Social Security and Divorce

Marcus Dillender 1
  • 1 W.E. Upjohn Institute for Employment Research, 300 S. Westnedge Ave., Kalamazoo, MI 49007-4686, USA
Marcus Dillender

Abstract

This paper studies how the likelihood and timing of divorce are influenced by Social Security’s 10-year rule, which provides spousal benefits to divorced people if their marriages lasted at least 10 years. Bunching analysis indicates that approximately 2 % of divorces occurring in the 6 months after 10-year anniversaries would have occurred earlier if not for Social Security’s 10-year rule. For older couples, who are likely more focused on retirement and have greater earning disparities, divorces are approximately 9 % higher in the 2 years after 10-year anniversaries than would be predicted without the abrupt change in Social Security benefits. The increase in divorces after 10 years of marriage appears to come from couples with disparate earning records.

1 Introduction

When and why people divorce matters because divorce has both emotional and economic implications. 1 Divorcing couples are happier after the divorce than before it, which means that delaying divorce can have psychic costs (Gardner and Oswald 2006). An individual’s potential financial well-being after divorce is likely a factor in the decision to leave a marriage, but understanding how financial well-being outside of marriage affects divorce is complicated by a couple’s financial situation being related to many unobserved factors that also affect divorce probabilities. This paper examines how Social Security’s 10-year rule, which entitles divorced individuals to Social Security spousal benefits if their marriages lasted at least 10 years, affects divorce timing and likelihood. This arbitrary rule creates a sharp increase in the value of exiting a marriage at 10-year anniversaries for secondary earners relative to the value of exiting the marriage at 9 years and provides an opportunity to understand how financial factors affect divorce.

In economic models of marriage, people choose to be married when the value of being married exceeds the value of being single. Therefore, raising the value of being single for married people should theoretically increase divorces, and people who would benefit from Social Security’s 10-year rule should have an incentive to delay divorce until after 10 years of marriage. Despite a theoretical basis for Social Security’s 10-year rule affecting divorce, Dickert-Conlin and Meghea (2004) find its implementation in 1977 had little immediate impact on divorce timing using a difference-in-differences strategy with the length of marriage in years from Vital Statistics data. Goda, Shoven, and Slavov (2007) point out that the 10-year rule should have a larger influence on couples with disparate earnings’ histories. Using data from the Panel Study of Income Dynamics (PSID), they find small, statistically insignificant effects of the 10-year rule on vulnerable couples.

Other research on the link between financial incentives and divorce has found mixed results. Alm and Whittington (1997) use PSID data to study how income tax penalties affect marriage and divorce decisions. They find evidence that marriage decisions respond to tax penalties while divorce decisions do not. Bitler et al. (2004), on the other hand, find divorce propensities fall after the passage of welfare reforms that increase the value of being married relative to being single. Thus, the current state of the literature is inconclusive about whether or not financial incentives influence divorce.

The current paper studies how the 10-year rule affects divorce using 1985–1995 Vital Statistics divorce data with the length of marriage in months. If Social Security’s 10-year rule affects divorce through either a decrease in divorces before the 10-year mark or an increase after it, divorces should discontinuously increase immediately after 10-year anniversaries. Plotting divorces by the duration of marriage in months and estimating the discontinuity in the divorce rate at 10 years of marriage show clear evidence of a distortion in the distribution of divorces around 10-year anniversaries. I then implement bunching analysis to quantify how many divorces are delayed and for how long. The basic approach involves using how divorces trend with marriage duration away from 10-year anniversaries to estimate how they would trend near them if not for the benefit change occurring immediately at the 10th year of marriages. With the counterfactual distribution estimated, I then calculate the divorces missing from the distribution before 10-year anniversaries as well as the extra ones after them.

The bunching analysis indicates that about 2 % of divorces occurring in 6 months after 10-year anniversaries would have occurred before them if not for Social Security’s 10-year rule. Responses to the 10-year rule vary dramatically by age. For couples with the woman under the age of 25 at the time of marriage, I find only weak evidence of a small effect of the 10-year rule on divorce rates. For couples where the woman was 45 or older at the time of marriage, I find that there are 9 % more divorces in the 2 years after 10-year anniversaries than the estimated distribution predicts.

Since the Vital Statistics collection program ended in 1995 and because having individual-level economic data is necessary to examine characteristics of couples who divorce, I also draw on data from the 2008 to 2011 American Community Survey (ACS). People delaying divorce until their 10th anniversaries would cause the likelihood of being divorced to increase discontinuously after 10 years of marriage. For people who married at older ages, the likelihood of being divorced gaps up at 10-year anniversaries by 19.4 %. The marriages that end are those where one spouse worked in an occupation that earned at least 50 % more than the other spouse’s occupation and those with spouses with unequal education levels. Thus, it appears that Social Security’s 10-year rule affects couples where one member has a higher earnings potential than the other.

These results provide strong evidence that Social Security and financial considerations factor into divorce decisions, especially for older Americans. The current paper extends the previous work on Social Security’s 10-year rule in several ways. First, the paper focuses on data several years after the implementation of the 10-year rule, meaning people would be more likely to know about the 10-year rule and how to take advantage of it. 2 Second, the paper uses Vital Statistics data with the length of marriages in months. When examining a flow measure like divorce rates, knowing the length of marriage in months is crucial as it allows for examining divorces within a close range of 10-year anniversaries. Similarly, having the duration of marriages in months allows for examining whether or not divorces bunch around 10-year anniversaries. Finally, the large data sets used in the analysis allow for exploring heterogeneous effects of the 10-year rule based on age at the time of marriage. Understanding heterogeneity by age is important as changes in family structure in old age have become increasingly common. Stevenson and Wolfers (2007) document the recent rise in marital formation of older Americans, while Brown and Lin (2012) study the dramatic increase in divorces for older Americans, which they term the “gray divorce revolution.”

2 Background

2.1 Social Security’s 10-Year Rule

People contribute to Social Security through payroll taxes, and employers match the employee contribution. Upon retiring, workers can receive Social Security benefits if they accumulated at least 40 quarters of earnings over their work lives. The size of the benefit, or the Primary Insurance Amount (PIA), is computed based on the average of the worker’s highest 35 years of indexed monthly earnings. 3

If their ex-spouses are still alive, former spouses are eligible for spousal benefits of 50 % of the primary earners’ PIA if their marriages lasted at least 10 years before ending in divorce. If their ex-spouses are deceased, former spouses are eligible for spousal benefits equal to the primary earners’ full PIA if their marriages lasted at least 10 years before ending in divorce. 4 Even former spouses who qualify for Social Security on their own earnings’ histories can still receive spousal benefits if the spousal benefits are greater than what they would receive based on their own earnings. Divorced people whose marriages lasted fewer than 10 years are not eligible for any spousal benefits.

The Social Security Administration (SSA) defines eligibility based on formal marriage length. A separation delayed until 10 years of marriage would still be considered intact. Remarrying results in the individual no longer being eligible for spousal benefits from a previous marriage; however, if a subsequent marriage ends in divorce, the person can be eligible for spousal benefits from any previous marriages that lasted at least 10 years. An ex-spouse remarrying does not affect an individual’s eligibility (Social Security Administration 2013). 5

The 10-year rule was part of a 1977 Social Security law and went into effect in 1979. While the main purpose of the 1977 law was to ensure the financial stability of Social Security, it also changed the length of marriage requirement for spousal benefits from 20 to 10 years because marriages were ending more quickly than before (Dickert-Conlin and Meghea 2004).

The vast majority of spousal benefits go to women since they tend to have lower PIAs. In 2006, approximately 8 % of people receiving Social Security received it through the spousal benefit, and approximately 98 % of people receiving spousal benefits were women. The ex-husband’s full PIA is much more likely to be larger than the woman’s PIA than half of the ex-husband’s PIA is. For this reason, a majority of divorced wives will receive benefits based on their deceased ex-husbands’ PIA if their ex-husbands die (Butrica and Smith 2012). As more women enter the labor force and earn higher wages, more women are receiving Social Security without the spousal benefit (Social Security Administration 2013 and Goda, Shoven, and Slavov 2007).

2.2 Conceptual Framework – Heterogeneity by Age at the Time of Marriage

The impact of the 10-year rule likely varies with age. As retirement is nearer for older people, they are likely more focused on Social Security benefits. Young people, on the other hand, tend to be myopic in thinking about retirement. Young people also likely do not know the value of the spousal benefit as earnings typically peak later in life, whereas older people generally have a better idea about whether or not spousal benefits would increase their Social Security payments. 6

Even if young people are perfectly rational and forward thinking, they still may not be influenced by Social Security’s 10-year rule because they have time to marry again and to achieve spousal benefits through another spouse. Since married people are no longer eligible for spousal benefits from previous marriages, young people would have to go through most of their adult lives unmarried or have subsequent marriages end in divorce to claim spousal benefits from divorces that occurred in their twenties or early thirties. Approximately 69 % of women and 78 % of men remarry after divorce (Schoen and Standish 2001), and young divorced people are much more likely to remarry than older divorced people (Brown, Lee, and Bulanda 2006), suggesting young divorced people likely expect to remarry. Not remarrying is a smaller price to pay for older adults. Older couples are also more likely to respond to the 10-year rule because a higher percentage of couples from older cohorts have disparate earnings’ records. Younger generations of women have more parity with their husbands and would be less likely to benefit from the 10-year rule.

Throughout the remainder of the paper, I consider how the 10-year rule influences divorce for the full sample as well as for three broad age groups. For the couple-level Vital Statistics data, the groups are couples with women younger than 25 at marriage, couples with women 25–44 at marriage, and couples with women 45 or older at marriage. With the individual-level ACS data, the groups are women younger than 25 at the start of marriage, women ages 25–44 at the start of marriage, and women 45 or older at the start of marriage. I focus on the age of the woman since women are more likely to receive spousal benefits. To ensure that these broad age groups are appropriate, I also consider smaller age bins, which produce noisier results but provide a fuller picture of how the response to the 10-year rule varies by age.

3 Bunching in Divorces around 10-Year Anniversaries

3.1 Data

The first set of results uses Vital Statistics data from 1985 to 1995. The Vital Statistics data were compiled by the National Center for Health Statistics and contain information from divorce certificates collected at the state level. About half of all states participated in the program. While some states provided a random sample to the National Center for Health Statistics, other states provided data on all divorce certificates. These data are not nationally representative as the non-reporting states come disproportionately from the South and Mountain West. 7 I restrict the sample to couples who divorced within 4 years of their 10-year anniversaries.

The Vital Statistics data have several advantages. First, they contain information on the month and year of marriage and divorce, meaning I can calculate the duration of marriage in months. Second, the data set is large. For the years 1985–1995, the data contain information on 2,008,923 divorces. Of these, 1,818,591 contain the ages of the spouses and the information necessary to compute the duration of marriage in months. Finally, these data come straight from divorce certificates and are likely very accurate.

3.2 Empirical Strategy

3.2.1 Discontinuity at 10 Years since Marriage

If Social Security’s 10-year rule leads to couples delaying divorce until their 10-year anniversaries, then the divorce rate would discontinuously increase at 10 years since marriages began. Thus, as a simple test of whether or not the 10-year rule influences divorce rates, I begin by examining whether or not a discontinuity exists at 10 years since marriage by estimating the following equation:

ym=α+fm,λ+Dmβ+ηm,
where m indexes marriage duration in months, y is the number of divorces happening at a given duration, f is a smooth function representing the duration profile of divorces, D is an indicator variable equal to 1 if the divorce occurs after at least 10 years of marriage, and ƞ is an unobserved error component. I estimate eq. [1] by modeling as a quadratic polynomial on either side of the 10-year threshold. In addition to estimating eq. [1] with the number of divorces as the dependent variable, I also include the log of the number of marriages at a given duration as the dependent variable, which will allow the β coefficients to be interpreted as estimates of the percentage discontinuities in divorce rates.

3.2.2 Bunching

Testing for a discontinuity at 10 years of marriage allows for establishing whether or not Social Security’s 10-year rule affects divorce rates, but it does not allow for examining whether or not the increase in divorces at 10-year anniversaries are retimed or to examine the length of any retiming. Thus, I also employ bunching analysis.

To examine bunching in divorces around the 10-year marks of marriages, I exclude data from around the 10-year cutoff and estimate how divorces would trend with months since marriage in the absence of the abrupt change in incentives at 10-year anniversaries. I then use this counterfactual distribution to examine bunching behavior around 10-year anniversaries. Persson (2013) uses a similar strategy to study marriage timing in response to changing survivor’s insurance in Sweden. 8

To implement this approach, I first estimate the following equation:

ym=α+gm,λ+ηm,
where m indexes the duration of marriage in months, y is the number of divorces happening at a given duration, g is a polynomial in the marriage duration, and ƞ is an unobserved error component. I estimate eq. [2] excluding divorces around 10-year anniversaries. I then use the parameter estimates to compute yˆbefore and yˆafter, estimates of the number of divorces that would have occurred in the omitted bunching region without a change in divorce incentives occurring at the 10-year marks of marriages. I next calculate the bunching estimates Bˆbefore and Bˆafter as the difference between the number of divorces predicted from the counterfactual distribution and the actual number of divorces occurring in the bunching regions, where Bˆbefore estimates the missing mass of divorces for couples before 10-year anniversaries and Bˆafter estimates the bunching that occurs immediately after 10-year anniversaries.

This framework requires two main assumptions. The first is that g would trend smoothly if not for Social Security’s 10-year rule. A possible concern with this assumption is that there may have always existed something related to 10 years of marriage that causes marriages to end that is completely unrelated to Social Security’s 10-year rule, which could be the case since many factors enter into divorce decisions that cannot be observed in administrative data. For example, a gap in divorces at 10-year anniversaries would exist if couples wanted to hold out until the 10-year milestone before divorcing for psychological reasons or if there was another unobserved change happening at 10 years of marriage. As a test of the assumption that g would trend smoothly if not for the 10-year rule, I replicate the Vital Statistics analysis using data from before the 10-year rule’s implementation in Appendix A. I find no evidence of any trend break at 10 years of marriage before the 10-year rule was implemented, which suggests that Social Security’s 10-year rule is indeed responsible for the altered divorce distribution. 9

The second assumption is that data from outside the bunching region can be used to approximate g within the bunching region. This assumption would be violated if people delayed divorce from very early on in marriages to benefit from the 10-year rule. This assumption relates to the choice of bunching regions, which is not immediately clear. The bunching region needs to be wide enough to exclude divorces affected by the 10-year rule; however, making the bunching region too wide can result in the loss of precision and results in eq. [2] being used to estimate ŷ’s far out of sample. Therefore, I report results for a range of bunching regions. I begin by setting the bunching region to be 6 months before and 6 months after 10-year anniversaries. This bunching region allows for a focused examination within a close range of 10-year anniversaries but assumes that couples do not delay divorce for more than 6 months and that divorces delayed because of the 10-year rule happen quickly after 10-year anniversaries. To consider the possibility of longer delays and to allow couples more time to divorce after reaching their 10-year anniversaries, I also show results that set the bunching regions to be 1, 1.5, and 2 years before and 2 years after 10-year anniversaries. For the main analysis, g is fitted as a cubic polynomial. In Appendix B, I consider the sensitivity of the results to specifying g as different polynomials.

Standard errors for the bunching estimates are estimated by bootstrapping. To do this, I draw a random sample with replacement from the original sample that is equal in size to the original sample. I then replicate the procedure described above 1,000 times to produce a distribution of Bˆbefore and Bˆafter. The standard errors for the bunching estimates are then calculated as the standard deviation of the distribution of bootstrapped estimates.

3.3 Results

Figure 1 shows the number of divorces by marriage length for all ages and for different age groups. For older couples in particular, divorces gap up immediately after 10-year anniversaries. For couples in the middle age group, divorces also appear to increase slightly at 10-year anniversaries. For couples where the woman was 25 or younger at the time of marriage, divorces appear to trend more smoothly.

Figure 1:
Figure 1:

Divorces by marriage length, from the 1985 to 1995 Vital Statistics data.

Citation: The B.E. Journal of Economic Analysis & Policy 16, 2; 10.1515/bejeap-2015-0168

The β coefficients from eq. [1] are shown in Table 1 and reveal a statistically significant increase in the divorce rate for people 25 and older at the time of marriage as well as for the full sample. For couples where the woman was at least 45 or older at marriage, divorces increase by 23 % at 10-year anniversaries. For middle-aged couples, divorces increase by over 4 % immediately after 10-year anniversaries. For younger couples, there is only weak evidence of an increase in divorces after 10-year anniversaries. For the full sample, divorces increase by 3.3 %. 10

Table 1:

Discontinuity in divorces at 10 years of marriage.

Divorces at 9 yearsDiscontinuity at 10 years
Number of divorcesLog of number of divorces
All couples73,575166***0.033***
(56)(0.009)
Married younger than 2546,188600.020*
(44)(0.011)
Married from 25 to 4425,77579***0.043***
(30)(0.014)
Married older than 441,61227***0.230***
(8)(0.064)

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. Robust standard errors are in parentheses below the estimates. The data are from the 1985 to 1995 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. All regressions control for quadratics in length of marriage in months on either side of 10-year anniversaries of marriages. The discontinuity in divorces at 10 years is per month.

These results provide evidence that age is a major factor in how divorce decisions respond to the 10-year rule. To further explore how the response to the 10-year rule varies with age, I replicate the analysis using 5-year age bins in Table 2. With one exception, the point estimates rise with each age bin, though they are generally not statistically significantly different from each other. These results reveal a nonlinear response to the 10-year rule. For women in the three older age bins, the estimated discontinuities are statistically significant and large. For women who married at 56–60, the estimated increase in divorces is over 60 %. 11

Table 2:

Discontinuity in divorces at 10 years of marriage narrower age bins.

Divorces at 9 yearsDiscontinuity at 10 years
Number of divorcesLog of number of divorces
Married from 16 to 2024,780300.021
(27)(0.013)
Married from 21 to 2524,094410.022
(32)(0.016)
Married from 26 to 3012,33546**0.047**
(19)(0.018)
Married from 31 to 355,895170.048
(13)(0.030)
Married from 36 to 402,823110.068
(11)(0.046)
Married from 41 to 451,334−20.011
(7)(0.069)
Married from 46 to 5070412**0.232**
(5)(0.096)
Married from 51 to 5534567**0.263**
(3)(0.122)
Married from 56 to 601578***0.613***
(2)(0.168)

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. Robust standard errors are in parentheses below the estimates. The data are from the 1985 to 1995 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. All regressions control for quadratics in length of marriage in months on either side of 10-year anniversaries of marriages. The discontinuity in divorces at 10 years is per month.

I next implement the bunching procedure described in Section 3.2 to produce estimates of the missing mass of divorces during the 6 months before 10-year anniversaries and the bunching that takes place during the 6 months after 10-year anniversaries. The results are shown in the top panel of Table 3 and suggest that approximately 2 % of divorces occurring in 6 months after 10-year anniversaries are delayed from the 6 months before. For couples with women who were 45 or older at marriage, approximately 11.8 % of divorces occurring during the 6 months after 10-year anniversaries appear to be retimed. For couples with women who married younger, 1.8 % of divorces occurring in the 6 months after divorce are retimed.

Table 3:

Bunching of divorces around 10 years of marriage.

Panel AEstimated divorces from 9.5 to 10 years without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.Estimated divorces from 10 to 10.5 years without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.
All couples36,156−763212−2.ll***34,0296761981.99***
Married younger than 2522,762−337165−1.48**21,7053841621.77***
Married from 25 to 4412,587−367124−2.91***11,5872051151.77*
Married older than 44807−5931−7.28*737873111.76***
Panel BEstimated divorces at year 9 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.Estimated divorces at year 10 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.
All couples75,050−1,475364−1.96***66,528−77326−0.12
Married younger than 2546,978−790282−1.68***42,759−254261−0.59
Married from 25 to 4426,384−609213−2.31***22,367391810.17
Married older than 441,688−7653−4.481,402138519.86***
Panel CEstimated divorces at years 8.5 to 9 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.Estimated divorces at years 10 and 11.5 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.
All couples115,803−782499−0.6834,0396662061.96***
Married younger than 2571,867−168396−0.2321,6864031701.86***
Married from 25 to 4441,225−478295−1.1611,6051871201.61
Married older than 442,711−13676−5.01*748763210.13***
Panel DEstimated divorces at years 8 and 9 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.Estimated divorces at years 10 and 11 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.
All couples159,457−182874−0.11125,2278517460.68
Married younger than 2598,497586930.0681,6631846180.23
Married from 25 to 4457,228−161523−0.2841,0404374161.07
Married Older than 443,732−79139−2.132,5242301079.10**

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The data are from the 1985 to 1995 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. The estimates of divorces that would have occurred without the 10-year rule come from fitting a polynomial excluding data from the bunching region and then using the parameter estimates to estimate the divorces that would have occurred in the bunching region. Standard errors are calculated from the bootstrapping procedure described in the paper.

In panels B and C, I set the bunching regions to be 1 and 1.5 years, respectively. The results from both bunching regions suggest that there are extra divorces after 10-year anniversaries for couples with women who were 45 or older at marriage. For couples with women who married younger, the results from panels B and C are contradictory. In panel B, the results imply that there is a missing mass of divorces immediately prior to 10-year anniversaries for couples with women who were younger than 45 at marriage. In panel C, the results imply that there are extra divorces after 10-year anniversaries for couples with women who were younger than 45 at marriage. The results from the bunching analysis for younger women being sensitive to the bunching window likely suggests that g does a poor job predicting what would happen within the bunching region using data far from the cutoff for younger women. Thus, the results for the larger bunching windows should be interpreted with caution for younger couples.

Panel D displays results with the bunching region set to be 2 years before and after 10-year anniversaries. With this wide bunching region, there is only evidence of bunching for couples where the woman was 45 or older at the time of marriage. There are approximately 9.1 % more divorces occurring during years 10 and 11 than the estimated distribution predicts. The estimate of divorces missing during years 8 and 9 is statistically indistinguishable from 0, suggesting that the extra divorces may not be merely retimed from years 8 and 9. These results for older couples are consistent with the results from the 1- and 1.5-year bunching windows.

The results from the wider bunching regions suggest that older people may not have divorced if they would never have received spousal benefits or that they delay divorce from very early on in marriages for spousal benefits. These results make interpreting the results with the bunching region set to be 6 months before and after difficult to interpret for older individuals. It appears that the 10-year rule affects divorce rates for older couples into the 11th year after marriage, meaning estimates of ŷbefore and ŷafter might be biased downward when the bunching region is set to 6 months.

The results presented in this section are most comparable to Dickert-Conlin and Meghea (2004), who find no immediate impact of the 10-year rule’s implementation using Vital Statistics data from 1975 to 1980. 12 These differences suggest that while people may not have changed their behavior because of Social Security’s 10-year rule immediately, older people in particular soon began adjusting their divorce timing so that they could receive spousal benefits after divorce.

4 Changes in the Likelihood of Being Married

4.1 Data

While the Vital Statistics data are ideal for examining how the 10-year rule influences the distribution and timing of divorces, the Vital Statistics data lack many demographic and labor force variables, meaning they do not allow for knowing characteristics of couples the 10-year rule influences. Examining characteristics of couples is important because economic theory suggests that couples with large disparities between the primary and secondary earners should be the most responsive to the 10-year rule. The Vital Statistics records are also strictly a flow measure, whereas we are also interested in if and how the stock of marriages changes at the 10-year mark. Because of these issues and to focus on more recent data, I examine Social Security’s 10-year rule using Integrated Public Use Microsample Series (IPUMS) ACS data from 2008 to 2011.

Beginning in 2008, the ACS began asking people the year their most recent marriage began. I subtract people’s answers to this question from the survey year to calculate the years since their marriages began. I focus on people over the age of 17 who married 5–14 years prior. Because I am interested in how the likelihood of being divorced changes, only non-widowed ever-married people are included in the sample. With the ACS, the unit of observation is the individual because I can only identify both members of the couple when the couple is still married. Because women make up the vast majority of people receiving spousal benefits and to keep the results concise, I focus on women with the ACS analysis. 13 Since the SSA considers couples who have separated as having intact marriages, I code separated people as still being married. As with the Vital Statistics, the ACS also has the advantage of being large. The final sample contains 809,912 observations. This large sample size contrasts with the PSID, which also has information on marriage histories. The PSID only has data on 16,361 marriages, the vast majority of which are comprised of young people.

Despite its advantages, the ACS has three issues. The first is that we only know the length of marriage in years, which means that brief delays in divorce are difficult to observe. Unless divorces are delayed for considerable lengths of time or there is a spike in divorce rates after 10 years that is not from retimed marriages, we may not find any evidence the 10-year rule affects the likelihood of divorcing even if it does. For this reason, the ACS analysis focuses primarily on the divorces of older individuals. The second limitation is that the ACS only asks respondents the years of their most recent marriages. This means that if someone gets a divorce after 10 years of marriage and then remarries, she will not show up in the data as having been divorced after 10 years of marriage. Since divorce rates rise after 10-year anniversaries, the estimates would be biased toward 0 if this happens. Asking about the year of the most recent marriage but not the year of divorce also means that I do not know the length of marriages for divorced people, which leads to me studying the years since marriages began and current marital statuses. Third, Social Security PIA is calculated based on lifetime labor earnings, while the ACS only asks about current earnings. Using current earnings is especially problematic with older individuals since many are retired. To mitigate this concern, I use people’s education levels and the average earnings of their prior occupations to study what kind of marriages end immediately after 10-year anniversaries.

To explore the characteristics of couples who divorce, I create a series of indicator variables that capture within-couple specialization that are equal to 1 if the woman is in a certain type of marriage and 0 otherwise. Since people who are no longer married have a value of 0 for these indicator variables, these variables will allow for understanding what types of marriages end at 10-year anniversaries. The first indicator variable equals 1 if both members of the couple are in the labor force at the time of the interview. I create a separate variable equal to 1 if an individual is married to someone with a different labor force participation status than herself. Since current labor force participation is likely a poor proxy for lifetime earnings, I also take advantage of the ACS question that asks people if they have ever worked over the last 5 years. I create a variable equal to 1 if the woman is married to someone with the same answer to this question as herself and a separate variable equal to 1 if the woman is married to someone with a different answer to this question.

If people report that they have worked in the last 5 years, the ACS asks them about their occupation. I compute the mean earnings of each occupation and then create a variable equal to 1 if women are in marriages where one member is in an occupation that earns at least 50 % more on average than the other’s occupation. I create another variable equal to 1 if the woman is in a marriage with an occupational earnings difference of less than 50 %. When people have not worked during the last 5 years, they are assigned an occupational earnings of 0 and are thus identified as being in marriages with wide earning disparities if their spouses worked at all. Finally, I create an indicator variable equal to 1 if women are married to men with the same education levels as well as an indicator equal to 1 if women are married to men with different education levels. Education is an attractive measure because it predicts lifetime earnings and has the advantage of generally being fixed from an early age.

Means of key variables are shown in Table 4. As with the estimates, the means are weighted using IPUMS weights. An important difference between older and younger women in the sample is how many times they have been married. Only 6 % of younger women have been married more than once, while 83 % of the older sample has had multiple marriages.

Table 4:

Means of key variables from the America community survey.

Within 5–14 years of marriage9 years since marriage10 years since marriage
Full sample
Married0.8300.8290.816
Divorced0.1700.1710.184
Black0.1000.1010.104
White0.7630.7630.761
Hispanic0.1600.1590.162
Age39.67539.37740.275
College0.3250.3270.319
High school0.9080.9070.902
Married more than once0.3120.3160.315
Married to spouse with same LFP0.5050.4990.494
Married to spouse with diff. LFP0.2290.2350.229
Married to spouse with same LFP over last 5 years0.5970.5920.577
Married to spouse with diff. LFP over last 5 years0.1370.1410.145
Married to spouse in Occ with similar wages0.3400.3370.327
Married to spouse in Occ with different wages0.4900.4920.489
Married to spouse with same education level0.2890.2870.286
Married to spouse with diff. education levels0.5410.5420.530
n809,91282,28384,612
Married before age 25
Married0.8280.8250.820
Divorced0.1720.1750.180
Black0.0750.0740.077
White0.7690.7690.768
Hispanic0.2180.2190.225
Age30.61330.12931.066
College0.2720.2760.270
High school0.8870.8830.877
Married more than once0.0600.0560.057
Married to spouse with same LFP0.4640.4510.458
Married to spouse with diff. LFP0.2570.2680.260
Married to spouse with same LFP over last 5 years0.5770.5720.563
Married to spouse with diff. LFP over last 5 years0.1440.1460.155
Married to spouse in Occ with similar wages0.3440.3430.333
married to spouse in Occ with different wages0.4840.4820.487
Married to spouse with same education level0.2920.2910.291
Married to spouse with diff. education levels0.5350.5340.529
n258,28526,30926,754
Married from 25 to 44
Married0.8300.8300.816
Divorced0.1700.1700.184
Black0.1140.1160.119
White0.7490.7470.747
Hispanic0.1390.1380.137
Age41.23340.93441.881
College0.3740.3760.363
High school0.9240.9250.921
Married more than once0.3700.3780.378
Married to spouse with same LFP0.5290.5260.516
Married to spouse with diff. LFP0.2100.2110.208
Married to spouse with same LFP over last 5 years0.6170.6100.594
Married to spouse with diff. LFP over last 5 years0.1220.1270.130
Married to spouse in Occ with similar wages0.3520.3480.340
Married to spouse in Occ with different wages0.4780.4820.476
Married to spouse with same education level0.2900.2880.286
Married to spouse with diff. education levels0.5410.5430.530
n453,58645,89447,635
Married older than 44
Married0.8350.8360.805
Divorced0.1650.1640.195
Black0.1060.1070.111
White0.8190.8270.817
Hispanic0.0850.0770.089
Age61.06361.20261.855
College0.2350.2330.232
High school0.8900.8890.879
Married more than once0.8250.8260.825
Married to spouse with same LFP0.5070.5040.487
Married to spouse with diff. LFP0.2420.2510.240
Married to spouse with same LFP over last 5 years0.5560.5620.528
Married to spouse with diff. LFP over last 5 years0.1940.1930.199
Married to spouse in Occ with similar wages0.2580.2550.237
Married to spouse in Occ with different wages0.5770.5810.569
Married to spouse with same education level0.2740.2700.267
Married to spouse with diff. education levels0.5610.5660.538
n98,04110,08010,223

Note: The sample is from the 2008 to 2011 IPUMS ACS and consists of women whose most recent marriages began between 5 and 14 years prior to the survey year. All means are computed using IPUMS weights.

4.2 Empirical Strategy

With the ACS data, I consider what happens to the likelihood of being married at 10-year anniversaries by estimating the following equation:

yi=γst+Xiα+fli,λ+Diβ+ηi,
where i indexes the individual, t indexes the year, s indexes the state, y is either an indicator equal to 1 if the individual is married, X is a set of individual covariates that includes years of education, a vector of indicator variables for race, and a vector of indicator variables for age, γ is a full set of year and state interactions, f is a smooth function representing the profile of the dependent variable with respect to the length of marriage l in years, and D is an indicator equal to 1 if the marriage happened at least 10 years prior. I model f as a quadratic polynomial on either side of 10-year anniversaries. Including state-by-year fixed effects means eq. [3] accounts for any state differences and any state-specific shocks over time. For example, state-wide law changes regarding the division of assets and changes to a state’s economy are both captured by these fixed effects.

I omit several indicator variables from eq. [3] to avoid multicollinearity. Specifically, I omit the indicator for Alabama in 2008, the indicator for being white, and the indicator for the earliest possible age in each regression. The β coefficients can be interpreted as the percentage-point change in the likelihood of being married at 10 years since the beginning of marriages for people who have ever married.

4.3 Results

Figure 2 shows how the likelihood of remaining married trends with years since marriages began. The likelihood of remaining married trends smoothly at 10-year anniversaries for all women except for those who were 45 or older at the time of marriage. For women 45 and older at the time of marriage, the likelihood of being married gaps down at 10-year anniversaries.

Figure 2:
Figure 2:

Probability of remaining married by years since the marriage began, from the 2008 to 2011 ACS.

Citation: The B.E. Journal of Economic Analysis & Policy 16, 2; 10.1515/bejeap-2015-0168

The results from estimating eq. [3] with indicator variables for remaining married as the dependent variable are shown in Table 5. The estimates suggest that the likelihood of being married falls by 3.2 percentage points or 3.8 % as marriages of older women cross the 10-year threshold. This estimate implies that the likelihood of being divorced increases by 19.4 % as marriages of older women cross the 10-year mark. Since I classify separated women as being married, the likelihood of being married would trend smoothly at 10-year anniversaries if people were moving out before 10-year anniversaries but waiting to file the paperwork. The large coefficient indicates that older people are delaying changes in living arrangements as well.

Table 5:

Changes in the likelihood of remaining married at 10 years since marriage.

nIndicator variable for being married
Full sample809,912−0.006
(0.004)
Married younger than 25258,285−0.000
(0.007)
Married from 25 to 44453,586−0.003
(0.005)
Married older than 4498,041−0.032***
(0.011)

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The sample is from the 2008 to 2011 IPUMS ACS and consists of women whose most recent marriages began between 5 and 14 years prior to the survey year. Robust standard errors are shown in parentheses. All regressions use IPUMS weights and control for age, education, race, sex, year, and state interactions, and a quadratic with respect to years since marriages began on either side of 10-year anniversaries.

These results are consistent both with couples retiming their divorces and with there being extra divorces after 10-year anniversaries for older women. Couples opting to forgo divorcing at 9 years of marriage would mean the 10-year rule causes the likelihood of remaining married at 9 years to be artificially high. When these divorce-delaying couples divorce at 10 years of marriage, the likelihood of being married would gap down even if the 10-year rule only affected marriage timing. On the other hand, if there were extra divorces at 10-year anniversaries that did not come from couples delaying divorce, the likelihood of remaining married would also gap down at 10-year anniversaries.

Table 6 shows results from using narrower age bins. As in Table 2 with the Vital Statistics data, Table 6 indicates nonlinearities in response to the 10-year rule based on age. The estimated change in the likelihood of remaining married is negative and statistically significant for women who married from age 41 to age 55. For women who married at younger ages, there is no evidence of changes in the likelihood of remaining married at 10-year anniversaries.

Table 6:

Changes in the likelihood of remaining married at 10 years since marriage narrower age bins.

nProbability of remaining married at 9 years of marriageIndicator variable for being married
Married from 16 to 2087,2210.790−0.009
(0.013)
Married from 21 to 25213,5200.8480.002
(0.007)
Married from 26 to 30174,0450.848−0.005
(0.008)
Married from 31 to 35109,8940.8190.008
(0.011)
Married from 36 to 4077,7550.801−0.001
(0.013)
Married from 41 to 4557,5360.813−0.025*
(0.015)
Married from 46 to 5040,0670.823−0.040**
(0.018)
Married from 51 to 5523,1970.838−0.040*
(0.022)
Married from 56 to 6011,8090.852−0.019
(0.029)

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The sample is from the 2008 to 2011 IPUMS ACS and consists of women whose most recent marriages began between 5 and 14 years prior to the survey year. Robust standard errors are shown in parentheses. All regressions use IPUMS weights and control for age, education, race, sex, year, and state interactions, and a quadratic with respect to year since marriages began on either side of 10-year anniversaries.

Next I estimate variations of eq. [3] to examine the characteristics of marriages that end at 10 years. 14 I focus the discussion on older people since they are the ones for whom a discontinuity was documented in Table 5. In column 1 of the top panel of Table 7, the dependent variable equals 1 if the individual is in a marriage where both members have the same labor force participation. A negative and significant β coefficient would imply that the marriages that end abruptly at 10-year anniversaries are those where members have the same labor force participation. The coefficient on crossing the 10-year mark is statistically indistinguishable from 0, suggesting that it is not couples with identical labor force participation statuses that drive the fall in marriage probabilities documented in Table 5. In column 1 of the bottom panel of Table 7, the dependent variable equals 1 if the individual is married to someone with a different labor force status than herself. The coefficient of –0.035 suggests that the marriages that end are those where spouses had different labor force statuses.

Table 7:

Types of marriages that end at 10 years since marriage.

nIndicator variable for being married to spouse with same LFPIndicator variable for being married to spouse with same LFP over last 5 yearsIndicator variable for being married to spouse in Occ with similar earningsIndicator variable for being married to spouse with same education level
Full sample809,9120.002–0.0010.0030.003
(0.005)(0.005)(0.005)(0.004)
Married younger than 25258,2850.008–0.000–0.0020.008
(0.009)(0.009)(0.009)(0.008)
Married from 25 to 44453,586–0.0020.0040.0090.002
(0.007)(0.006)(0.006)(0.006)
Married older than 4498,0410.004–0.026*–0.0050.000
(0.014)(0.014)(0.012)(0.012)
nIndicator variable for being married to spouse with diff. LFPIndicator variable for being married to spouse with diff. LFP over last 5 yearsIndicator variable for being married to spouse in Occ with diff. earningsIndicator variable for being married to spouse with diff. education levels
Full sample809,912–0.007*–0.005–0.009*–0.009*
(0.004)(0.003)(0.005)(0.005)
Married younger than 25258,285–0.0040.0040.002–0.008
(0.008)(0.006)(0.009)(0.009)
Married from 25 to 44453,586–0.003–0.008*–0.012*–0.005
(0.005)(0.004)(0.007)(0.007)
Married older than 4498,041–0.035***–0.005–0.027*–0.032**
(0.012)(0.011)(0.014)(0.014)

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The sample is from the 2008 to 2011 IPIJMS ACS and consists of women whose most recent marriages began between 5 and 14 years prior to the survey year. Robust standard errors are shown in parentheses. All regressions use IPIJMS weights and control for age, education, race, sex, year, and state interactions, and a quadratic with respect to year since marriages began on either side of 10-year anniversaries.

Column 2 of Table 7 displays β estimates from eq. [3] with the indicator variables based on whether or not both members of the couple have ever worked over the last 5 years as the dependent variables. These results indicate that couples whose marriages end are those where both members answered the same to having ever worked in the last 5 years. In column 3, the dependent variables are based on average earnings of the spouses’ occupations. The coefficient of –0.027 when the dependent variable is an indicator for couples having been in occupations with different average earnings suggests that it is couples with different earning potentials who divorce at their 10-year anniversaries.

The last column of Table 7 evaluates how similar the education levels are for couples who divorce at 10-year anniversaries. In the last column of the top panel of Table 7, the dependent variable is 1 if women are married to husbands with the same education levels. In all cases, the coefficients are statistically insignificant. In the bottom panel, the dependent variable is an indicator equal to 1 if the woman is married to someone with a different education level. The estimate of –0.032 for older couples indicates that the couples who divorce are those where one member has more education than the other member.

These results are most comparable to Goda, Shoven, and Slavov (2007), who also study how marriage stocks change at 10-year anniversaries and find small, statistically insignificant differences in divorce probabilities at 10 years of marriage between couples with and without large earning disparities between the primary and secondary earners. My estimates of the change in the likelihood of being divorced at 10-year anniversaries are small for the full sample as well. Unlike the PSID, though, the ACS allows for studying heterogeneity based on age and reveals important differences in age at marriage.

We would not expect divorces from the 10-year rule to comprise a large share of total divorces since the vast majority of marriages are not near their 10-year anniversaries and because many factors unrelated to Social Security influence divorce. Nevertheless, the ACS allows for performing a back-of-the-envelope calculation to better understand the percent of divorces for people who married over the age of 44 that are influenced by the 10-year rule. According to the ACS, 98,611 marriages where the woman was 45 or older at marriage ended in divorce in 2011. Of these divorces, 9,049 ended at marriage durations of 10 and 11. The Vital Statistics bunching estimate implies that 9.1 %, or 823, of those divorces in years 10 and 11 are additional divorces that would not have occurred in years 10 and 11 without the sudden change in spousal benefits after divorce at 10-year anniversaries. These numbers imply that the 10-year rule influenced about 0.84 % of divorces in 2011 for older couples, which is a small but nontrivial share of divorces.

4.4 Heterogeneity, Robustness, and Placebo Tests

I now test for heterogeneity based on marital status, consider the robustness of these results to various estimation choices, and conduct placebo analyses. Table 8 displays results with the dependent variable being an indicator for being married, but the results are similarly robust for other outcome variables.

Table 8:

Robustness of estimates with dependent variable being an indicator for being married.

(1)(2)(3)(4)(5)(6)(7)(8)(9)
Full sample–0.003–0.011–0.003–0.007**–0.006–0.006–0.007*0.004–0.001
(0.005)(0.007)(0.004)(0.003)(0.004)(0.004)(0.004)(0.004)(0.004)
Married younger than 25–0.0020.0120.005–0.007–0.0010.002–0.0020.007–0.007
(0.007)(0.033)(0.008)(0.005)(0.007)(0.007)(0.007)(0.007)(0.007)
Married from 25 to 44–0.004–0.001–0.001–0.003–0.003–0.005–0.0040.002–0.002
(0.006)(0.009)(0.006)(0.004)(0.005)(0.005)(0.005)(0.005)(0.005)
Married older than 44–0.008–0.037***–0.037***–0.026***–0.031***–0.030***–0.033***–0.0010.021*
(0.028)(0.012)(0.012)(0.008)(0.011)(0.011)(0.011)(0.010)(0.011)
Sample/estimation alterationMarried onceMarried at least twiceCoding separations as unmarriedUnweighted estimatesInteracting demographics with fControlling for health, income, and childrenAverage marginal effects from ProbitPlacebo test at 9 yearsPlacebo test at 11 years

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The sample is from the 2008 to 2011 IPIJMS ACS and consists of women whose most recent marriages began between 5 and 14 years prior to the survey year. Robust standard errors are shown in parentheses. All regressions use IPIJMS weights and control for age, education, race, sex, year, and state interactions, and a quadratic with respect to years since marriages began on either side of 10-year anniversaries. The estimates consider the robustness of the estimates in column 1 of Table 5. In column 1, the sample size for each row is 537,443, 243,890, 278,413, and 15,140. In column 2, the sample size for each row is 272,469, 14,395, 175,173, and 809,912. In columns 3 through 7, the sample size for each row is 809,912, 258,285, 453,586, and 98,041. In column 8, the sample size for each row is 817,298, 257,745, 456,076, and 103,477. In column 9, the sample size for each row is 801,014, 258,821, 450,334, and 91,859.

I first test for differences between people in their first marriages and people who have been married more than once. Heterogeneity by marital history may exist because people who have been married multiple times may be more aware of divorce rules than people who are on their first marriages. Alternatively, people who are on subsequent marriages may be eligible for spousal benefits from previous marriages and may therefore not care about reaching the 10-year mark in their current marriages. Column 1 displays results with the sample restricted to people who have only married one time, while column 2 displays results with the sample restricted to only people who have had multiple marriages. For people married more than once, the point estimate of the fall in the likelihood of remaining married at 10 years is larger in absolute value for the full sample and for people who married at 45 or older. For people married only once, the fall in the likelihood of remaining married is statistically insignificant. These results suggest that older people in subsequent marriages may be more responsive to the 10-year rule.

For the main analysis, I coded an individual as being married if she was separated because the SSA uses the official marriage length in determining eligibility for the spousal benefit. In column 3, I test the sensitivity of the results to defining being married as being 0 if people are separated. The results are similar to the original estimates.

All of the ACS estimates are weighted using the IPUMS sample weights. In their review of econometric issues associated with survey weights, Solon, Haider, and Wooldridge (2013) suggest considering both weighted and unweighted estimates. Column 4 tests the sensitivity of the estimates to not using weights. The point estimate for the change in the likelihood of being married at 10 years of marriage for older people is statistically indistinguishable from the prior estimate and from 0. 15

Another possible concern is that the controls in eq. [3] may be inadequate, which would be the case if the effects of demographic variables change at the cutoff. To consider this possibility, I supplement eq. [3] with interactions of years of education, age, and the nonomitted race indicators with quadratic polynomials with respect to years since the start of marriages. The results are shown in column 5 and are almost identical to the original results, suggesting returns to demographic variables changing at the threshold do not drive the fall in the likelihood of remaining married at 10-year anniversaries. A related concern is that many factors affect divorce that I do not control for, such as health, income, and the presence of children in the household. These factors changing discontinuously at 10-year anniversaries could potentially bias the results. 16 In column 6, I supplement eq. [3] with controls for the presence of children, the individual’s annual earnings, and indicator variables for having a cognitive difficulty, an ambulatory difficulty, self-care difficulty, vision difficulty, or difficulty living alone. The estimates are very similar to the original estimates.

An alternative to estimating eq. [3] using ordinary least squares (OLS) would be to estimate the model using a probit regression. Column 7 of Table 8 displays average marginal effects from probit regressions of eq. [3]. The estimates of the average change in the likelihood of being married at 10-year anniversaries from the probit regressions are nearly identical to the OLS estimates.

One might also be concerned that f is not flexible enough to capture the likelihood of remaining married trending smoothly with years since marriages began for older people. In columns 8 and 9, I replicate the analysis using marriage durations of 9 and 11 years as placebo cutoffs. One of the estimates of abrupt changes in the likelihood of being married at these marriage lengths is statistically significant at the 10 % level. An issue with using the years immediately before and after the cutoff is that f can be influenced by the cutoff at 10 years. As a further check, I replicate the analysis setting marriage durations of 6 through 8 and 12 through 14 as the cutoffs. The point estimates are significant at least the 10 % level 12.5 % of the time, which is close to the 10 % we would expect from chance. These results provide evidence that f can sufficiently account for the marriage profile trending smoothly.

Finally, I include widows in the sample and consider how the likelihood that women are widows changes at 10-year anniversaries. The likelihood that a woman classifies herself as a widow could change at 10-year anniversaries if divorced women are more likely to consider themselves widows rather than divorced if their ex-husbands die while they are receiving spousal benefits. However, we would be concerned if the likelihood of being a widow changes dramatically at 10-year anniversaries.

Figure 3 shows how the likelihood of being a widow changes with years since marriages began. Estimates of eq. [3] with the dependent variable being an indicator equal to one if the woman is a widow are shown in Table 9. The profiles with respect to years since marriage do not provide evidence of large spikes at 10-year anniversaries. The change in the likelihood of being a widow is estimated to increase by 0.3 percentage points and is significant at the 5 % level. None of the estimated discontinuities are statistically significant for any of the age groups. Although statistically significant for the full sample, the estimate is 91 % smaller than the estimate for the decrease in the likelihood of remaining married after 10 years of marriage and does not indicate that widowhood increases dramatically at 10-year anniversaries.

Figure 3:
Figure 3:

Probability of being widowed by years since the marriage began, from the 2008 to 2011 ACS.

Citation: The B.E. Journal of Economic Analysis & Policy 16, 2; 10.1515/bejeap-2015-0168

Table 9:

Changes in the likelihood of being widowed at 10 years since marriage.

nProbability of being widowed at 9 years of marriageIndicator variable for being widowed
Full sample832,5860.0260.003**
(0.001)
Married younger than 25260,1350.0080.001
(0.002)
Married from 25 to 44461,3040.0180.003
(0.002)
Married older than 44111,1470.1210.013
(0.008)

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The sample is from the 2008 to 2011 IPUMS ACS and consists of women whose most recent marriages began between 5 and 14 years prior to the survey year. Robust standard errors are shown in parentheses. All regressions use IPUMS weights and control for age, education, race, sex, year, and state interactions, and a quadratic with respect to years since marriages began on either side of 10-year anniversaries.

5 Conclusion

Social Security is a key part of retirement plans and retirement income for Americans. This paper provides evidence that Social Security’s requirement that spouses be married for at least 10 years before qualifying for spousal benefits influences divorce timing and propensities. Around 2 % of divorces occurring within the 6 months after 10-year anniversaries would have occurred before them if not for Social Security’s 10-year rule. For older couples, the effects are even more dramatic. It appears that many older couples would not have divorced if they could not receive spousal benefits after divorce or that they delay divorce for many years to benefit from the 10-year rule. Even many middle-aged couples, who account for over 40 % of all divorces in the sample, delay divorcing until after their 10-year anniversaries.

The likelihood of being divorced gaps up at 10-year anniversaries for women 45 or older at the time of marriage, suggesting Social Security’s 10-year rule is not only affecting the timing of divorce paperwork. Instead, people delay changing living arrangements until after 10 years of marriage. The marriages that end are ones where one member of the couple likely earned significantly more than the other, which speaks to the importance of the spousal benefit for women who specialize in home production. As the 10-year rule means that many secondary earners are better off from divorcing after 10-year anniversaries relative to divorcing before 10-year anniversaries, the results from this paper provide evidence that financial well-being after divorce is a consideration for people when making the decision to leave a marriage.

Appendix

A Trend Breaks before the 10-Year Rule

To test for similar discontinuities at 10 years of marriage before the 10-year rule was implemented, I use data from 1966 to 1974. The reason for allowing several years of data before the passage of the 1977 law is that the law applied to all existing divorces, not just the ones happening after the law was passed. This means divorce rates might change before the law if people were anticipating the change from a 20-year to a 10-year requirement.

I first conduct Kolmogorov–Smirnov tests to determine whether or not the distributions of marriage lengths are different for the two time periods. A failure to reject that they are different would cast doubt on the 10-year rule being responsible for the discontinuity observed at 10 years of marriage presented in the main text. Both for the full sample and for each age group, the Kolmogorov–Smirnov tests reject the null hypotheses that the distributions are the same at the 1 % level. These tests provide suggestive evidence that the 10-year rule influences divorce timing.

As divorce norms and economic factors changed between these two time periods, it is possible that the distributions could be different even without the 10-year rule being implemented. As such, I next replicate the approach taken in the main analysis to test for discontinuities at 10 years of marriage before the 10-year rule was implemented. Graphs of the divorce profiles before the 10-year rule was implemented are shown in Figure 4. As with the main Vital Statistics results, the number of divorces falls with the duration of marriages. In none of the graphs does there appear to be any gaps or bunching associated with 10 years of marriage. Table 10 shows estimates of the discontinuities at 10-year anniversaries using data from 1966 to 1974. All of the estimates are statistically indistinguishable from 0 and provide no evidence of discontinuities before the 10-year rule was implemented.

Figure 4:
Figure 4:

Divorces by marriage length, from the 1966 to 1974 Vital Statistics data.

Citation: The B.E. Journal of Economic Analysis & Policy 16, 2; 10.1515/bejeap-2015-0168

Table 10:

Discontinuity in divorces at 10 years of marriage before 10-year rule.

Divorces at 9 yearsNumber of divorcesLog of number of divorces
All couples17,950–14–0.004
(23)(0.016)
Married younger than 2510,72630.011
(19)(0.021)
Married from 25 to 446,395–15–0.028
(13)(0.025)
Married older than 44829–2–0.020
(15)(0.080)

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. Robust standard errors are in parentheses below the estimates. The data are from the 1966 to 1974 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. All regressions control for quadratics in length of marriage in months on either side of 10-year anniversaries of marriages. The discontinuity in divorces at 10 years is per month.

Sensitivity of Bunching Estimates to Polynomial

The bunching results presented in Section 3 modeled g as a cubic polynomial in eq. [2]. Tables 11 through 14 show the sensitivity of the bunching results to using polynomials of different degrees. The results using bunching regions of 6 months, 1 year, or 1.5 years are very similar regardless of the polynomial used. When the bunching region is 2 years wide, the estimate of extra divorces after 10-year anniversaries for older couples is not significant for higher polynomials. However, the estimates are similar in size and do not provide evidence that contradicts the results presented in the paper.

Table 11:

Bunching of divorces around 10 years of marriage – sensitivity to polynomial choice.

Estimate of divorces from 9.5 to 10 years without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.Estimate of divorces from 10 to 10.5 years without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.
Second-degree polynomial
All couples36,220–827211–2.28***33,9657401982.18***
Married younger than 2522,790–365165–1.6**21,6774121621.90***
Married from 25 to 4412,618–398123–3.15***11,5552371162.05**
Married older than 44812–6431–7.83**732923112.49***
Fourth-degree polynomial
All couples36,127–734231–2.03***34,0017042142.07***
Married younger than 2522,751–326180–1.43*21,6943951721.82**
Married from 25 to 4412,561–341132–2 71***11,5612311252.00*
Married older than 44815–6734–8.26**746783310.46***
Fifth-degree polynomial
All couples36,145–752233–2.08***33,9847212142.12***
Married younger than 2522,775–350182–1.54*21,6704191721.93***
Married from 25 to 4412,556–336133–2.68***11,5662261241.96*
Married older than 44813–6534–8.03*748763210.16***

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The data are from the 1985 to 1995 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. The estimates of divorces that would have occurred without the 10-year rule come from fitting a polynomial excluding data from the bunching region and then using the parameter estimates to estimate the divorces that would have occurred in the bunching region. Standard errors are calculated from the bootstrapping procedure described in the paper and are shown in parentheses. The results in Table 3 modeled g as a cubic polynomial.

Table 12:

Bunching of divorces around 10 years of marriage – sensitivity to polynomial choice.

Estimate of divorces at year 9 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.Estimate of divorces at year 10 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.
Second-degree polynomial
All couples75,301–1,726353–2.29***66,2761753280.26
Married younger than 2547,095–907276–1.93***42,643–138260–0.32
Married from 25 to 4426,503–728205–2.75***22,2481581860.71
Married older than 441,704–9250–5.38*1,3861545211.13***
Fourth-degree polynomial
All couples75,478–1,903440–2.52***66,957–506417–0.76
Married younger than 2547,323–1,135344_2 4***43,104–599326–1.39*
Married from 25 to 4426,446–671252–2.54***22,429–23238–0.1
Married older than 441,709–9764–5.691,423117638.19*
Fifth-degree polynomial
All couples75,550–1,975455–2.61***66,885–434413–0.65
Married younger than 2547,411–1,223355–2.58***43,016–511325–1.19
Married from 25 to 4426,432–656263–2 48***22,443–37232–0.17
Married older than 441,707–9568–5.571,426114628.02*

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The data are from the 1985 to 1995 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. The estimates of divorces that would have occurred without the 10-year rule come from fitting a polynomial excluding data from the bunching region and then using the parameter estimates to estimate the divorces that would have occurred in the bunching region. Standard errors are calculated from the bootstrapping procedure described in the paper and are shown in parentheses.

Table 13:

Bunching of divorces around 10 years of marriage – sensitivity to polynomial choice.

Estimate of divorces at years 8.5 to 9 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.Estimate of divorces at years 10 and 11.5 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.
Second-degree polynomial
All couples116,681–1,660444–1 42***34,0896162051.81***
Married younger than 2572,277–578358–0.821,7093801691.75**
Married from 25 to 4441,639–892256–2 14***11,6281641191.41
Married older than 442,766–19164–6.9***75173329.67**
Fourth-degree polynomial
All couples115,784–763558–0.6634,0326732311.98***
Married younger than 2571,803–104451–0.1421,6604291891.98**
Married from 25 to 4441,201–453316–1.111,5951971361.7
Married older than 442,781–20683–7.4***77648356.16
Fifth-degree polynomial
All couples116,137–1,116688–0.9634,0746312371.85***
Married younger than 2572,224–525556–0.7321,7103791931.74*
Married from 25 to 4441,118–371413–0.911,5852071421.78
Married older than 442,795–220113–7.88*77846375.93

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The data are from the 1985 to 1995 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. The estimates of divorces that would have occurred without the 10-year rule come from fitting a polynomial excluding data from the bunching region and then using the parameter estimates to estimate the divorces that would have occurred in the bunching region. Standard errors are calculated from the bootstrapping procedure described in the paper and are shown in parentheses.

Table 14:

Bunching of divorces around 10 years of marriage – sensitivity to polynomial choice.

Estimate of divorces at years 8 and 9 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.Estimate of divorces at years 10 and 11 without 10-year ruleDiff. between observed divorcesStandard errorPercent diff.
Second-degree polynomial
All couples160,478–1,203790–0.75124,2061,8727731.51***
Married younger than 2599,017–462637–0.4781,1437046250.87
Married from 25 to 4457,687–620458–1.0840,5808974522.21**
Married older than 443,773–120120–3.172,48427011810.87**
Fourth-degree polynomial
All couples160,951–1,6761,764–1.04126,721–6431,736–0.51
Married younger than 25100,360–1,8051,394–1.883,526–1,6791,367–2.01
Married from 25 to 4456,8012661,0640.4740,6138641,0622.13
Married older than 443,790–137268–3.612,5821722626.68
Fifth-degree polynomial
All Couples161,266–1,9911,962–1.23126,407–3291,659–0.26
Married Younger than 25100,756–2,2011,529–2.1883,130–1,2831,346–1.54
Married from 25 to 4456,6324351,2190.7740,7836949731.7
Married older than 443,878–225311–5.82,49326123710.46

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. The data are from the 1985 to 1995 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. The estimates of divorces that would have occurred without the 10-year rule come from fitting a polynomial excluding data from the bunching region and then using the parameter estimates to estimate the divorces that would have occurred in the bunching region. Standard errors are calculated from the bootstrapping procedure described in the paper and are shown in parentheses.

C Discontinuities in Divorces at Prior Cutoff

I now test for evidence of a discontinuity at 20-year anniversaries using data from 1966 to 1974, which is when the SSA required 20 years of marriage before divorced spouses were eligible for spousal benefits. Because few people marry older than 44 and divorce around 20th anniversaries and because the people who marry at 25 or older are in their mid-forties or older at their 20-year anniversaries, I divide couples into two age groups instead of three. Figure 5 shows how divorces trend with marriage duration near 20-year anniversaries. It appears as though divorces may increase abruptly after 20-year anniversaries for couples who married older than 24 but not dramatically so.

Figure 5:
Figure 5:

Divorces by marriage length, from the 1966 to 1974 Vital Statistics data.

Citation: The B.E. Journal of Economic Analysis & Policy 16, 2; 10.1515/bejeap-2015-0168

I next estimate eq. [2] setting 20 years as the cutoff and using data from 1966 to 1974 on divorces that occurred within 4 years of 20th anniversaries. The results are shown in Table 15. Although I cannot rule out a large discontinuity for older couples, neither the graphs nor the estimates provide compelling evidence of a discontinuity in divorces at 20 years of marriage. This ambiguous evidence contrasts with the strong evidence of bunching around 10-year anniversaries. These results may suggest that people who have been married 20 years are less sensitive to Social Security incentives than people who have been married for 10 years. Alternatively, they may also indicate that the passage of the 10-year rule raised awareness that Social Security provided spousal benefits to marriages that lasted certain lengths before divorce.

Table 15:

Discontinuity in divorces at 20 years of marriage prior to 10-year rule.

Divorces at 19 yearsDiscontinuity at 20 years
Number of divorcesLog of number of divorces
All couples7,372–0–0.004
(17)(0.030)
Married younger than 254,285–13–0.041
(14)(0.040)
Married older than 24652130.043
(11)(0.044)

Notes: *, **, and *** indicate significance at 10 %, 5 %, and 1 %, respectively. Robust standard errors are in parentheses below the estimates. The data are from the 1966 to 1974 Vital Statistics data on divorces and include all divorces that occurred within 4 years of 10-year anniversaries. All regressions control for quadratics in length of marriage in months on either side of 10-year anniversaries of marriages. The discontinuity in divorces at 10 years is per month.

References

  • Alm, J., and L. A. Whittington. 1997. “Income Taxes and the Timing of Marital Decisions.” Journal of Public Economics 64 (2):219–40.

    • Crossref
    • PubMed
    • Export Citation
  • Ananat, E. O., and G. Michaels. 2008. “The Effect of Marital Breakup on the Income Distribution of Women with Children.” Journal of Human Resources 43 (3):611–29.

  • Bastani, S., and H. Selin. 2014. “Bunching and Non-bunching at Kink Points of the Swedish Tax Schedule.” Journal of Public Economics 109:36–49.

    • Crossref
    • Export Citation
  • Bedard, K., and O. Deschenes. 2005. “Sex Preferences, Marital Dissolution, and the Economic Status of Women.” Journal of Human Resources 40 (2):411–34.

  • Bitler, M. P., J. B. Gelbach, H. W. Hoynes, and M. Zavodny. 2004. “The Impact of Welfare Reform on Marriage and Divorce.” Demography 41 (2):213–36.

    • Crossref
    • PubMed
    • Export Citation
  • Brown, S. L., and I. - Fen Lin. 2012. “The Gray Divorce Revolution: Rising Divorce among Middle-Aged and Older Adults, 1990–2010.” The Journals of Gerontology Series B: Psychological Sciences and Social Sciences 67 (6):731–41.

    • Crossref
    • Export Citation
  • Brown, S. L., G. R. Lee, and J. R. Bulanda. 2006. “Cohabitation among Older Adults: A National Portrait.” The Journals of Gerontology Series B: Psychological Sciences and Social Sciences 61 (2):S71–9.

    • Crossref
    • Export Citation
  • Butrica, B. A., and K. E. Smith. 2012. “The Retirement Prospects of Divorced Women.” Social Security Bulletin 72 (1):11–22.

    • PubMed
    • Export Citation
  • Couch, K. A., C. R. Tamborini, G. Reznik, and J. W. R. Phillips. 2013. “Divorce, Womens Earnings, and Retirement Over the Life Course.” In Lifecycle Events and Their Consequences: Job Loss, Family Change, and Declines in Health, edited by Kenneth, C. Mary, D. and Julie, Z., 133.

  • Dickert-Conlin, S., and C. Meghea. 2004. “The Effect of Social Security on Divorce and Remarriage Behavior.” CRR WP 2004–09, April 2004.

  • Gadalla, T. M. 2008. “Gender Differences in Poverty Rates after Marital Dissolution: A Longitudinal Study.” Journal of Divorce & Remarriage 49 (3–4):225–38.

    • Crossref
    • Export Citation
  • Gardner, J., and A. J. Oswald. 2006. “Do Divorcing Couples Become Happier by Breaking Up?.” Journal of the Royal Statistical Society: Series A (Statistics in Society) 169 (2):319–36.

    • Crossref
    • Export Citation
  • Genadek, K. R., W. A. Stock, and C. Stoddard. 2007. “No-Fault Divorce Laws and the Labor Supply of Women with and Without Children.” Journal of Human Resources 42 (1):247–74.

  • Goda, G. S., J. B. Shoven, and S. N. Slavov. 2007. “Social Security and the Timing of Divorce.” NBER Working Paper No. 13382.

    • Crossref
    • Export Citation
  • Kennedy, S., and S. Ruggles. 2014. “Breaking Up Is Hard to Count: The Rise of Divorce in the United States, 1980–2010.” Demography 51 (2):587–98.

    • Crossref
    • PubMed
    • Export Citation
  • Kleven, H. J., and M. Waseem. 2013. “Using Notches to Uncover Optimization Frictions and Structural Elasticities: Theory and Evidence from Pakistan.” The Quarterly Journal of Economics 128 (2):669–723.

  • Kopczuk, W., and D. J. Munroe. 2015. “Mansion Tax: The Effect of Transfer Taxes on the Residential Real Estate Market.” American Economic Journal. Applied Economics 7 (2):214–257.

  • Lavelle, B., and P. J. Smock. 2012. “Divorce and Womens Risk of Health Insurance Loss.” Journal of Health and Social Behavior 53 (4):413–31.

    • Crossref
    • Export Citation
  • Manoli, D. S., and A. Weber. 2011. “Nonparametric Evidence on the Effects of Financial Incentives on Retirement Decisions (No. w17320).” National Bureau of Economic Research.

    • Crossref
    • Export Citation
  • Page, M. E., and A. H. Stevens. 2004. “The Economic Consequences of Absent Parents.” Journal of Human Resources 39 (1):80–107.

    • Crossref
    • Export Citation
  • Persson, P. 2013. “Social Insurance and the Marriage Market”.

  • Peters, E. H., K. Simon, and J. R. Taber. 2014. “Marital Disruption and Health Insurance.” Demography 51 (4):1397–421.

    • Crossref
    • PubMed
    • Export Citation
  • Ratcliffe, C. E., G. P. Acs, T. Dore, and D. Z. Moskowitz. 2008. “Assessment of Survey Data for the Analysis of Marriage and Divorce at the National, State, and Local Levels.” The Urban Institute, September.

    • Crossref
    • Export Citation
  • Saez, E. 2010. “Do Taxpayers Bunch at Kink Points?.” American Economic Journal: Economic Policy 2 (3):180–212.

  • Schoen, R., and N. Standish. 2001. “The Retrenchment of Marriage: Results from Marital Status Life Tables for the United States, 1995.” Population and Development Review 27 (3):553–63.

    • Crossref
    • Export Citation
  • Shryock, H., and J. Siegel. 1973. “The Methods and Materials of Demography.” Bureau of the Census.

  • Smock, P. J., W. D. Manning, and S. Gupta. 1999. “The Effect of Marriage and Divorce on Women’s Economic Well-Being.” American Sociological Review 64 (6):794–812.

  • Solon, G., S. J. Haider, and J. Wooldridge. 2013. “What Are We Weighting for?.” No. w18859. National Bureau of Economic Research.

    • Crossref
    • Export Citation
  • Stevenson, B., and J. Wolfers. 2007. “Marriage and Divorce: Changes and Their Driving Forces.” The Journal of Economic Perspectives 21 (2):27–52.

  • U.S. Social Security Administration. 2013. Annual Statistical Supplement to the Social Security Bulletin, 2012. Washington, D.C: U.S. Government Printing Office, http://www.ssa.gov/policy/docs/statcomps/supplement/2012/oasdi.pdf.

Footnotes

1

Much research studies the consequences of divorce on economic outcomes. For example, refer to Ananat and Michaels (2008), Bedard and Deschenes (2005), Couch et al. (2013), Gadalla (2008), Genadek, Stock, and Stoddard (2007), Lavelle and Smock (2012), Page and Stevens (2004), Peters et al. (2014), and Smock, Manning, and Gupta (1999).

2

Studying the implementation of the 10-year rule has added empirical challenges because the 10-year rule applied to divorces retroactively, meaning people could have delayed divorce before the law became effective if they anticipated the law. Alternatively, if people who would have delayed divorce until the 10-year mark had already divorced when the law was passed, it could take several years before the 10-year rule would display an effect.

3

People receive their full PIA if they retire at the full retirement age. Beginning in 2000, the full retirement age began to rise incrementally from age 65 to age 67. People can retire starting at age 62 and receive a reduced benefit (Social Security Administration 2013).

4

Spousal benefits are different than survivor benefits. A widow or widower is also entitled to the deceased spouse’s full PIA as long as the marriage lasted at least 9 months before the death of the spouse.

5

A divorced spouse receiving a spousal benefit does not affect the other spouse’s Social Security payment in any way. If a person has several marriages that last 10 years, all former spouses can claim spousal benefits under his PIA without affecting his benefit. Married people can receive spousal benefits after 1 year of marriage.

6

For both older and younger people, Social Security benefits can be difficult to calculate, and people often have a difficult time estimating their benefits. To give people a better idea of their benefits, the SSA began mailing of benefits in 1995.

7

Refer to Shryock and Siegel (1973) for more information about Vital Statistics divorce data and to Kennedy and Ruggles (2014) for more information about the history of the collection of divorce data in the United States more generally.

8

Bunching analysis has also been used to study responses to tax thresholds by Bastani and Selin (2014), Kleven and Waseem (2013), Kopczuk and Munroe (2014), and Saez (2010). Manoli and Weber (2011) use bunching analysis to study how employer-provided retirement benefits affect retirement timing.

9

In results available upon request, I also test for discontinuities in divorce rates at 5, 15, and 20 years of marriage and find none, which provides more evidence that people do not delay divorce to reach certain milestones.

10

In Appendix C, I test for evidence of bunching in divorces at 20 years of marriage using data from 1966 to 1974, which is when Social Security required 20 years of marriage before ex-spouses would be eligible for spousal benefits. I do not find strong evidence that people were delaying divorces to reach 20-year anniversaries. This null result suggests that people who have been married for longer may be less responsive to Social Security’s rules regarding divorce or that the switch to the 10-year rule raised awareness of Social Security’s divorce rules.

11

I do not consider higher age bins because there are not couples divorcing at each marriage duration in months for older ages.

12

Dickert-Conlin and Meghea (2004) use a difference-in-differences strategy to identify the immediate effect of the law by comparing divorces in 1978–1980 to divorces in 1975–1977. When I compare the discontinuities from eq. [1] from these two time periods, I find no significant differences, which corroborates the Dickert-Conlin and Meghea result of no immediate impact of the law and suggests that differences in methods are not the reason I find the 10-year rule influences divorce.

13

I also restrict the sample to only women to avoid counting people from the same marriages twice. The results for men are similar to women except that it appears as though there is a slight gap at 10 years since marriage for men in the middle age group. This is not entirely surprising given the fact that it is likely the age of the woman and not the age of the man that matters since women are likely the ones benefit from the 10-year rule.

14

I focus on understanding what kind of marriages end at 10 years of marriage because I can only observe characteristics of both members of the couple when the couple is living together. Once a woman is divorced and no longer living with her husband, I can no longer determine characteristics of her marriage.

15

In their analysis of marriage and divorce data, Ratcliffe et al. (2008) evaluate 20 data sets containing marital information along a variety of dimensions, including representativeness, and conclude that the ACS is one of the three best. Unfortunately, though, no other data set with marriage lengths is large enough to verify that the ACS information on years since marriages began is nationally representative. Since I use the ACS to verify that the likelihood of being married and divorced changes abruptly at 10 years of marriage and to study the characteristics of marriages that end at 10 years, the sample does not need to be representative for the results to be valid. Rather, the likelihood of being included in the survey cannot change at 10 years of marriage, which it presumably does not.

16

For these factors to be driving the changes at 10 years of marriage, they would have to change at 10-year anniversaries other than through divorce. A concern with controlling for these factors is that divorce likely affects health, income, and the presence of children.

If the inline PDF is not rendering correctly, you can download the PDF file here.

  • Alm, J., and L. A. Whittington. 1997. “Income Taxes and the Timing of Marital Decisions.” Journal of Public Economics 64 (2):219–40.

    • Crossref
    • PubMed
    • Export Citation
  • Ananat, E. O., and G. Michaels. 2008. “The Effect of Marital Breakup on the Income Distribution of Women with Children.” Journal of Human Resources 43 (3):611–29.

  • Bastani, S., and H. Selin. 2014. “Bunching and Non-bunching at Kink Points of the Swedish Tax Schedule.” Journal of Public Economics 109:36–49.

    • Crossref
    • Export Citation
  • Bedard, K., and O. Deschenes. 2005. “Sex Preferences, Marital Dissolution, and the Economic Status of Women.” Journal of Human Resources 40 (2):411–34.

  • Bitler, M. P., J. B. Gelbach, H. W. Hoynes, and M. Zavodny. 2004. “The Impact of Welfare Reform on Marriage and Divorce.” Demography 41 (2):213–36.

    • Crossref
    • PubMed
    • Export Citation
  • Brown, S. L., and I. - Fen Lin. 2012. “The Gray Divorce Revolution: Rising Divorce among Middle-Aged and Older Adults, 1990–2010.” The Journals of Gerontology Series B: Psychological Sciences and Social Sciences 67 (6):731–41.

    • Crossref
    • Export Citation
  • Brown, S. L., G. R. Lee, and J. R. Bulanda. 2006. “Cohabitation among Older Adults: A National Portrait.” The Journals of Gerontology Series B: Psychological Sciences and Social Sciences 61 (2):S71–9.

    • Crossref
    • Export Citation
  • Butrica, B. A., and K. E. Smith. 2012. “The Retirement Prospects of Divorced Women.” Social Security Bulletin 72 (1):11–22.

    • PubMed
    • Export Citation
  • Couch, K. A., C. R. Tamborini, G. Reznik, and J. W. R. Phillips. 2013. “Divorce, Womens Earnings, and Retirement Over the Life Course.” In Lifecycle Events and Their Consequences: Job Loss, Family Change, and Declines in Health, edited by Kenneth, C. Mary, D. and Julie, Z., 133.

  • Dickert-Conlin, S., and C. Meghea. 2004. “The Effect of Social Security on Divorce and Remarriage Behavior.” CRR WP 2004–09, April 2004.

  • Gadalla, T. M. 2008. “Gender Differences in Poverty Rates after Marital Dissolution: A Longitudinal Study.” Journal of Divorce & Remarriage 49 (3–4):225–38.

    • Crossref
    • Export Citation
  • Gardner, J., and A. J. Oswald. 2006. “Do Divorcing Couples Become Happier by Breaking Up?.” Journal of the Royal Statistical Society: Series A (Statistics in Society) 169 (2):319–36.

    • Crossref
    • Export Citation
  • Genadek, K. R., W. A. Stock, and C. Stoddard. 2007. “No-Fault Divorce Laws and the Labor Supply of Women with and Without Children.” Journal of Human Resources 42 (1):247–74.

  • Goda, G. S., J. B. Shoven, and S. N. Slavov. 2007. “Social Security and the Timing of Divorce.” NBER Working Paper No. 13382.

    • Crossref
    • Export Citation
  • Kennedy, S., and S. Ruggles. 2014. “Breaking Up Is Hard to Count: The Rise of Divorce in the United States, 1980–2010.” Demography 51 (2):587–98.

    • Crossref
    • PubMed
    • Export Citation
  • Kleven, H. J., and M. Waseem. 2013. “Using Notches to Uncover Optimization Frictions and Structural Elasticities: Theory and Evidence from Pakistan.” The Quarterly Journal of Economics 128 (2):669–723.

  • Kopczuk, W., and D. J. Munroe. 2015. “Mansion Tax: The Effect of Transfer Taxes on the Residential Real Estate Market.” American Economic Journal. Applied Economics 7 (2):214–257.

  • Lavelle, B., and P. J. Smock. 2012. “Divorce and Womens Risk of Health Insurance Loss.” Journal of Health and Social Behavior 53 (4):413–31.

    • Crossref
    • Export Citation
  • Manoli, D. S., and A. Weber. 2011. “Nonparametric Evidence on the Effects of Financial Incentives on Retirement Decisions (No. w17320).” National Bureau of Economic Research.

    • Crossref
    • Export Citation
  • Page, M. E., and A. H. Stevens. 2004. “The Economic Consequences of Absent Parents.” Journal of Human Resources 39 (1):80–107.

    • Crossref
    • Export Citation
  • Persson, P. 2013. “Social Insurance and the Marriage Market”.

  • Peters, E. H., K. Simon, and J. R. Taber. 2014. “Marital Disruption and Health Insurance.” Demography 51 (4):1397–421.

    • Crossref
    • PubMed
    • Export Citation
  • Ratcliffe, C. E., G. P. Acs, T. Dore, and D. Z. Moskowitz. 2008. “Assessment of Survey Data for the Analysis of Marriage and Divorce at the National, State, and Local Levels.” The Urban Institute, September.

    • Crossref
    • Export Citation
  • Saez, E. 2010. “Do Taxpayers Bunch at Kink Points?.” American Economic Journal: Economic Policy 2 (3):180–212.

  • Schoen, R., and N. Standish. 2001. “The Retrenchment of Marriage: Results from Marital Status Life Tables for the United States, 1995.” Population and Development Review 27 (3):553–63.

    • Crossref
    • Export Citation
  • Shryock, H., and J. Siegel. 1973. “The Methods and Materials of Demography.” Bureau of the Census.

  • Smock, P. J., W. D. Manning, and S. Gupta. 1999. “The Effect of Marriage and Divorce on Women’s Economic Well-Being.” American Sociological Review 64 (6):794–812.

  • Solon, G., S. J. Haider, and J. Wooldridge. 2013. “What Are We Weighting for?.” No. w18859. National Bureau of Economic Research.

    • Crossref
    • Export Citation
  • Stevenson, B., and J. Wolfers. 2007. “Marriage and Divorce: Changes and Their Driving Forces.” The Journal of Economic Perspectives 21 (2):27–52.

  • U.S. Social Security Administration. 2013. Annual Statistical Supplement to the Social Security Bulletin, 2012. Washington, D.C: U.S. Government Printing Office, http://www.ssa.gov/policy/docs/statcomps/supplement/2012/oasdi.pdf.

FREE ACCESS

Journal + Issues

The B.E. Journal of Economic Analysis & Policy (BEJEAP) is an international forum for scholarship that employs microeconomics to analyze issues in business, consumer behavior and public policy. Topics include the interaction of firms, the functioning of markets, the effects of domestic and international policy and the design of organizations and institutions.

Search

  • View in gallery

    Divorces by marriage length, from the 1985 to 1995 Vital Statistics data.

  • View in gallery

    Probability of remaining married by years since the marriage began, from the 2008 to 2011 ACS.

  • View in gallery

    Probability of being widowed by years since the marriage began, from the 2008 to 2011 ACS.

  • View in gallery

    Divorces by marriage length, from the 1966 to 1974 Vital Statistics data.

  • View in gallery

    Divorces by marriage length, from the 1966 to 1974 Vital Statistics data.