Socio-economic determinants of divorce in early twentieth-century Sweden

Abstract

Using a combination of census data and aggregated divorce statistics, this study investigates how socio-economic conditions influenced the risk of divorce among men in different occupations during the 1920s and 1930s in Sweden. The results support the theoretical presupposition that the stability of marriage was associated with the degree of economic interdependence between spouses. Rural, low-income, single-provider households with many children exhibit a significantly lower probability of divorce than urban, dual-provider, high-income households with few children. This lends support to a socio-economic growth hypothesis stating that lower levels of marriage stability first developed in the more affluent strata of society living in urban settings. The tendency of decreasing marriage stability then successively spread to the middle and lower classes as the divorce rate continued to increase during the course of the twentieth century.

Keywords:

• Divorce
• Sweden
• Early twentieth century
• Social stratification
• Female labor force participation

Acknowledgments

I particularly wish to express my gratitude to Professor Gunnar Andersson for organizing the session on divorce held at Nordic Demographic Symposium in Lund, June 2010 as well as to the session-participants for their valuable comments on an earlier draft of this paper.

Notes

1 Principles for the classification of urban and rural municipalities are described in Statistics Sweden (1923, pp.3–26).

2 Optimally we would have representative samples of individual level data to investigate these issues, as the variability on the individual level is reduced when one is forced to use group level means, which can result in aggregation bias (for a discussion see e.g. Firebaugh, 1978; Susser, 1994). Analysis of relationships on the aggregate level is, however, common in social sciences as micro-level data can be hard or even impossible to obtain. The rationale is that a failure to falsify a hypothesis on the aggregate level makes it considerably more probable that there indeed is also a connection on the level of the individual and that group level effects are of interest in their own right as they can uncover contextual influences (Flanigan & Zingale, 1985; Freedman, 2004; Susser, 1994).

3 Children and female employment have a VIF of approximately 1.5 and the other variables are close to 1. Effects on standard errors due to collinearity are thus small. A VIF equaling 4 reflects a doubling of standard error due to collinearity, which is stated as a cautious cutoff level in the literature (e.g. Miles & Shevlin, 2001, p.130).

4 The original dataset included 130 cases. One case (lawyers in urban municipalities in 1920) was excluded from the analysis. This was motivated by the fact that this group had a very high mean income as compared to the other cases. Including this case did not, however, make the estimates unstable, and removing it had very small effects on the estimates. Despite this lack of influence, the case was removed from the analysis as the mean income was almost twice the second highest income value in the sample, which made it an obvious outlier in the distribution of income values in the sample.