Union Formation and Dissolution
In this chapter, we describe trends in educational differences for several dimensions of marriage and union formation—early marriage, nonmarital cohabitation, divorce, and husband’s education. Before presenting these figures, we describe the data used for these analyses.
The National Fertility Survey
The National Fertility Survey (NFS) has been conducted by the National Institute of Population Problems and Social Security Research (Japanese Ministry of Health, Labour and Welfare) every five years since the second survey in 1952.1 The analyses presented below are based on data from the 8th through the 14th surveys, conducted in 1982 and 2010, respectively. The NFS is a nationally representative survey of two populations: married women aged 18-49 and unmarried men and women aged 18-49.2 The NFS is based on a two-stage systematic sampling of all census enumeration districts in Japan. More specifically, respondents are randomly selected from a systematic sample of between 325-840 census tracts drawn from the roughly 1000 tracts surveyed by the Kokumin Seikatsu Kiso Chosa (Basic Survey of National Life, also conducted by the Ministry of Health, Labour and Welfare) 1The first survey was conducted in 1940. One exception to the five-year interval is the three-year interval between the 12th (2002) and 13th (2005) surveys, a change designed to align the NFS survey years with census years. The conducting institution changed its name from the Institute of Population Problems to the National Institute of Population Problems and Social Security Research in 1996.
2The 8th survey in 1982 was the first to sample unmarried men and women, motivating our focus on the 8th to the 14th surveys.
© The Author(s) 2017
J.M. Raymo and M. Iwasawa, Diverging Destinies,
Population Studies of Japan, DOI 10.1007/978-981-10-0185-7_3
which were systematically sampled from the full population of census tracts established for the most temporally proximate national census. Questionnaires were distributed to all 18-49-year-old unmarried men and women and all 18-49-year-old married women residing in the sampled census tracts. The total sample size ranges from 13,346 in 1982 to 19,544 in 1992. The composition of the surveys with respect to marital status has changed over time as the trend toward later marriage has become more pronounced in Japan. In the 8th survey in 1982, 63 % of respondents were married women, but this proportion declined to 43 % in 14th survey in 2010. The response rates for both the married and unmarried samples have been consistently high. Not surprisingly, response rates have been higher among married women and response rates for both groups have declined over time. For married women, the response rate declined from 95 % in 1982 to 87 % in 2010 and the corresponding decline for unmarried women was from 86 % to 74 %. While the decline in response rates may be a cause for concern, it is important to note that these response rates are much higher than for most recent social surveys in Japan. For example, the response rate for the 2010 round of the Japan General Social Survey was only 62 %.
The NFS is the best available source of data on family behavior over an extended period of time. Each survey contains information about the timing of women’s current marriage, the timing of all births within the current marriage, information about cohabitation and divorce experience, information about husband’s educational attainment, and information about women’s employment following marriage and childbirth. More recent surveys include information about the duration of cohabiting unions and the timing of divorce and remarriage, but we are not able to use this information to examine trends in educational differences over long periods of time. The 8th-14th NFS samples include women born between 1932 and 1991, but we focus on women born between 1940 and 1979, splitting the sample into four ten-year birth cohorts. Exclusion of women born outside of this range reduces potential problems with small cohort size and allows us to focus on women who have had sufficient time to experience the events of interest.
Importantly, the NFS has collected consistent measures of women’s educational attainment, allowing us to describe educational differences in family behavior and how those differences have changed over time. Educational attainment is measured with five categories representing the highest level of education completed at the time of the survey: junior high school, high school, vocational school, junior college, and university. The use of consistent categories is useful for making comparisons across time, but it is important to keep in mind the compositional shifts that have taken place with respect to educational attainment. Rising educational attainment means that university-educated women have become a less selective group over time while those who did not complete high school have become a more selective group. This compositional shift should be kept in mind when evaluating change over time in educational differences in family behavior. In her work on diverging destinies, McLanahan (2004) and McLanahan and Jacobsen (2015) circumvented this problem by using reported years of education to construct a relative measure that reflects rising educational attainment (i.e., low, medium, and high education). Because the NFS only provides information on women’s highest completed educational level, we are not able to implement a similar measurement strategy. Subsequent efforts to evaluate socioeconomic bifurcation in family behavior might profitably examine alternative specifications for the measure of socioeconomic status (SES) that are not plagued by this concern about cross-cohort comparability (due to changing patterns of educational selection).
The NFS includes data on both married and unmarried men, but we limit our focus to women for two important reasons. One is the fact that married men are not sampled directly. To examine men in the same way that we examine women would require us to construct records for married men based on the information provided by married women. This is problematic to the extent that the husbands of 18-49-year-old married women are not a representative sample of 18-49-year-old married men. A second reason is that most of the research on diverging destinies focuses on mothers, reflecting the centrality of nonmarital childbearing and single motherhood to our understanding of potential socioeconomic differences in children’s access to resources.
In addition to the absence of direct responses from married men, the cross-sectional design and sampling frame of the NFS introduce some other potential limitations. For example, the NFS data contain no information on women who would have been sampled had they not died prior to the survey and women who were sampled but did not respond to the survey. Very low mortality suggests that the absence of information for the first group will be of little substantive importance, but nonresponse is likely to pose a more serious problem. Despite the consistently high response rates, differential nonresponse with respect to the variables of interest (especially educational attainment) would introduce a potentially important source of bias into the analyses summarized below.
-  https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/34623 (accessed April 25, 2016).