Page 20 - IJPS-4-1

Transition to first marriage in China

four birth cohorts corresponding to the described four social-historical periods, representing those born in 1946–1955,
1956–1965, 1966–1975, and 1976–1993, respectively, with a relatively even distribution of each cohort (Table 1). In data
presentation, I showed descriptive statistics of the sample by gender and hukou origin. To further explore respondents’
age at first marriage, I conducted Kaplan–Meier survival analysis and Cox proportional hazard analysis with the sample
restricted to those aged 30 and above (n = 23,253). The rationale behind this sample restriction is that, to be able to
observe the full tempo of transition to first marriage of a cohort, we need to have an observation frame including the full
range of their marriage active years, which may last from their late teens until late 30s. In the case of China, previous
research indicates that, by the age of 30, the majority have married (Ji and Yeung, 2014; Tian, 2013). As such, using age
30 (relative to 35 or 40) as the cutting point in this analysis allows us to not only preserve the biggest sample size but also
observe the momentum of entry to marriage among youths, particularly of the youngest cohort.
The data for further analysis of homogamy patterns among married couples were constructed in two steps. First, I
coded four first-marriage cohorts based on the reported years of respondents’ first marriage formation, excluding those
who had never married (n = 1,743, 7.5%), those who were remarried (n = 465, 2%), those who were separated or divorced
(n = 628, 2.7%), or those who were widowed (n = 930, 4%). The rationale of focusing on first marriage unions is two-
folded. For one thing, the substantive topic in this paper concerns transition to first marriage as a marker of adulthood status
attainment. However, in cases of remarriage, divorce, or widowhood, we cannot guarantee valid information pertaining
to first marriages. For another, given that the majority of reported marriages in the data are first-time marriages (90.2%),
which reflects a relatively low proportion of divorce. This is likely to introduce a moderate but not substantial level of
upward bias in the estimates of homogamy patterns (Mu and Xie, 2014). In the second step, I compiled a couple-education
profile data using respondents’ reports of their own and spouse’s educational attainment. After list-wise deletion of missing
data (n = 147), we obtain a final analysis sample with 19,340 couples. I performed log-linear and log-multiplicative layer
effect analysis (Xie, 1992) to examine the strength of educational homogamy across historical periods.

4.2. Measurement
4.2.1. Dependent variables

In the analysis of marriage timing, the dependent variable is the age at marriage, which was calculated from two variables
in the original data: Respondents’ year of birth and their reported year of marriage (the latter minus the former). In
homogamy analysis, I employed a four-level educational categorization to identify couples’ educational profiles: Primary
School or less, junior middle school, high school, and college or above.

4.2.2. Covariates
Respondents’ birth cohort and marriage cohort are represented as multiple dichotomized variables (1 = yes and 0 = no).
The three measures of family background were coded as follows: hukou origin (1 = urban hukou); father’s education (1
= illiterate and primary school, 2 = junior middle school, and 3=high school and above), and employment status during
respondents’ adolescence (1 = full-time employed, including those who had “stable employment,” who held a position in
family businesses, who were business owners, and who retired from stable employment with benefits; 2 = farming; and 3
= others, including a variety of vulnerable job situations, such as migrants, temporary laborers, laid-off workers, and those
who were economically inactive or lost earning capacities).
Demographic variables include dichotomous measures for gender (male vs. female) and ethnicity (Han vs. non-Han).
Dummy variables as indicators of geographic regions (east, central, west. and northeast) were controlled for in Cox
models to account for the regional heterogeneity of marriage patterns (Ji and Yeung, 2014). As mentioned earlier, dummy
variables of the three survey years were constructed to control for potential effects of survey time on outcome variables.

5. Results

5.1. Summary statistics
Table 1 presents the weighted descriptive statistics for the sample by hukou status and gender. Respondents’ educational
attainment varies greatly by the intersection of their hukou status and gender, with rural females receiving lowest education
(5.55 years of completed education; 6% attended college or higher), urban males receiving the highest (12.02 years;
37% attended college or above), and a glaring 6.5-year gap between them, reflecting the long-term gender inequality in
education among the Chinese population until recent cohorts (Treiman, 2013). If breaking down the educational measure

14 International Journal of Population Studies | 2018, Volume 4, Issue 1