25.01.2015 Views

CONNECTIONS - INSNA

CONNECTIONS - INSNA

CONNECTIONS - INSNA

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>CONNECTIONS</strong><br />

Configurations of Homophily<br />

elements of the power set). With five<br />

sociodemographic dimensions there are 32<br />

possible configurations. Given a large number<br />

of individuals who are heterogeneous along<br />

several dimensions, it is not surprising that all<br />

configurations of homophily are represented.<br />

The first row is the null set, absence of<br />

homophily on all five dimensions. The most<br />

common configuration is number 16, which<br />

includes similarity on four of the five<br />

dimensions (Sex, Race, Education, and<br />

Religion). This configuration contains 27%<br />

more relationships than the second most<br />

frequent configuration (number 32), which is the<br />

complete homophily configuration.<br />

Configuration Frequency<br />

To determine which configurations occur more<br />

often than expected by chance it is necessary to<br />

compare the observed number of relations in<br />

each configuration to a baseline level that<br />

represents random association. The theoretical<br />

baseline population in this research comprises<br />

all possible relations in the United States in<br />

1985. Like many surveys of the general<br />

population, the GSS underrepresents males<br />

(Davis & Smith, 1996) and the marginal<br />

distributions of other demographic dimensions<br />

do not match U.S. population estimates. To<br />

create a representative baseline, the data were<br />

weighted to reflect the general population using<br />

U.S. census estimates from 1985 to weight by<br />

age, sex, and race and from 1990 to weight by<br />

education (U.S. Census Bureau, 1990, 2000).<br />

Respondents were not weighted by religion as<br />

the GSS comprises the best available national<br />

estimate of religious identification. Using the<br />

GSS as the basis for computing the baseline,<br />

rather than census estimates that don’t include<br />

religion, has the advantage of preserving the<br />

consolidation of parameters. Maintaining the<br />

correlation of dimensions within individuals is<br />

vital when examining how multiple dimensions<br />

simultaneously draw people together. An<br />

expected proportion was computed for every<br />

configuration of homophily and can be seen in<br />

the Baseline column of Table 1 (computation of<br />

the baseline is presented in the Appendix).<br />

Examination of Table 1 reveals that the<br />

distribution of dyads across configurations<br />

varies from that expected by chance. For<br />

instance, the top row represents relations that are<br />

different on all five dimensions. Examining the<br />

“All” column we see that the 0.1% of dyads<br />

observed in this configuration is well below the<br />

3.1% expected by chance. Conversely, the<br />

12.8% of dyads in the bottom row, homophily<br />

on all dimensions, is well above the 1.6%<br />

expected by chance. To evaluable how<br />

dimensions interact to affect the frequency of<br />

observed dyads, each configuration was coded<br />

as exceeding the baseline frequency or not. This<br />

was determined using a z-test for proportions<br />

with an alpha of .05. Those configurations<br />

whose proportion was significantly higher than<br />

the baseline proportion were coded as 1;<br />

otherwise they were coded as 0. Ten<br />

configurations occurred more often than<br />

expected by chance while twenty-two did not.<br />

Role relations correspond to common foci that<br />

bring people together and provide the<br />

opportunity for relationships to develop<br />

(Marsden, 1990). Examining role relations<br />

independently can provide insight to the<br />

multidimensional nature of homophily induced<br />

through substructures. The GSS allowed<br />

respondents to select multiple ways in which<br />

they were connected to each alter. This research<br />

utilizes the following relations: kin, coworkers,<br />

friends, and group members. Since kin relations<br />

evolve through a different process than most<br />

other relations a distinction was made between<br />

kin and nonkin. Relations with parents, siblings,<br />

children, and other family members were<br />

collapsed into the kin relation. To help ascertain<br />

the homophily induced by substructural foci and<br />

not pre-existing kin relations, only nonkin<br />

friends, coworkers, and group members were<br />

considered. Table 1 reports the distribution of<br />

dyads across homophily configurations by type<br />

of role relation. The proportion of dyads of each<br />

relation type in each configuration was<br />

compared to the baseline, with those exceeding<br />

26

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!