Inflation and other aggregate determinants of the trend in US divorce rates since the 1960s

Abstract

This article extends empirical research on the determinants of divorce in two ways. First, I examine the effect of inflation on divorce. Second, the use of a structural time-series modelling approach attributes unobservables and omitted variables to an unobserved component, which allows for the model's parameters to be estimated consistently. Inflation is statistically significant, positive and persistent. I show that the effects of inflation are robust to the inclusion of additional explanatory variables and various trend specifications. The long-run implications of inflation are also substantial. I conclude that price stability has the potential to reduce divorce rates.

Acknowledgements

I thank C. L. Baum, E. A. Eff, G. Givens, A. W. Hogan, T. Minor, P. Morris, M. F. Owens, A. Seals, J. Zietz and an anonymous referee for their assistance and helpful comments.

Notes

¹ A few of these studies examine the increase and levelling of divorce rates, which refers to the stock of divorces, not the number of new divorces. This article focuses on new divorces.

² South (1985) and Bremmer and Kesselring (1999, 2004) are exceptions.

³ I use the terms structural time-series and unobserved component models interchangeably throughout this article. The structural time-series methodology has been used to analyse a variety of different economic relationships (e.g. Abeysinghe, 2000; Muscatelli and Tirelli, 2001; Scuffham, 2003; Hon and Yong, 2004; Dimitropoulos et al., 2005; Mazzocchi et al., 2006; Adhikari et al., 2007).

⁴ The returns associated with marriage are usually attributed to the couple's ability to specialize in market and household work. For example, increases in consumption, leisure and the production of one's own children have been cited as determinants of marriage.

⁵ See Becker et al. (1977).

⁶ The results in the majority of these studies support the theory and findings of Becker et al. (1977). Charles and Stevens (2004) find that job displacement, measured as layoffs, increases the risk of divorce. However, they find that disability and plant closings have no effect on divorce. Their results cast doubt on pecuniary motives of divorce, since disability, plant closings and layoffs have similar long-run consequences.

⁷ Marriage-specific capital could be the production of children, investments in joint assets and investing in additions to human capital for spouses.

⁸ In fact, a simple Ordinary Least Square (OLS) regression of the growth rate of GDP on the unemployment rate, and vice versa, yields a negative, contemporaneous relationship between the two variables. As a result, South's estimates of the change (or percentage change) in GNP could be downwardly biased, which could be the reason for the negative effect found with respect to changes in GNP.

⁹ See the lower portion of Fig. 3.

¹⁰ Bremmer and Kesselring (1999, 2004) are the only studies in this list to examine the aggregate relationship between divorce rates and female-labour force participation. Bremmer and Kesselring (1999) attempt to determine the causal direction of the two variables and find that the divorce rate ‘Granger’ causes female labour-force participation. Bremmer and Kesselring (2004) examine the long-run relationship between divorce rates and female labour-force participation. Their results suggest that rising divorce rates increases female labour-force participation and that rising female-labour force participation increases divorce rates.

¹¹ A related issue is the role of female earnings in divorce decisions. Ressler and Waters (2000) and Kesselring and Bremmer (2006) examine the relationship between female earnings and divorce. Ressler and Waters (2000) find evidence that the two variables are jointly determined. Kesselring and Bremmer (2006) find evidence confirming the results found by previous research; that is, the rising economic power of women increases the risk of divorce. After comparing these findings, it is difficult to determine the causal relationship between the two variables.

¹² Since unit root tests rely on autoregressive models, Harvey (1997) contends that such tests may exhibit poor statistical properties. In fact, Harvey and Jaegar (1993) show with simulations that unit root tests do not typically detect variables that are I(2). Detecting a unit root process usually results in the researcher concluding that the series is I(1).

¹³ See Harvey (1989) for a detailed description of the Kalman filter and its application in structural time-series models. The statistical package used–Structural Time-Series Analyser, Modeller and Predictor (STAMP)–offers a canned procedure for the Kalman Filter.

¹⁴ If the variance of the disturbance term η_t equals zero and the variance of the disturbance term ξ_t is nonzero, the model takes the smooth-trend specification, which is integrated of order two (Harvey, 1997).

¹⁵ I adopt the empirical methodology advocated by the London School of Economics (LSE). Each set of parameter restrictions are validated by checking the statistical properties of the model. The LSE approach assumes that all models are false. The goal of the LSE approach is to find an adequate model; one that captures the data generating process.