Study of Age at First Marriage in Contemporary China: A Relative Distribution Approach

Shige Song, University of California, Los Angeles

The rising age at first marriage during the past half century in China, as documented by increasing mean or median age, is often attributed to the effective state family planning policy. However, both mean and median measure differences in averages while leave information of other distributional changes out of the analysis. The difference in averages is not the best indicator of the effectiveness of the state family planning policy. The family planning agency pays different attention to those who would have married early from those who would have married late, it cares more about delaying marriage for those who would have married before 20 than further delaying marriage for those who would have married at their 30s. In other words, a rapid increase in the lower tail (e.g. the 10th percentile) provides strong supportive evidence of the effectiveness of the family planning policy; a constant or even decrease in the lower tail shows that the family planning policy is not that effective, even if there is an increase in the upper tail (e.g. 90th percentile). The contrast between these two changes provides more reliable assessment of the effectiveness of family planning policy in delaying age at first marriage. I use the Phase II China in-depth fertility survey data (1987) for this analysis. The public available data includes four provinces: Guangdong, Beijing, Liaoning, and Shandong. I use the method of relative distribution – a way to explore full distributional information (as opposed to the central tendency based statistics like mean or median) in comparing two or more distributions or the same distribution at different time periods. This method combines intuitive graphical representation and hypothesis testing oriented numerical index in a coherent framework. The observed change can be conveniently decomposed into the part that is caused by the shift in median and that is caused by the shape change. In addition, the effects of covariates (both discrete and continuous) can be easily adjusted. The analysis in this paper follows 3 steps: 1) analysis of changes in the lower & upper tail of the age distribution; 2) analysis of changes in the full age distribution (with the decomposition into location shifts and shape changes); and 3) analysis of the influence of differential educational attainment. Preliminary results show that the percentage of late marriage has been increasing from 1950s till 1970s while the percentage of early marriage has been decreasing at the same time; but the increase of late marriage is more prominent than the decrease of early marriage. This trend stops in 1980s when the percentages of both early marriage and late marriage start to decrease. Decomposition shows that from 1950s to 1970s, the upward shifts in location are predominant (compared with the shape changes). The upward shifts in location stop in 1980s. After controlling for the location shift, the shape changes provide some evidence that is directly relevant to the central theoretical interest. The relative distribution between 1950s and 1960s is a U shape curve: both early marriage and late marriage increase during that period of time. From 1960s to 1970s, the percentage of early marriage continues to increase while the percentage of late marriage starts to decrease. From 1970s to 1980s, percentage of both early marriage and late marriage start to decrease. The results cast some doubts of the family planning policy explanation of the change in age at first marriage. The increasing of percentage in early marriage in both 1960s and 1970s, as revealed by the shape change in the decomposition analysis, shows a very different picture as revealed by the mean and median change. Even during the period that the family planning policy works (1980s), the effect is two sided: it reduces the percentage of early marriage and the percentage of late marriage at the same time. This interesting pattern deserves further study.