What has been the result of an increasing liberalization in divorce laws in the United States?

Chile and Divorce Research

The results are presented in the following sections. Section 5.1 shows the estimations for the static specifications, while Sect. 5.2 presents the outcomes when the model is enhanced to properly capture the dynamic response of divorce rates. In Sect. 5.3, control variables are added to the static and dynamic models to account for observed heterogeneity, and in Sect. 5.4, alternative empirical approaches are followed to determine if the main conclusions continue to be valid.

Static specifications

Table 2 reports estimates of the static effects on divorce rates when unilateral legislation is adopted. The estimates suggest that unilateral divorce raises divorce rates in Mexico. All coefficients of unilateral are statistically significant. The first specification in column (1) does not include fixed effects, and it is observed that its coefficient for unilateral is the largest. It captures not only the effect of the modification in the divorce legislation but also other changes on divorce patterns over time and across states. To improve the model, controlling for the average differences in states and years, specification (2) includes year and state effects. The coefficient indicates that the adoption of unilateral divorce raises the divorce rate by 0.32 annual divorces per thousand people. While the year effects capture evolving unobserved characteristics at a country level, and the state effects control for constant factors over time that influence divorce decisions; specifications (3) and (4) represent more flexible models where attributes that affect divorce propensities in each state are allowed to change over time. The results exhibit a smaller effect of no-fault divorce when linear and quadratic state trends are included.

The F statistics for the state trends in columns (3) and (4) show that the significance level of the test equals zero, reflecting that state trends are jointly significant, both linear and quadratic. In addition, moving across the columns, the adjusted R2 increases from 0.89 in specification (2) to 0.95 in specification (4), supporting the inclusion of state trends as relevant to the model. A possible explanation for the modest variation in the unilateral coefficient when state trends are added, compared to other countries such as the USA, might be the homogenous gender inequality that is predominant in all Mexican states to this day. Women’s decision-making power within the household is limited in the country, and therefore only an external shock such as an unexpected change in the divorce legislation triggers a structural change in the marriage market, disrupting traditional gender roles and stereotypes. It may also be the case that the main factors that have an impact on divorce rates within states have not changed much over the period analyzed. In Sect. 5.3, the results are presented when some of these potential factors are explicitly included in the estimations. As an only exception, in specification (4), the F test for the year effects fails to reject the null hypothesis that the coefficients for all years are jointly equal to zero, suggesting that there is no need to include year-fixed effects in the model. Table 7 in Appendix provides the estimations for all specifications excluding year effects. The impact of unilateral legislation on divorce rates remains positive, significant and similar in magnitude.

Considering that Friedberg (1998) used specifications similar to those in Table 2 for the USA and obtained a variation between 0.004 and 1.80 in annual divorces per thousand people due to unilateral legislation, it can be argued that in the case of Mexico, regardless of the model used, the static effects of unilateral legislation do not vary much across specifications, from 0.23 to 0.39. This suggests that the model is appropriate for the country and that there is a strong and steady relationship between changes in divorce law and divorce rates in Mexico. The unilateral coefficient in specification (3), for instance, represents 34.9% of the average divorce rate of 0.85 annual divorces per 1000 population for the period analyzed. Moreover, the adoption of unilateral legislation has increased the divorce rate by 26.4% in the shifting states during the period 2009–2015.

An issue for the robustness of the results presented above is the number of years considered in the analysis before the policy shock, to properly account for preexisting state trends. This is less of a problem for those states that have shifted to unilateral divorce more recently but remains a controversy for those that started earlier, such as Mexico City (2008) or Hidalgo (2011). Table 3 reports the static effects on divorce rates for the period 2001 to 2015. Comparing Tables 2 and 3, it is observed that the inclusion of additional years pre-reform plays no major role in the analysis. Estimations for a larger period, from 1993 to 2015, are also provided in Table 8 in Appendix, and the findings remain unchanged. It is to be noted that adding data where all states are untreated (1993 to 2004) tends to increase the unilateral coefficient. For instance, in Table 2, specification (4) indicates that no-fault legislation raises divorce rates by 0.23 annual divorces per thousand people, whereas in Table 3, specification (4) shows an increase in 0.29 annual divorces per thousand people. Contrary to what is observed, it is expected that adding data where all states are untreated would reduce the coefficient. This finding might reflect the almost null variation in divorce rates during the pre-reform years at the national level, reinforcing the effect of the change in the divorce legislation rather than diluting it when the data are extended back. According to data from INEGI, in 1990 and 2000, there were seven divorces for every 100 new marriages. By 2005, this rate rose to 11.8, and in 2015, it reached 22 per 100 new marriages (see Fig. 2 in Appendix).

(Source: National Institute of Statistics and Geography (INEGI))

Divorces per 100 new marriages in Mexico

The aim of this section is to examine the potential bias resulting from unmeasured confounders. As mentioned earlier, outcomes from Eq. (1) might be biased measures of the causal effect of unilateral divorce on divorce rates because the unilateral coefficient is not allowed to change after the adoption of no-fault divorce, confounding preexisting trends with the dynamic effects of the policy shock. When a policy shock takes place, depending on the circumstances, the impact may be immediate or occur with considerable delay; it either has a permanent effect or dies out at a relatively fast pace. Wolfers (2006) analyzes the short-, medium- and long-run effects of the adoption of unilateral law in the USA. In the case of Mexico, the shift toward no-fault divorce is a recently enacted legislation, starting in 2008, so the analysis is focused on the short and medium term. Table 4 presents the effects that unilateral legislation has on divorce rates within the first 2 years of the change in the law, during years 3 and 4 and after 5 or more years. All unilateral coefficients are statistically significant, with the exception of column (4) after 5 years or more. State trends are jointly significant, and the adjusted R2 increases from specification (1) to (4).

According to estimates in columns (2) to (4), the introduction of unilateral reforms increases divorce rates in the short run from 0.21 to 0.28 annual divorces per thousand people. Over years 3 and 4, the effect increases in size for specifications (2) and (3) and remains very similar for specification (4). Finally, 5 or more years after the reform, the impact is still positive but starts to diminish, affecting divorces rates by 0.29 and 0.25 annual divorces per 1000 people, according to specifications (2) and (3), respectively. Tests have been performed on the equality of the three coefficients of unilateral in each specification, rejecting the hypothesis that they are similar for specifications (3) and (4) at standard confidence levels, supporting the strategy followed in this section. A potential explanation of the higher effect of the change in law in years 3 and 4, rather than during the first 2 years, is that initially, the changes in the divorce regime are not widely known by the population, taking time for the information to be disseminated. Time is also necessary for divorce to become more acceptable, and people gradually become more open to ending a marriage that no longer works as more couples get separated. In addition, the process of filing for divorce under different rules can be difficult to understand at the beginning, delaying the decision. The positive but smaller size of the effect on divorce rates of no-fault divorce after 5 or more years indicates that although the dynamic response to the policy shock persists in the medium term, the effect of the law change over the following years might gradually be reduced as an adjustment to a temporary boom of inefficient marriages breaking up immediately after the reform. It is important to highlight that comparing the static and dynamic estimates for unilateral in Tables 2 and 4, the coefficients do not vary much and remain very similar, confirming a close relationship between changes in divorce legislation and divorce rates, regardless of the approach that is followed.

Control variables

To explicitly account for observed heterogeneity, five variables are included in the analysis: education,12 female labor force participation, fertility rates,13 gross domestic product (GDP) and unemployment. The inclusion of these controls aims to reassess the impact of unilateral legislation on divorce rates when some state-level variables are added to the model. The results for the static and dynamic specifications, reported in Tables 9 and 10 in Appendix, are virtually identical to those presented in Sects. 5.1 and 5.2 for the effect of divorce legislation, validating the inclusion of state-fixed effects and trends in the analysis in order to capture the effect of other factors that affect divorce rates.

In terms of the new variables added to the model, only unemployment turned out to be significant in most specifications. However, contrary to what the literature suggests (Becker et al. 1977), an increase in unemployment leads to an unexpected reduction in divorce rates in Mexico. An explanation for this is that divorce itself costs money, so the inability to afford a divorce for individuals facing unemployment, and the fact that it costs more for a couple to live separately than together, may be preventing married couples in developing countries from filing for divorce when unemployment rates are higher. Another possible explanation is that marriage might be seen as some sort of informal insurance against unemployment, becoming more valuable when unemployment is high.

Unweighted specifications and changes in the functional form

All of the previous estimations have been performed using weighted least squares (WLS) to correct for the presence of heteroscedasticity generated by the use of state-level divorce rates rather than individual data on divorce decisions. However, it has been argued that estimations under WLS and ordinary least squares (OLS) should be similar if the unobserved heterogeneity is adequately addressed (Kim and Oka 2014). Following Droes and Lamoen (2010), the transformed model using analytical weights is:

$$\begin{aligned} {\text{Divorce Rate}}_{s,t} \sqrt {{\text{pop}}_{s,t} } = & \beta {\text{Unilateral}}_{s,t} \sqrt {{\text{pop}}_{s,t} } \\ & + \mathop \sum \limits_{s} {\text{State fixed effects}}_{s} \sqrt {{\text{pop}}_{s,t} } + \mathop \sum \limits_{y} {\text{Year fixed effects}}_{y} \sqrt {{\text{pop}}_{s,t} } \\ & + \mathop \sum \limits_{s} {\text{State}}_{s} *{\text{Time}}_{t} \sqrt {{\text{pop}}_{s,t} } + \mathop \sum \limits_{s} {\text{State}}_{s} *{\text{Time}}_{t}^{2} \sqrt {{\text{pop}}_{s,t} } + \varepsilon_{s,t} \sqrt {{\text{pop}}_{s,t} } \\ \end{aligned}$$

(3)

where \({\text{pop}}\) is the state population in thousands. It is observed that the coefficient for unilateral divorce remains equal after the transformation. Lee and Solon (2011), and Droes and Lamoen (2010), using Wolfers (2006) and Friedberg’s (1998) data, estimate the effect of unilateral divorce using OLS. In addition, Lee and Solon (2011) perform estimations using the logarithm of the divorce rate, claiming that this is also a valid functional specification. The results for the USA suggest that the change in law has no effect on divorce rates, neither when OLS regressions are estimated nor when the dependent variable in the analysis is the divorce rate in log, casting doubt on the true effect of unilateral legislation in that country.

Weighting by population to correct for heteroscedasticity in order to obtain efficient estimators relies on the strong assumption of homoscedastic and independent error terms for individuals within the state. However, if individual error terms share a common state-level error component, the unweighted state-average error terms are closely homoscedastic. In this scenario, the use of WLS would exacerbate any existing heteroscedasticity, and OLS estimation would be more efficient than WLS. Large discrepancies between the results obtained using WLS and OLS might be an indication of functional form or model misspecification. Therefore, estimations based on OLS without weighting are also important to perform and report. Likewise, given the nature of the dependent variable used within this context, an always positive divorce rate, it is possible to consider different functional specifications, such as the logarithm of divorce rates. Typically, the results based on changes in functional form assumptions are expected not to be extremely sensitive to these modifications, supporting previous findings and providing compelling evidence for the main conclusions in the analysis.

To determine if the results obtained for Mexico are still valid following these approaches, Tables 11 and 12 in Appendix report the OLS estimates, and Tables 13, 14, 15 and 16 present the estimations when using the log of the divorce rate. As discussed by Lee and Solon (2011), the OLS coefficients obtained are smaller than the WLS estimates, given that WLS places more weight on those states that are more populated, and given that unilateral divorce has larger effects on these states. However, in contrast to the results for the USA, the coefficients obtained for unilateral legislation continue to be positive and statistically significant in practically all specifications. These findings provide compelling evidence that unilateral divorce has an effect on the divorce rates in Mexico, regardless of the estimation methods or the functional form assumed.

Page 2

1. Source: Author’s elaboration based on the standing legislation in each state
2. Legislation of the remaining 14 states not included in this table was also verified (Baja California, Campeche, Chiapas, Chihuahua, Durango, Guanajuato, Jalisco, Oaxaca, Queretaro, San Luis Potosi, Sonora, Tabasco, Veracruz and Zacatecas). Unilateral divorce is not valid in any of them
3. Last updated: January 2017