The Blackwell Companion to Sociology (73 page)

designed by Jonathan Kelley and Mariah Evans, in 1987 (with nine nations

participating) and 1992 (with 18 nations participating), which have generated

invaluable information on subjective aspects of stratification; a third inequality module was carried out in 1999. In addition to the ISSP project, three other

projects have provided highly comparable data of interest to stratification

researchers: (a) the 12±nation International Social Justice Project conducted

surveys in 1991 and 1996 that include many items concerned with subjective

aspects of stratification (see Kluegal et al., 1995); (b) the International Survey of Economic Attitudes, recently launched by Jonathan Kelley and his colleagues,

thus far includes data from six nations and plans tri-annual surveys on an

expanding set of nations; (c) the six-nation survey of Social Stratification in

Eastern Europe after 1989, headed by Szelenyi and Treiman (see Treiman and

Szelenyi, 1993), includes extensive intergenerational and event history data for six formerly communist Eastern and Central European nations.

In addition to these new data collection efforts, Ganzeboom and Treiman have

for some years been engaged in a project to collate and standardize unit-record

data for all known national surveys conducted in the twentieth century that

contain variables pertinent to the status attainment process, in order to develop a comprehensive account of to what extent and in what ways the processes of

educational and occupational attainment vary over time and across societies

and what macrosocial factors account for such variations ± some of the results

of which have been reported above. Thus far, they have obtained some 250

sample surveys from about 40 nations covering most of the twentieth century.

Of these, 113 have been standardized and a common subset of variables has

been made available for public use in the International Social Mobility and

Occupations, Stratification, and Mobility

311

Politics File (Nieuwbeerta and Ganzeboom, 1996). All of the internation-

ally comparable data sets mentioned here are listed in the appendix to this

chapter.

Multilevel Models

One of the most important recent developments in stratification research has

been the ascendancy of multilevel analytic designs (for a review see DiPrete and Forristal, 1994). Generically, these are designs that include at least two levels of data: a micro-level ± almost always individuals ± and a macro-level, which

specifies a social context. The basic idea is to study how the behavior of

individuals varies according to the social context. There are several ways to

analyze multilevel data. The most common approach is to carry out micro-

level analysis for each context separately and then to compare the results,

either formally or informally. This is the implicit design of any comparison of

two or more contexts. Often such comparisons are limited to two points in

time or to two or at most a handful of nations. Such comparisons suffer from

what is known as thè`degrees-of-freedom problem.'' Because any pair of

nations or time points may differ in any number of ways, it is difficult to be

certain that the particular macro-level differences one adduces to explain any

particular differences in the micro-level outcomes are, in fact, the causal agents.

For this reason, comparisons of small numbers of contexts are more useful

for establishing cross-context regularities than for explaining cross-context differences.

The obvious way around the degrees-of-freedom problem is to compare

enough contexts to be able to treat contexts as observations. In this approach,

a micro-level process is modeled separately for each context. Then, in a second

step, the contexts are treated as observations and the coefficients from the micro-model are treated as dependent variables (by using multilevel modeling proced-

ures, both levels can be estimated simultaneously, which is advantageous

statistically). The variation in the micro-level coefficients is explained by variation in the contextual variables. Note that in this approach complete explanation of variation in the micro-level coefficients is not necessary; instead, the coefficients are treated as stochastic, i.e. themselves subject to error.

An interesting new development in stratification research, which exploits the

multilevel approach, is the treatment of time as a variable, studying the impact of temporal change on various stratification outcomes. While many previous studies have compared outcomes at two points in time ± which is logically equival-

ent to comparing two nations ± until recently there have been few studies that

have taken calendar time as a context, studying year-by-year variations in

stratification outcomes, and linking variations in outcomes to macrosocial his-

torical events, such as depressions, wars, revolutions, and social policy changes.

This kind of research has burgeoned in recent years as comparable data have

accumulated over time.

A limitation of both cross-national and cross-temporal multilevel designs is

that it is difficult to generate enough cases to sustain anything other than the 312

Donald J. Treiman

simplest analysis at the macro-level. We seldom have more than 30 or 40 nations

or years, which is hardly enough cases from which to estimate a model with

more than one or two contextual variables. An obvious solution is to cross years by nations. This is the approach Ganzeboom and Treiman are taking in the

analysis of status attainment mentioned above: by crossing data from 40 nations

by 15 five-year birth cohorts, it is possible to generate data for up to 600

(ˆ 15  40) contexts, although in practice there are fewer contexts since not

all cohorts are available for every nation. With such data it is possible to

systematically test hypotheses regarding the effect of contextual variables that vary both over time and across nations; for example, ``parental occupational

status has a smaller effect on educational attainment in welfare states than in

laissez-faire systems,'' where the attributes of the welfare system are measured in each nation separately for each five-year period.

Conclusion

Great progress has been made in the study of social stratification and social

mobility in the past half century, and particularly in the years since the publication of The American Occupational Structure by Blau and Duncan (1967).

Advances in statistical methods, research designs, measurement standardization,

and the availability of high quality cross-nationally comparable data have made

it possible to show that modern societies have become increasingly open over the course of the twentieth century and as nations have industrialized, so that at the beginning of the twenty-first century inequality in occupational status and

income arises more from differences in education than from direct occupational

or income inheritance. Moreover, educational differences are mostly achieved

rather than ascribed, depending more on personal attributes than on social

status. This review has covered only a small fraction of these developments,

but perhaps enough to give a sense both of the current accomplishments of this

area of research and of the remaining challenges.

APPENDIX: CROSS-NATIONAL COMPARATIVE DATA SETS

International Social Survey Project (ISSP). Ann Arbor: Inter-university Consortium for Political and Social Research (distributor); consult http://www.icpsr.umich.edu/

archive1.html

Kelley, J., Zagorski, K. and Evans, M. D. R. International Survey of Economic Attitudes.

International Survey Centre, Institute for Advanced Study in the Behavioral Sciences, Australian National University, Canberra (distributor); consult http://www.

international-survey.org/index.html

Nieuwbeerta, P. and Ganzeboom, H. G. B. (1996) The International Social Mobility and Politics File (a CD-ROM database) Amsterdam: Steinmetz Archive. This is a public use file of a selected set of internationally standardized variables from 113 surveys, a subset of the approximately 250 surveys collected by Harry B. G. Ganzeboom and

Donald J. Treiman. The complete database is catalogued on the Internet: http://

www.fss.uu.nl/soc/hg/ismf/index.htm. Access to the catalogued surveys must be

Occupations, Stratification, and Mobility

313

arranged with Ganzeboom ([email protected]) since use of some of the data sets requires the permission of the original investigators.

Szelenyi, I. and Treiman, D. J. (1993) Social Stratification in Eastern Europe after 1989. Los Angeles: Social Science Data Archive, Institute for Social Science Research, UCLA (distributor); download from http://www.sscnet.ucla.edu/issr/da/SSEE/SSEE.

home.html

22

Social Networks

Bonnie Erickson

Network analysis is thriving in almost every corner of sociology, in every other social science, and in still further territories like mathematics and information science. Network analysis is so widely powerful because it is the serious study of social structure. Social structure is important everywhere, and in addition structural patterns can be abstracted from particular applications and freely applied to many other settings, so that theoretical and methodological approaches

become very widely useful. Structure is as beautiful as it useful. It is lovely

visually, since network structures have attractive patterns that are fun to look at and think about. It is lovely conceptually, since structural thinking produces both the sheer intellectual elegance and the paradoxical surprises that have made formal sociology a charmer since Simmel's day (and Simmel is every true networker's spiritual ancestor and continuing inspiration).

When I define network analysis as the serious study of social structure, some

people are surprised: do not we all study social structure? Not really. Social

structure means the pattern of social relationships linking social actors. The

actors can be of many kinds: people, companies, families, nations, and so on.

The social relationships can be of many kinds: love, hate, cooperation, competi-

tion, admiration, disdain, talking face-to-face, sending e-mail, and so on. But

whatever the details, structure is the concrete network of some ties among some

actors, and this is usually not what people study. Some describe overall features of a structure impressionistically, without actually recording and analyzing

relationships. Many study individual attributes; for example, equating centrality in the world system with national wealth instead of powerful positions within

networks of trade, diplomacy, war, and so forth. Some confuse social structure

with the distribution of some variable, as in thèàge structure'' of a population, which provides important information about the relative size of age groups but

gives us no information about anyone's social ties. Some study one relationship

Social Networks

315

at a time, even though a single relationship is always subject to influence from surrounding relationships. For example, there are many studies of marriages that study each couple in isolation; yet the more wife and husband share their

networks, the more they share domestic tasks.

To illustrate differences in approaches, consider elite Florentine families dur-

ing Cosimo di Medici's rise to power. Why did some families support the Medici,

while others supported the rival faction of oligarchs? Some have tried to explain this in terms of family attributes such as wealth, recency of elite status, or

neighborhood. But Padgett and Ansell (1993) showed that supporters of the

Medici and of the oligarchs were quite similar in such attributes. Padget and

Ansell recorded marriage and business ties among the families and mapped their

overall organization, grouping together families who had similar ties to other

families. They found that families supported the faction they were most linked

to, and also found good structural reasons for the Medici's eventual victory.

Medici supporters had ties to the Medici and almost none to other elite families, including each other. This put the Medici in a powerful position, since their

supporters depended on them for indirect ties to the rest of the elite, and their faction proved to be relatively centralized and cohesive. Meanwhile the oligarchs had many ties among themselves, with several well connected families whose

competition for leadership weakened their party.

The power of the Medici's structural position was not just a fluke of

Florentine society or politics, but is far more general. Burt (1992) finds much

the same for very modern managers: the ones who get ahead faster are the

ones who have more links to separate sets of people who are poorly (if at all)

linked to each other, so that the fortunate manager who links them can control

the flow of information to his or her advantage. The same kind of structural

position pays off in laboratory studies of social exchange, in which the people

who gain the most are the ones with exchange partners who have no one else to

bargain with. Burt traces this analysis back to Simmel and his discussion of

tertius gaudens, or the third party who profits from division between two others.

The same structural pattern has the same outcome in classic theory and in

modern experiments, in Florence centuries ago and in management hierarchies

today.

Thus structural thinking can lead to generalizations that leap across time,

place, and type of setting (as Simmel's examples do). There are many other