896 1914 1928 1933 1938 1900 1914 1928 1938 1 Number of firms 212 323 377 405 361 249 242 369 409 2 Connected firms 156 292 366 389 346 226 193 329 375 3 Isolated firms (%) 26.4 9.6 2.9 4.0 4.2 9.2 20.2 10.8 8.3 4 All interlocks 513 3081 12374 8177 6967 1579 1466 2538 2091 5 Directed interlocks 136 438 1416 1222 1156 468 395 812 715 6 Directed interlocks (%) 26.5 14.2 11.4 14.9 16.6 29.6 26.9 32.0 34.2 7 Interlocks per firm 2.4 9.5 32.8 20.2 19.3 6.3 6.1 6.9 5.1 8 Density (%) 1.61 4.23 10.8 6.66 7.0 3.2 3.34 2.49 1.64

In Germany, a total of 513 interlocks were maintained by 156 connected firms in 1896. This figure rose to 12,374 interlocks by 1928. In the United States, a total of 1,579 interlocks were available in the network in 1900, a figure that had risen to 2,538 by 1928. The amount of social capital was still larger in the United States than in Germany in 1900, but after 1914, German firms had more interlocks. These dif- ferences are also evident in line 7 (number of interlocks per firm) and line 8 (density). Density is a standardized unit of measurement and can be interpreted as a percentage. It indicates the share of possible interlocks in a network that have actually materialized.

It is shown in section 5 that the distribution structure of social capital can be approximated by a power law distribution. In section 6, a network effect is analyzed (the concentration of social capital in the metropoles of New York and Berlin). Section 7 answers the question about the characteristics of firms that accumulated a great deal

8 If an executive manager of firm A sits on the BoD of firm B as a non-executive/external director, this person creates a directed interlock. If a director sits on the BoD of firms A and B as a non-executive/external director, this person creates an undirected interlock between these two firms. 8 of social capital. Section 8 examines whether causal relations were stable in the years 1896 to 1938 (multilevel regression).

5. The unequal distribution of social capital In his work La courbe de la répartition de la richesse, Pareto turns his attention to the structure of income distribution in various countries. He shows that the income distribution can be estimated by a distribution function that later became known in a sim- plified form as the 20:80 rule: 20 percent of the population owns about 80 percent of the wealth.9 With the help of tax statistics, Pareto (1967:3) proves that the income distribution in England and Prussia follows a distribution “law,” presented in a mathe- matical formula explained below in the equations (1) and (2).

It can be shown that the distribution of social capital among large firms can be approximated by this Pareto distribution. The analytical units in this case are not individuals, but organizations. A large company could be linked to numerous other companies over its boards of directors. The larger the number of firms with which it is directly linked, the larger the amount of social capital at its disposal (= degree).

Figure 1a shows the frequency distribution of degree for all German firms in 1896 included in the sample survey (N = 212). Degree is represented on the x-axis; frequency on the y-axis. Isolated firms (degree = 0) and firms with a degree of 1 were combined. The figure shows that 100 firms had a degree of 0 or 1, meaning they had practically no social capital at their disposal in the network of big companies. Only ten firms had a degree of thirteen or higher. The empirical frequencies also show an extremely unequal distribution in the networked capital. There are many isolated or marginalized firms (100) and only a few companies with many interlocks. In figure 1b, the empirical distribution is estimated by using the Pareto distribution and takes on the following form: (1) p(x) = Cx-α | x: degree; p(x) = y = frequencies (2) ln(y) = c – αln(x)

The coefficient α is called the Pareto constant. The larger α becomes, the more unequal becomes the distribution of the variable being tested – in this case, the distribu-

9 A historical study on income inequality in the United States (1913-1998) is found in Piketty and Saez (2003). 9 tion of social capital. After logarithmic transformation (linearization), we get equation (2), the power law distribution. Various studies have shown that there are many variables in the social and natural sciences whose distribution can be estimated by this function. For example, the distribution of the number of inhabitants living in the cities of a country has such a distribution structure: there are a great many cities with a small population size, but only a few cities with a very large population size. Another example can be found in research, where a great many articles are written that are cited very rarely if ever (“isolated”), while a select few are constantly being cited within the profession.10 Figure 1c shows that the distribution of social capital among big German companies in 1896 approximates a power law distribution.

Figure 1a: Frequencies

100

60 Frequency 40

0 0 5 10 15 20 Germany Degree 1896

Ø: 3.7; SD: 3.9; N=212

Figure 1b: Pareto distribution

100,0 Observed Inverse

80,0

60,0

Frequency 40,0

20,0

0,0

0,0 5,0 10,0 15,0 20,0 Germany Degree 1896

Regression: y= 1/xα R²= 0.92

10 On power law distribution in networks, see Newman et al. (2002) and Newman (2005). 10

Figure 1c: Power law distribution

5,0 Observed Linear

4,0

3,0

2,0

1,0

0,0

0 0,5 1 1,5 2 2,5 3 Germany Ln(degree) 1896 Regression: Ln(y) = c - α ln(x) R² = 0.88 α = 1.46

Table 2: Unequal distribution of social capital United States Germany Year Pareto R² Gini Hoover Skew Pareto R² Gini Hoover Skew 1900 1.17 0.74 0.54 0.41 1.64 1.46 0.88 0.60 0.46 1.85 1914 1.03 0.79 0.61 0.46 1.61 0.94 0.82 0.57 0.44 1.75 1928 1.20 0.77 0.53 0.40 1.07 0.60 0.60 0.48 0.36 1.57 1938 1.45 0.87 0.51 0.38 1.59 0.71 0.65 0.47 0.34 1.87 The column "Pareto" gives unstandardized coefficients (α) for the following regression equation: ln(y) = c – αln(x); y: frequencies; x: degree. The coefficient of determination (R²) refers to this regression equation. The Hoover coefficient is equal to the portion of the total social capital that would have to be redistributed to attain perfect equality. Skew: skewness (third central moment of distribution). In- equality coefficients have been computed with ungrouped data (high sampling resolution).

Table 2 lists various coefficients that can be used to describe the degree of inequality in a distribution structure. For 1896, the Pareto constant in Germany equals 1.46; it falls to 0.71 by 1938. The development in the United States is just the reverse. The constant increases from 1.17 (1900) to 1.45 (1938). In Germany, the amount of social capital grows tremendously between 1896 and 1938, and the degree of inequality decreases. In the United States, there appears to be a contrary trend. The density decreases continuously, and the degree of inequality increases. However, this trend is not confirmed by the Gini or Hoover coefficients. Therefore, we assume that the degree of inequality in the United States did not change during this period.

Pareto (1967: 4) calculated an α for England and Prussia for the years from 1843 to 1886 that varied from 1.35 to 1.73. He assumed that the income distribution in mo-

11 dern mass societies could be described exactly with the help of these constants.11 The Pareto constants that we calculated for the distribution of social capital between big companies in Germany (1896) and in the United States (1938) lie in this realm. However, the following qualification needs to be considered: the R2 for the linearized Pareto function shows that the adjustment is insufficient for several years (e.g. Ger- many in 1928: R2 = 0.60). The Pareto constant needs to be interpreted with caution.12 The distribution of social capital in the network of big companies only approximately follows a power law distribution.

The question to be answered in this section is not why some firms possess relatively large amounts of social capital.13 That question will be addressed later in section 8 (regression analysis). What is to be explained in this section is the distribution structure of social capital. Why is this structure roughly as unequal as the distribution structure of income and wealth?

We assume that the unequal distribution of social capital can be explained with the help of three effects: a network effect, an inclusion/exclusion effect, and a path- dependency effect.

The network effect maintains that peripheral firms try to interlock with companies that already enjoy a central position in the network, meaning companies with access to a great deal of social capital. The efforts of the peripheral firms end up positively rein- forcing the position of the few key companies, because the latter companies are the ones who accumulate more social capital. In this way, the cycle repeats itself. Gran- ted, the peripheral firms are so able to establish interlocks with a major actor, but they themselves accumulate very little social capital. Instead they are subjected to the logic of “structural holes,” and the key company can act as a broker for information and resources (Burt 2001).

11 For criticism of Pareto's notion of "law", see Béraud (2005); Prévost (2002). 12 The regression equation Ln(y) = c – αLn(x) gives a biased estimate of Pareto’s constant (α). We used this estimate to compare our results directly with Pareto’s calculation of income inequality in the late nineteenth century. An unbiased estimate provides the maximum-likelihood estimation: α = 1+n(∑ln xi/xmin). See Newman (2005). We estimated the Pareto constant using this formula. We also computed the Atkinson and Theil coefficients (normalized entropy). The results are available from the author upon request. 13 In 1928, Guaranty Trust (NYC) had a degree of 51, General Electric of 45. In contrast, Ford Motor Co. had a degree of 1 and Eastman Kodak was “isolated” (degree = 0). 12

Networks set themselves off from their environment through processes of inclusion and exclusion (Luhmann 1995). By way of controlled inclusion and selection, they create identities for themselves based on reputation and prestige. Bourdieu describes this process as “interreconnaissance”: only actors legitimized by prestige and/or reputation have access to this network (e.g. investment banks, large success- ful companies).14

Social capital is based on past investments in a corporate network. A broken interlock incurs sunk costs that cannot be replaced. Thus, companies nurture the ties in which they have already invested heavily. In this way, a path dependency evolves within the network that stabilizes distribution structures. In the following section, we will analyze path dependency more closely (the key positions of the metropoles in the networks).

6. The regional distribution of social capital Until the mid-nineteenth century, the American market was still divided into nearly autonomous regions with little economic trade among them. Not until the expansion of the railroad network did the United States become an integrated market on which economies of scale could be used. Connected to this was the emergence of big companies, for whom the entire continent became a potential market (Chandler 1977).

In several studies it has been argued that the network links firms that produce in various regions and can use interlocking directorates to coordinate their performance and exchange information. In this way, the network overcomes problems of physical distance and can socially integrate a national economy (Kono et al. 1998; Burt 2006). In this section, we will examine whether big companies in the first half of the twentieth century were integrated into a comprehensive network that encompassed the national market. If so, it would mean that the social capital was divided up to an approximately equal degree among the firms of the various regions.

14 Reputation is based on past achievement; prestige is an ascribed characteristics. 13 It has been further argued that the network promotes the creation of an integrated national economic elite (Baltzell 1979). The boards of directors serve as meeting places for managers from various firms and regions. By meeting the same people from distant regions of the country several times a year to make joint decisions and perform supervisory functions, these regional elites can integrate themselves into a national economic elite. We will also test this argument.

For a start, we divided the United States and Germany into states and different regions, respectively. Firms were then grouped according to these divisions.15 We then calculated regional distribution matrices that show the percentage of social capital concentrated within each region and the scope to which the regions were linked to one another.

Table 3: Regional distribution of social capital (%) U.S. Intrastate NewYork State ↔ NY ∑NY CR3 All (N) (1) (2) (3) (4) (5) (6) 1900 41.3 32.8 22.8 55.6 73.91579 1914 37.4 26.3 24.2 50.5 72.81466 1928 45.2 28.3 20.4 48.7 68.72538 1938 46.8 25.3 19.7 45.0 63.92091

Germany Intraregion Berlin Region ↔ Berlin ∑Berlin CR3 All (N) 1896 43.5 13.2 16.7 29.8 58.5513 1914 34.6 16.9 22.6 39.5 70.43081 1928 33.5 16.2 22.4 38.7 75.212374 1933 34.6 16.4 19.9 36.3 71.48177 1938 36.2 13.5 18.8 32.3 71.66967 Note: CR3: Concentration ratio for 3 states/regions with highest proportion of social capital.All (N): Total number of interlocking directorates

Table 3 offers an overview of the regional distribution structure of social capital in the United States and Germany. Column 1 (intrastate/intraregion) shows the percentage of interlocks existing between firms located in the same state or region. We define this percentage as intraregional interlocking. Within Germany in 1896, 43.5 percent of all interlocks were concentrated within each respective region. Correspondingly, (100 – 43.5 =) 56.5 percent of all interlocks existed between firms located in some other region (interregional interlocking). The percentage of the interregional interlocking can be interpreted as an indicator of the degree of spatial integration for a national

15 US: a total of 25 individual states; Germany: a total of 10 regions. A firm was assigned to the state in which its executive office was located. In 1928, for example, 119 firms had their executive offices in New York state (see Table A1, appendix). 14 network. Striking is the fact that the percentage of intraregional interlocks drops in Germany between 1896 and 1938 from 43.5 percent to 36.2 percent but rises in the United States from 41.3 percent to 46.8 percent. In other words, the network in Germany tends to become more “national,” while in the United States it tends to become more “regional.” According to this trend at least, the American network is more concentrated on individual states in 1938 than was the case in 1900. This is all the more astounding since the transportation technology made enormous progress between 1900 and 1938 (railway, automobile, airplane). We could have expected the opposite trend.16

Column 2 shows the percentage of social capital that was concentrated in New York and Berlin. In the United States, New York was the center of corporate interlocking. In 1900, nearly a third of all interlocks (32.8 percent) linked firms whose executive offices were located in New York. In comparison, Berlin played a less dominant role. Over the course of time, the importance of New York waned. The percentage of intraregional interlocks concentrated in New York dropped from 32.8 percent to 25.3 percent.

Column 3 indicates the percentage of interregional interlocks linking New York or Berlin to other states or regions (state↔NY). These figures therefore show the power of attraction exerted by the network center on firms that seek to enhance their geo- graphically rather peripheral positions by linking themselves to firms situated in the geographical center of the network. New York’s power of attraction tends to decrease over the course of time from 22.8 percent to 19.7 percent. That of Berlin fluctuates around the mean of about 19-20 percent during the same period.

The figures in column 4 are the sums of those in columns 2 and 3, and represent the percentage of all interlocks that exist directly between either firms within the network center or the center and the geographic periphery. Although the importance of New York decreases, 45 percent of social capital is still concentrated in the city in 1938, either intraregionally or interregionally. The percentage of social capital concentrated

16 Even in 1976 it still held that US firms tended to be regionally interlocked: “… the American interlock network was divided into regional groupings within which corporations maintained denser connections to local concerns than to distant companies” (Bearden and Mintz 1985: 241). 15 in Berlin overall is less than this and drops noticeably between 1914 and 1938 from 39.5 percent to 32.3 percent.

Column 5 shows the concentration ratio for the three regions (CR3) in which the greatest percentage of social capital is concentrated. In the United States, these are the states of New York, Pennsylvania, and Illinois; in Germany, they are the regions of Berlin, Rhineland-Westphalia, and Saxony. Table 3 shows that at least two-thirds (and often more) of the total social capital was concentrated in three regions throughout nearly the entire period. These regions made up the strategic center of an economy. Although the economy was being increasingly organized “nationally” with regard to the geographical diversification of production and the integration of the market, its strategic headquarters for management and communication were concentrated only in a few regions.

The argument that social capital was highly concentrated on a regional basis is therefore confirmed by these results. In the United States, 71 percent of all social capital was concentrated around five cities in 1900 17 – namely, New York, Chicago, Phila- delphia, Boston, and Pittsburgh. These five cities held onto their leading positions throughout the period ending in 1938 (68 percent). While the importance of New York declined, the share of social capital found in Boston and Chicago grew. In Germany, 48 percent of all social capital was concentrated around five German cities in 1896. This percentage rose to 55 percent by 1928 and then fell again to 33 percent by 1938. Only Berlin and Hamburg were among the five cities with the most social capital for the entire period from 1896 to 1938. Overall, the geographical degree of concentration of social capital in Germany was less than in the United States.

In a study on the “Philadelphia Gentlemen,” Baltzell (1979) puts forth the thesis that a national economic elite arises in the United States in the early twentieth century “which cuts across local boundaries to include fashionable families in all the older urban centers from San Francisco to New York, Boston, Philadelphia, or Baltimore” (pp. 5, 24). Baltzell names four institutions that were particularly important for the integration of an “inter-city aristocracy.” The first three are “fashionable boarding

17 This percentage is compiled as follows: percentage of interlocks between firms located in the same city (e.g. Chicago) plus the percentage of interlocks existing interregionally with firms in this city (e.g. between Chicago and San Francisco). 16 schools, universities and clubs” (p. 13). Also important for the social integration of the upper class was the “social register,” in which the upper class members could be registered. This register thus became a document of exclusivity (pp. 19-24).

The analysis of the geographical distribution of social capital presented here questions the idea that the economic elite in the United States actually united to form a national elite. It was shown that the percentage of ties between firms within the same city rose in the period from 1900 to 1938 (intraregional interlocks). Thus the network did not become more national, but more regional. Furthermore, the results indicated that New York continued to play the dominant role. Last of all, it was shown that more than two-thirds of social capital was concentrated in five cities. Three of these are located on the East Coast and one, Pittsburgh, is in the eastern United States; only Chicago can be said to be from the Midwest. Neither the major economic centers of the western United States (San Francisco, Los Angeles) nor those serving the South (St. Louis, New Orleans) played a major role in this network. By the late nineteenth century, the American economy had developed into a nationally integrated production system. However, this production system was directed by an economic elite located predominantly on the East Coast.

These findings are supported by a study from Sweezy (1962: 168) on Interest Groups in the American Economy. Sweezy identifies eight “interest groups” that were interlocked with one another and exerted a dominant influence on the 200 largest industrial enterprises in 1938. In five of these interest groups, the center of the network consisted of families or certain individual firms. These were J.P. Morgan and the First National Bank (New York), Rockefeller and the Chase National Bank (New York), Kuhn & Loeb (New York), Mellon from Mellon National Bank (Pittsburgh), and DuPont (Wilmington, Delaware). Three interest groups were concentrated in cities, namely, Chicago, Boston, and Cleveland.

In conclusion we can say that more than two-thirds of the social capital in both Germany and the United States was concentrated in three regions and that one of these regions in each country – namely, Berlin and New York – assumed the role of being the national center of the network. These findings do not really prove the existence of a national network that became integrated spatially and socially by way

17 of strong interregional interlocking. However they are proof of a network effect.18 The firms attempted to integrate themselves into networks that already had many members. Major metropoles can attribute their hegemonial position not only to above- average provisions of their material and cultural infrastructures, but also to the fact that they are the network centers and monopolize a high percentage of all social capital.19

Networks are the result of past investments. They require continual care in order to prevent them from turning into dead ends. If a firm leaves the metropole, it loses its investments (sunk costs). This indicates that its dominant position can be explained not only by the network effect, but also by path dependency: firms remain in the metropole in order to avoid taking a loss on past investments in social capital. With regard to the unequal distribution of social capital among firms, it can be argued that physical location is an important explanation for the centrality of a company in the network. Firms whose headquarters are located in a metropole have access to more contacts than firms located on the periphery.

7. Regression analyses: Who has the most social capital? We have seen so far that social capital is distributed very unequally among firms. In this section, several variables will be identified that can explain this unequal distribution. Which characteristics do firms have that possess a relatively large amount of social capital, and what distinguishes them from companies that are isolated or marginalized in the network?

We measure social capital with the aid of actor degree centrality (Wasserman and Faust 1994: 178). This is defined as the number of companies with which a firm is linked through the BoD. The degree is the dependent variable in the regression analysis.

The regional distribution of social capital that we analyzed in the previous section is an important independent variable. It is quantified by the number of companies in a

18 As put by Katz and Shapiro, “…the utility that a given user derives from the good depends upon the number of other users who are in the same ‘network’ as is he or she” (Katz and Shapiro 1985: 424). 19 Cf. the concept of a “new geography of centrality” in Sassen (2001). 18 region. We assume that regions in which a relatively large number of big companies have settled (e.g. New York and Berlin) offer more opportunities to interlock than remote regions in which only a few big companies are found. Thus, the network effect is measured by the number of companies in a region (see table A1 in the appendix).

In many studies on interlocking directorates, banks were identified as belonging to a major economic sector that features a particularly large number of interlocks. There- fore, we have added the financial sector and a few other economic sectors as dummy variables in the regression. In this way, the relative importance of the financial sector can be determined in a multivariate analysis.

Another variable used to help quantify the influence of the financial sector is the number of bank managers who sit on the boards of directors of other firms. The argument is that companies with many bankers on their board possess a relatively large amount of social capital. Banks hold a major position in the network and often act as “brokers” between other companies. However, the direction of the causality does pose a problem: do companies possess a high degree because many bankers sit on their boards? Or do bankers try to be selected to sit on the boards of companies that are already important to the network, companies that already have a high degree? We will try to determine the direction of causality by using additional information.

Company size is also incorporated into the regression as another explanatory variable. Size is quantified by using the log of equity capital. Equity capital is the only size criterion for which we have information for a sufficiently large number of companies over the entire period of this study. Furthermore, we have compiled the founding year for each company (age). The argument is that the development of networks requires time; hence, older companies should therefore have a greater degree.

The regression analyses were calculated separately for each of the years 1896/1900, 1914, 1928, and 1938. These analyses show which variables have determined the centrality of a company in the network for each of these various years.

19 Table 4 presents in detail the findings of these analyses that we summarize briefly here. In every year, the most important explanatory variable is the number of bankers sitting on the boards of directors. Companies with many bankers on their BoD have a high degree and therefore possess a relatively large amount of social capital. For example, Western Union Telegraph and Northern Pacific Railroad each had six bankers on their boards in 1914. In Germany, AEG had seven bankers on its supervisory board. The unstandardized regression coefficient for the variable “number of bankers on BoD” is 11.1 for Germany in 1928. This means that a company (with average characteristics) increases its degree by 11.1 for each additional banker it selects to sit on its supervisory board. For the United States the figure for the unstandardized coefficient is 4.1.

Table 4 : OLS-Regressions Germany – United States Independent Germany United States variables 1896 1914 1928 1938 1900 1914 1928 1938 Banker 0.453 0.473 0.583 0.473 0.673 0.613 0.543 0.493 Assets (ln) 0.263 0.413 0.313 0.363 0.132 0.223 0.263 0.243 Region 0.172 0.132 0.123 0.05 0.122 0.143 0.06 0.163 Bank 0.172 -0.03 0.0 -0.01 0.132 0.223 0.263 0.213 Steel/RR 0.01 0.06 0.102 0.132 0.152 0.173 0.112 0.07 Intercept 0.13 -7.03 -17.43 -11.63 3.03 1.91 4.03 1.71 R² 0.49 0.59 0.56 0.44 0.69 0.73 0.50 0.49 Firms (N) 190 305 374 359 235 226 344 370

Dependet Variable: actor degree centrality (all interlocks). Figures in table are standardized regression coefficients (β). 3: α ≤ 0.000; 2: α ≤ 0.01; 1: α ≤ 0.05. Germany: steel; U.S.: Railroads (RR)

The second most important variable is the size of the company (assets). The larger the company, the more numerous its contacts to other companies. In third place is the variable ‘region.’ Social capital is accumulated especially in the big metropoles, which become the centers of the contact networks. Firms move their headquarters to these cores in order to integrate themselves into the contact networks.20

20 We also computed interaction effects. For instance, the interaction effect “number of bankers” x “region” is highly significant, meaning that many banks are located in New York/ Berlin. However, when this interaction is entered into the regression, the main effects of the variables “banker” and “region” are considerably reduced, while the explanatory power of the model (R²) is not improved. Therefore, Table 4 only shows the main effects. Interaction effects are available upon request from the author. 20 In the United States, the variable ‘banks’ and ‘railroads’ (RR) have a significant influence on the degree. Companies belonging to these economic sectors possess ap- preciably more social capital than other companies (ceteris paribus). In Germany, the influence of the economic sectors varies. The variable ‘banks’ is only significant in 1896; after that the banks no longer have a higher degree, compared to nonfinancial companies with comparable characteristics. In 1928 and 1938, the iron and steel industry had a significant influence on degree. The unstandardized regression coefficient for the iron and steel industry is 12.9 for Germany in 1928. This means that a company from the iron and steel industry (with average characteristics) had nearly thirteen more contacts to other companies than did companies from other sectors. The coefficients for the United States are 7.2 for banks and 3.3 for railroads in 1928.

The variable ‘company age’ is not significant in any year. Therefore, it was not included in table 4. In the early twentieth century, the age of a company did not play a role with regard to the accumulation of social capital.21

It is surprising to find that the variable ‘banker’ maintains the greatest explanatory value for the entire period. However, as was pointed out above, this information does not clarify the direction of the causality. The first assumption can be that the variable only describes a selection effect: bankers seek out companies that are located at the center of the network and therefore already have a high degree. In other words, bankers are merely jumping on the bandwagon. The alternative explanation is that bankers actually “cause” the high degree of companies. The following example is meant to illustrate that bankers were an important cause for the accumulation of social capital for companies and that the results of the regression analysis are not the statistical artifact of a selection process.

For each year in the study, we compiled a list of big linkers, meaning those managers and directors who had the most mandates in the network. In 1900, the following four bankers topped the list for the United States:

Charles Steel, partner at J. P. Morgan, 10 positions James Stillman, CEO of National City Bank, New York, 9 positions

21 However, in the 1990s, a company’s age did have significant influence on its centrality in the network (Windolf 2002: 45-6). 21 Robert Bacon, partner at J. P. Morgan, 8 positions George F. Baker, CEO of First National Bank, New York, 8 positions

These four bankers all sat together on the BoD of Northern Pacific Railroad. All told, they held thirty-five positions in the network and thereby created (35 – 4 =) thirty-one interlocks to other companies for Northern Pacific Railroad, which was linked to a total of forty other companies (degree). A simple calculation led to the finding that these four bankers created 55 percent of all ties for Northern Pacific Railroad.22 Figure 2 shows the ego network of Northern Pacific Railroad. The black dots represent those companies in which at least one of the four bankers sat in the BoD. There is another interesting detail in this context: of the forty companies with whom Nor- thern Pacific was linked, twenty-six (65 percent) had their executive offices in New York. The executive office of Northern Pacific sat in St. Paul, Minnesota.

Figure 2: Ego-Network of Northern Pacific RR (1900)

22 When calculating these figures, it was taken into account that Steele and Bacon not only sat together on the board of Northern Pacific, but also at American Bridge and the Erie Rail- road. This double interlocks were subtracted (dichotomization). For Germany, a comparable calculation shows that three bankers created 51.6 percent of all ties for the Gelsenkirchener Bergwerks AG (degree: 155 in 1928). 22 Note: NP-RR: Northern Pacific RR in the center of the network. Black dots: 22 firms in which at least one of the four bankers sat on the BoD. RR: railroad; B: Bank; In: Insurance company. JPM: J.P. Morgan & Co. The Northern Pacific RR is connected to 40 other firms (degree). Clustering coefficient: 0.290 (= density of the dichotomized network).

It thus becomes clear that bankers dominated the network and often served as go- betweens for industrial firms. Northern Pacific Railroad accumulated social capital by appointing social-capital accumulators to its BoD. Therefore, the significant relationship between the number of bankers sitting on the BoD and the degree of a company is not a spurious correlation. It can be explained by the fact that a small number of bankers who hold many positions meet repeatedly and regularly in a small number of firms and thus increase the number of contacts among these companies. Big linkers who hold N positions in the network do not increase the number of interlocks linearly, but multiplicatively by the factor N*(N-1)/2.

This also reveals a mechanism that explains the extreme inequality in the distribution of social capital discussed above. The number of interlocks maintained by a single individual is limited by that person’s time budget and talent for establishing contacts. Only a few network-virtuosos accumulated substantial “wealth.” The organization can again cumulate the social capital of the big linkers by appointing them to its BoD. We can call this the “leverage” effect of organizations.

Figure 2 shows the ego network of Northern Pacific Railroad in 1900, about a year before the stock market crash of 1901. This crash was caused in no small measure by the conflict between rival bank groups fighting for control of Northern Pacific. This episode illustrates that not only cooperation is organized in networks, but that con- flicts and rivalries are also played out there. In the ego network of Northern Pacific Railroad in 1900, we find the rival parties represented: Edward Harriman (Union Pacific) and Jacob Schiff (Kuhn & Loeb) on one side; J. P. Morgan and James Hill (Great Northern Railroad) on the other. The main actors involved in the struggle for control of Northern Pacific sit in various combinations on the boards of directors of three companies: Northern Pacific Railroad, Erie Railroad, and Western Union Tele- graph. Through their membership on these boards, they could keep an eye on each other, exchange information, enter coalitions and leave them again (Carosso 1970). Social capital can be used to strengthen cooperation and trust (social embedded-

23 ness). Social capital can also be used by rivaling individuals to secure advantages in the competitive struggle (brokerage, structural holes).

8. Stability over time: multilevel regression analysis The regression analysis presented in Table 4 covers a time period of about forty years (1896 – 1938) during which the First World War, the Great Depression, and the first phase of the Nazi regime occurred. The historical context in which the companies operated in the first half of the twentieth century has changed considerably. We could assume that the causal relations analyzed in the preceding section were not stable during this period. This would mean that the regression coefficients mea- suring the strength of a variable’s influence vary relatively greatly during these years.

Figures 3 and 4 show the regression lines of the variable ‘banker’ for Germany and the United States for each sample year.23 It appears as if the variation of the regression coefficients in Germany would be clearly stronger than in the United States. In Germany, it rises from 2.1 (1896) to 11.1 (1928). This variation is definitely smaller in the United States. We label the variance of the regression coefficients for the 2 variable ‘banker’ as σ u1. If this variance were significant, it would be reasonable to assume that the causal relation had changed over time.

Figure 3: Regression lines - Germany

100

Degree 20

0 012345678910 -20

-40 Banker (N) 1896 1914 1928 1938 Figure shows regression lines for the relationship between the number of bankers sitting on the supervisory board (x-axis) and the degree of the firm in the network (y-axis). Unstandardized regression coefficients: 1896: 2.1; 1914: 5.6; 1928: 11.1; 1938: 9.9

23 The unstandardized regression coefficients come from the regression whose findings are presented in Table 4. 24 Figure 4: Regression lines – U. S.

50 45 40 35 30 25

Degree 20 15 10 5 0 012345678910 Banker (N) 1900 1914 1928 1938 Unstandardized regression coefficients: 1900: 4.4; 1914: 3.9; 1928: 4.1; 1938: 3.0

The samples for the United States and Germany were pooled and a multilevel regression was calculated.24 The first levels consist of variables that are already included in table 4 (fixed part). The second level consists of the four years that we interpret as varying historical context (random effects). The results and technical details are found in Table A2 (appendix). The findings are briefly summarized here.

At first, the coefficients of the pooled samples (fixed part) support the results that are already presented in Table 4: all variables have a significant influence on the degree of companies. The coefficients for the pooled sample are, in a certain sense, an average for the period 1896-1938. Compared to the OLS regression, the variable ‘banker’ now only has a slightly lower level of significance. This can be explained by the fact that this variable now varies on level 2 (random effect).

However, no random effect is significant overall, meaning we could not prove that the strength of the variable’s influence varies between the years 1896 and 1938. First, we can explain this finding technically: On level 2 there are only four years (N=4). Therefore, the standard error of the estimations is very large. The only variable for which the estimation procedure delivers interpretable results is the variable ‘banker.’ For all other variables, the procedure does not produce any converging estimations. No random effect can be estimated for the United states for the intercept either. This

24 We could not calculate a panel analysis because the companies were not identical throughout the entire time period. Between 1896 and 1938, only about 25-30 % of the firms are in each year of the sample. For these firms, the autocorrelation could become a problem in the pooled samples. The Durbin-Watson Test for Germany results in a figure of 1.96, and for United States, 1.86. Therefore, autocorrelation does not play a role. 25 can be explained by the small variance on level 2 (1.2 percent, next to the last line in Table A2).

A substantive interpretation of the results means that the causal relations found in Table 4 are relatively stable compared to the variation of the historic context. Throughout the entire period, bankers in both the United States and Germany played a major role in the network. These big linkers are major figures in the network of the economic elite. They have monopolized social capital both at the personnel level as well as that of the company. Similar considerations also hold for the remaining variables: the influence of company size, the (regional) network-effect, and the influence of the economic sector (e.g. railroads in the United States) were also relatively stable over the entire period. Germany experienced several catastrophes in the first half of the twentieth century. It appears as if the economic elite was able to protect their privileged network of contacts against these disasters relatively well.

9. Summary At the end of the nineteenth century and with the spread of managerial capitalism, a relatively dense network of interlocks developed that could be used by its members for different purposes. Access to this network was a valuable resource because market opportunities were linked to it in a broader sense for the economic elite. The social capital incorporated in the networks was translated into “capital” in a twofold sense for managers and supervisory board members of big companies: it was the result of past involvement in interlocking relationships and therefore represented an “investment.” At the same time, social capital was an unspecific resource, which – like money – could be used in pursuing a variety of interests.

Among big companies, social capital was unequally divided. The distribution structure of social capital could be approximated by a power law distribution introduced by Pareto to demonstrate the unequal distribution of income in industrialized societies. In addition to economic and human capital, social capital represents another important dimension of social inequality.

The units of analysis were companies and the network of contacts linking them. It should be taken into account that the corporate network is produced by people who

26 become go-betweens for the big companies because they serve on several boards of directors. The unequal distribution of social capital is therefore also always an unequal distribution of interlocks available to people due to their formal position in the network. A top manager who heads a major company with a high degree has more market opportunities at his disposal than the manager of an isolated company. The unspecific character of social capital makes it possible to pursue business interests as well as personal career interests in the network of contacts.

Big linkers who have dozens of mandates create junctions at which paths of information and communication intersect. Companies that appoint several big linkers to their BoDs can accumulate social capital very quickly and come directly into contact with a large circle of companies. In this way, the organization compounds the capital accumulated by individuals and then makes it available to that exclusive circle of individuals who belong to the top echelons of management in the organization (leverage effect).

The focus of this analysis did not concentrate on people, but on the characteristics of companies that accumulate relatively large amounts of social capital. The analysis of regional distribution structures produced particularly clear results for the United States: about 70 percent of social capital was concentrated in five cities, wherein New York played a particularly prominent role. Interlocking directorates did not integrate the major companies in a national network spanning the continent. The form of integration can be better described by using the terms center and periphery: firms located in the large metropoles created junctions for paths of communication and decision-making, and they exerted a strong power of attraction. Accordingly, the top managers in firms located in the metropoles had more social capital at their disposal than their counterparts in peripheral firms.

Bankers played a major role in the networks. The fact that they held many positions and, at least in the first half of the twentieth century, dominated the networks does not mean that they “ruled” industrial enterprises (Hilferding 1955). However, bankers (as a collective) did have access to the most contacts when compared with other managerial groups, and they created the main information junctions. Even though they did not directly influence business decisions as a rule, they were indeed in-

27 formed about all decisions and could make their information available to a select few. The bankers possessed – more than any other group of managers – privileged “market opportunities” in the network.

Appendix

Table A1: Regional interlocking of large US firms 1928 - Number of interlocks per firm State NY PA IL NJ MA DE MI VA OH MT MD CT CA Outdeg NY 10.14 0.84 1.03 0.74 0.72 0.57 0.29 0.32 0.61 0.30 0.20 0.15 0.22 16.9 PA 3.62 6.66 0.41 0.34 0.03 0.21 0.07 0.24 0.00 0.24 0.21 0.07 0.14 12.3 IL 2.90 0.26 6.43 0.17 0.21 0.10 0.02 0.10 0.55 0.24 0.10 0.00 0.14 11.4 NJ 6.69 0.85 0.62 0.31 0.23 0.38 0.23 0.23 0.31 0.31 0.08 0.08 0.08 10.9 MA 4.06 0.06 0.50 0.17 4.72 0.00 0.00 0.00 0.00 0.17 0.28 0.44 0.28 11.1 DE 5.60 0.60 0.40 0.40 0.30 0.40 0.70 0.20 0.20 0.20 0.00 0.00 0.20 9.6 MI 5.83 0.67 0.17 0.50 0.00 0.83 0.00 0.17 0.83 0.17 0.17 0.00 0.00 9.8 VA 4.22 0.44 0.33 0.33 0.00 0.33 0.11 0.44 1.44 0.22 0.22 0.00 0.33 8.9 OH 3.24 0.10 1.05 0.24 0.00 0.10 0.19 0.62 2.95 0.10 0.14 0.00 0.00 9.0 MT 3.44 0.67 1.00 0.44 0.22 0.11 0.00 0.33 0.33 0.44 0.33 0.00 0.33 8.3 MD 2.88 1.00 0.38 0.13 0.50 0.00 0.13 0.25 0.25 0.38 1.25 0.00 0.25 7.5 CT 2.83 0.33 0.17 0.17 1.50 0.00 0.00 0.00 0.00 0.00 0.00 2.17 0.33 7.7 CA 1.32 0.21 0.26 0.05 0.26 0.05 0.00 0.05 0.00 0.11 0.11 0.05 3.37 6.2 Indegree 16.6 12.2 11.4 10.8 12.0 10.5 9.5 9.0 9.1 8.9 7.9 7.3 6.5 Firms (n) 119 29 42 13 18 10 6 9 21 9 8 6 19 309 Note: Figures show number of interlocks per firm. Figures in the diagonal refer to intrastate interlocking; figures off-diagonal refer to interstate interlocking. The number of firms varies greatly between states. Thus, the size of the network varies greatly, making it difficult to compare densities. Consequently, the number of interlocks per firm has been computed. The number of firms in each state is shown in the last row (n). States with less than six firms have been excluded. Regional matrices have been computed for Germany and the United States for all years. They are available upon request from the author. Example: There are 119 firms located in New York State. On average, these firms are connected to 10.14 firms also located in NY (intrastate interlocks); they are connected to 0.84 firms located in Pennsylvania (interstate interlocks).

28 Table A2: Multilevel Regression Germany – United States Germany United States B T Sig. B T Sig. Fixed Part Intercept -10.40 -4.49 0.003 2.96 10.5 0.000 Banker 6.75 3.05 0.047 3.85 10.4 0.000 Assets (ln) 7.52 14.83 0.000 1.99 10.9 0.000 Region 0.08 4.90 0.000 0.02 5.5 0.001 Bank -2.38 -1.64 0.101 5.01 11.0 0.000 Steel/RR 5.62 3.80 0.000 3.07 5.6 0.000 Random Part Variance Wald Z Sig. Variance Wald Z Sig.

Intercept (σ²uo) 12.86 1.26 0.209 - - -

Banker (σ²u1) 18.74 1.32 0.188 0.47 1.29 0.197

Covariance (σuo1) 13.26 1.25 0.210 - - - Pearson's r 0.85 0.210 - - -

Model fit -2Log σ²e -2Log σ²e Intercept (base line) 11727.6 586.8 9063.6 73.7 + fixed effects 10752.5 353.8 7463.6 32.3 + random effects 10546.2 295.2 7432.0 31.2 Level-2 variance 24.4% 1.2% Number of firms (N) 1231 1178 Level 1: firms (pooled sample); level 2: years (N=4). Steel: steel industry (Germany); RR: railroads (U.S.) -2Log: -2LogLikelihood. Three models have been computed: The baseline model (intercept only); fixed-effects model; random-effects model. -2Log and σ²e (unexplained residuals) show how the model fit improved stepwise. Level 2 variance: intraclass correlation (proportion of total variance at level 2.)

Model: yij = βo + β1x1ij + β2x2ij ...... + uoj + u1jx1ij ..... + eij

Fixed part (bold):

βo: intercept (variance of intercepts between years: σ²uo).

β1: β-coefficient for dependent variable x1ij (= number of bankers in firm i in year j); variance of β1-coefficients between years: σ²u1

β2: β-coefficient for dependent variable x2ij (= Ln(share capital))

Random part (italics): uoj : error-term of intercept; var(uoj ) = σ²uo u1j: error term of β1-coefficient; var(u1j) = σ²u1 eij: residual term, level 1; var(eij) = σ²e cov(uoj, u1j): covariance between intercepts (uoj) and β1-coefficients (u1j) = σuo1. Cf. Goldstein (1995: 17-23).

Example: The cov(uoj, u1j) between the intercepts and the β1-coefficients for the variable "banker" for Germany is 13.26; variance of the intercepts = 12.86; variance in β1 = 18.74. Pearson's r between the intercepts and the β1-coefficients: 13.26/(√18.74*√12.86) = 0.85. Interpretation: The higher the average degree (density) is in a given year in Germany (intercept), the higher the strength of the influence of the variable "banker" (β1). However, Pearson's r is not significant (α≤ 0.210) because of the small number of years on level 2 (N=4).

29 Level-2 variance (intralevel-2 unit correlation = proportion of the total variance which is between years) is 24.4% for Germany, but only 1.2% for the U.S. Almost all variation in y is due to level 1 variables. Random effect of the intercept for the U.S. has not been computed as no convergence in the iteration process has been achieved. Method of estimation: maximum likelihood (ML).

References

Baltzell, Digby (1979): Philadelphia Gentlemen: The Making of a National Upper Class. Chicago: Quadrangle.

Bearden, James und Mintz, Beth (1985): Regionality and Integration in the American Inter- lock Network. Pp. 234-249 in: Frans Stokman et al. (eds.), Networks of Corporate Power. Cambridge: Polity Press.

Béraud, Alain (2005): La notion de loi dans l’œuvre de Pareto. Colloque International Lille, 22-24 septembre 2005 (http://www.u-cergy.fr/beraud/pareto.pdf).

Borgatti, Stephen et al. (1998): Network Measures of Social Capital. Connections 21: 27-36.

Bourdieu, Pierre (1980): Le capital social. Actes de la recherche en sciences sociales 31: 2- 3.

Bourdieu, Pierre (1983): Ökonomisches Kapital, kulturelles Kapital, soziales Kapital. Pp. 183- 198 in: Reinhard Kreckel (eds.), Soziale Ungleichheiten. Göttingen: Schwartz (Soziale Welt, Sonderband No. 2).

Burt, Ronald (2001): Structural Holes versus Network Closure as Social Capital. Pp. 31-56 in: Nan Lin et al. (eds.), Social Capital: Theory and Research. New York: De Gruyter.

Burt, Ronald (2006): Interlocking Directorates Behind the S&P Indices. University of Chicago Graduate School of Business (http://faculty.chicagogsb.edu/ronald.burt/research/IDS&P.pdf).

Carosso, Vincent (1970): Investment Banking in America. Cambridge: Harvard University Press.

Chandler, Alfred (1977): The Visible Hand. Cambridge: Harvard University Press (Belknap).

Coleman, James (1988): Social Capital in the Creation of Human Capital. American Journal of Sociology 94 (Supplement): S95-S120.

DeLong, J. Bradford (1991): Did J.P. Morgan’s Men Add Value? Pp. 205-236 in: Peter Temin (eds.), Inside the Business Enterprise. Chicago: The University of Chicago Press.

Fohlin, Caroline (2006): Finance Capitalism and Germany's Rise to Industrial Power. Cambridge: Cambridge University Press (im Erscheinen).

30 Fukuyama, Francis (1999): Social Capital and Civil Society. George Mason University. (http://www.imf.org/external/pubs/ft/seminar/1999/reforms/fukuyama.htm)

Goldstein, Harvey (1995): Multilevel Statistical Models. London: Arnold.

Hilferding, Rudolf (1955) [1910]: Das Finanzkapital. Berlin: Dietz.

Kadushin, Charles (2004): Too Much Investment in Social Capital? Social Networks 26: 75- 90.

Katz, Michael und Shapiro, Carl (1985): Network Externalities, Competition, and Compatibility. American Economic Review 75: 424-440.

Katz, Richard (1998): Japan the System that Soured. New York: Sharpe.

Kono, Clifford et al. (1998): Lost in Space: The Geography of Corporate Interlocking Directorates. American Journal of Sociology 103: 863-911.

Lévi-Strauss, Claude (2002)² [1947]: Les structures élémentaires de la parenté. New York: Mouton de Gruyter.

Luhmann, Niklas (1995): Inklusion und Exklusion. Pp. 237-264 in: Niklas Luhmann, Soziologische Aufklärung. Opladen: Westdeutscher Verlag.

Macneil, Ian (2000): Relational Contract Theory: Unanswered Questions. Northwestern University Law Review 94: 877-908.

Mintz, Beth und Schwartz, Michael (1985): The Power Structure of American Business. Chicago: University of Chicago Press.

Mizruchi, Mark (1982): The American Corporate Network 1904-1974. Beverly Hills: Sage.

Mizruchi, Mark (1992): The Structure of Corporate Political Action. Cambridge: Harvard.

Newman, M.E.J. (2005): Power Laws, Pareto Distributions and Zipf's Law. Contemporary Physics 46: 323-351.

Newman, M.E.J. et al. (2002): Random Graph Models of Social Networks. Proceedings of the National Academy of Sciences of the United States of America (PNAS) 99: 2566-2572.

Pareto, Vilfredo (1967) [1896]: Écrits sur la courbe de la répartition de la richesse. Oeuvres complètes, Tome III. Genève: Droz.

Pennings, Johannes (1980): Interlocking Directorates. San Francisco: Jossey-Bass.

Piketty, Thomas und Saez, Emmanuel (2003): Income Inequality in the United States, 1913- 1998. Quarterly Journal of Economics 68: 1-39.

Prévost, Jean-Guy (2002): Genèse particulière d'une science des nombres: L'autonomisation de la statistique en Italie entre 1900 et 1914. Actes de la recherche en sciences sociales 141/142: 98-109.

Putnam, Robert (1995): Bowling Alone. Journal of Democracy 6: 65-78.

Sassen, Saskia (2001): The Global City: New York, London, Tokyo. Princeton: Princeton University Press. 31

Simmel, Georg (1987) [1900]: Philosophie des Geldes. Berlin: Duncker & Humblot.

Stiglitz, Joseph (2000): Formal and Informal Institutions. Pp. 59-67 in : Ismail Serageldin und Partha Dasgupta (eds.), Social Capital: A Multifaceted Perspective. Washington: World Bank.

Sweezy, Paul (1962): Interest Groups in the American Economy. Pp. 158-188 in: Paul Sweezy, The Present as History. New York: Monthly Review Press.

Useem, Michael (1978): The Inner Group of the American Capitalist Class. Social Problems 25: 225-240.

Useem, Michael (1984): The Inner Circle. London: Oxford University Press.

Uzzi, Brian (1990): The Sources and Consequences of Embeddedness for the Economic Performance of Organizations. American Sociological Review 61: 674-698.

Wasserman, Stanley and Faust, Katherine (1994): Social Network Analysis. Cambridge: Cambridge University Press.

Weber, Max (1964): Wirtschaft und Gesellschaft, Erster Halbband. Köln: Kiepenheuer & Witsch.

Wellhöner, Volker (1989): Großbanken und Großindustrie im Kaiserreich. Göttingen: Vandenhoeck & Ruprecht.

Williamson, Oliver (1985): The Economic Institutions of Capitalism. New York: Free Press.

Windolf, Paul (2002): Corporate Networks in Europe and the United States. Oxford: Oxford University Press.