摘 要：使用2005年人口小普查数据研究了独生子女之间的婚配选择问题，发现独生子女更可能与独生子女结婚，且非独生子女也更可能与非独生子女结婚，所以认为独生子女与非独生子女之间“随机婚配”的假定并不符合实际。然而，以往有关生育政策调整的人口预测大都基于这一假定之上，这导致预测结果一方面低估了“双独”和“双非”夫妇的数量，另一方面高估了“单独”夫妇的数量。同时，本文还以独生子女的婚配选择问题为例讨论了嵌套模型中Logistic回归系数的可比性问题。直接比较Logistic回归系数可能导致错误的结论，借助“y*标准化”法和“KHB分解法”可以在一定程度上解决Logistic回归系数不可比的问题。

关键词：2005年小普查数据；独生子女；“随机婚配”假定；Logistic回归系数

中图分类号：C92-05 文献标识码：A 文章编号：1000-4149（2016）03-0043-09

Abstract：Based on the 2005 minicensus data， this paper studies the mating selection problem of only child under the age of 30 and finds that only child is more likely to marry only child and that people with siblings are more likely to marry people with siblings. Therefore， the “Random Mating” hypothesis is far beyond the truth. However， the previous studies concerning fertility policy adjustment were all based on this hypothesis， which probably underestimated the quantity of couples that both/neither are only child and overestimated the quantity of couples that either is only child. Taking the mating selection problem of only child for example， this paper also discusses the comparability of coefficients of nested Logistic regression and points out that direct comparison of the coefficients of nested Logistic models will result in incorrect conclusion. “y* standardization” and “KHB decomposition” are two methods that can resolve the incomparability problem of the coefficients of nested Logistic regressions.

Keywords：2005 minicensus data； only child； “Random Mating” hypothesis； coefficient of Logistic regression

一、导论

中国的计划生育政策，特别是30 多年来独生子女政策的全面推行不仅产生了数以亿计的独生子女[1]，而且创造了“双独”、“单独”和“双非”

“双独”夫妇指夫妇双方均为独生子女；“单独”夫妇指夫妇中只有一方是独生子女；“双非”夫妇指夫妇双方都不是独生子女。等独具中国特色的夫妇类型[2]。研究这些夫妇类型一方面有助于我们辨识一些特殊的家庭结构（如“四二一”家庭结构），另一方面也有助于当前和未来生育政策的制定（如“双独二孩”政策和“单独二孩”政策都以夫妇双方是否是独生子女为条件）。所以，研究“双独”、“单独”和“双非”夫妇的数量、结构及其在未来的发展趋势是关系到家庭养老和生育政策调整等一系列重大决策的重要问题。……