Relationship Dissolution

Abstract

This chapter provides an overview of recent trends in divorce and relationship dissolution in Australian society. It commences by describing historical and contemporary trends in marriage dissolution using Australian Bureau of Statistics (ABS) data. These data offer little insight into the dissolution of cohabiting relationships. To fill this gap we use unit record data from the first 10 waves of the Household, Income and Labour Dynamics in Australia (HILDA) survey to compare and contrast the stability of marital compared to cohabiting relationships since 1995. Next we examine the consequences of marital and cohabiting relationship dissolution for income and health outcomes for men and women. We conclude with a brief discussion of the extent to which unstable marriages have been replaced by cohabitation.

Appendix 5.1: Methodological Notes

The data used to examine dissolution from cohabiting relationships came from the first ten waves of The Household, Income and Labour Dynamics in Australia (HILDA) survey, collected between 2001 and 2010 (see Technical Appendix).

The sample for Table 5.1 included all respondents at wave 1 who indicated that they had married. The analysis was restricted to those in their first marriage as the processes of divorce surrounding remarriages are very different from those of first marriages (Carmichael et al. 1997; Sweeney 2010). Retrospective marriage history data were used. As only first marriages are under consideration in this analysis, if a respondent had married once the information about their present marriage was included in the calculation of the dependent variable. If the respondent had been married more than once then information about their first marriage was included but not information about subsequent marriages.

The sample for Table 5.2 and Figs. 5.6 and 5.7 includes respondents who formed a cohabiting or marital relationship between 1995 and 2010. Prior to 2000 retrospective relationship history data are used, and after 2000 panel data were used. If a respondent had formed more than one cohabiting relationship during that time we included their most recent or current cohabiting relationship. We restricted the marriage sample to those who entered their first marriage only. People who were cohabiting after marriage were also excluded from the analytic sample. To capture the main relationship processes identified by previous research (Qu et al. 2009), we differentiated between marriages, cohabitations that ended in marriage and cohabiting only relationships.

The sample for the analysis presented in Figs. 5.8, 5.9 and 5.10 includes all respondents in HILDA waves 1–10 who were married or cohabiting at wave 1 or who married or started cohabiting during the panel. We follow them over the panel and observe those relationships that end in separation. The model for Fig. 5.10 includes controls for relationship duration, age, employment status, highest level of education, and number of children in household 50 % or more of the time. The models for Figs. 5.9 and 5.10 include a range of basic controls including relationship duration, age, number of children under 18 in the household 50 % or more of the time, household income, employment status, highest level of education and health status at the previous wave.

Given that we had repeated observations on individuals over time, the structure of our data violates the assumption of independent observations and ordinary least squares regression would not be appropriate. Instead we used a linear fixed-effects model to account for clustering of observations by individual and control for between individual variation (Singer and Willett 2003). This approach is also appropriate for unbalanced panels. The fixed-effects model controls for unobserved heterogeneity because it produces estimates that are net of all observed and unobserved differences between individuals that are time-invariant. Models were estimated using the fixed effects option in xtreg in STATA 11.2 (StataCorp 2012).

For the results presented in Fig. 5.8, we use equivalised disposable annual household income as our main dependent variable. Our income measure was equivalised using the OECD-modified equivalence scale (Organisation for Economic Co-operation and Development 2008). In this approach the first adult within the household is assigned a value of 1, a value of 0.5 is assigned to each additional adult member (aged 15 or over) and a value of 0.3 is assigned to each child. We used this scale as it is the equivalence scale considered best suited to the Australian situation by the ABS (Australian Bureau of Statistics 2006). Preliminary analyses showed that using alternative equivalence scales, such as dividing income by the square root of the number of household members, did not lead to different conclusions. In addition we excluded extreme outliers on household income; respondents who reported a household income (not equivalised) of more than $300,000 AUD each year.

For Figs. 5.9 and 5.10 we used the mental and physical health domain measures derived from the Short-Form 36 (SF-36). The SF-36 is a self-completed measure of health status comprising 36 items that measure two main health domains as well as eight health constructs and is a well-validated tool for measuring population health (McHorney et al. 1993; Butterworth and Crosier 2004). For physical and mental health domains, scale scores ranged from 0 to 100, where lower scores indicated poor health and higher scores indicated excellent health (Ware et al. 2000).