The Effect of Social Security on Divorce and Remarriage Behavior* Stacy Dickert-Conlin

Michigan State University Cristian Meghea

Michigan State University

ABSTRACT

As the divorce rate increased through the 1970s, the policy climate for divorce also changed dramatically, raising the question of whether the secular trends in divorce were influenced by these policy changes or vice versa. One such policy change, passed in late 1977 in recognition of falling marriage durations, entitled divorced persons to claim auxiliary Social Security benefits on their ex-spouse’s record if the marriage lasted at least ten years, where previously 20 years of marriage were required. Using large samples of vital statistics divorce records, we consider the possibility that the financial incentives associated with a claim on auxiliary Social Security benefits discourages divorces from occurring immediately before the ten year duration is reached and encourages divorces that reached the ten year duration. We find no evidence that the divorce timing or propensities reacted in this way within the overall sample of divorces. There is suggestive evidence that older women may have delayed divorces to reach the ten-year mark. Further, we find that divorces ending after ten years do not follow longer periods of separation to reach the ten year duration. Finding little evidence that the law affected divorce propensities or timing, we consider whether providing a new claim on future Social Security benefits to all women with at least a ten-year marriage lowered remarriage probabilities, relative to women with only a nine-year marriage. Again, the incentives implicit in the law do not appear to have affected remarriage behavior. This evidence suggests that the effect of the law change was mainly redistributive and did not have the unintended consequence of changing divorce or remarriage behavior. Keywords: Marriage, Divorce, Social Security

* The authors are very grateful to participants at the American Economics Association Conference and to Hilary Hoynes for her discussion. We are also grateful to Mary Hamman for very thorough research assistance. And, thanks to Mike Conlin for his suggestions. Of course, all errors are ours.

1

"We can never insure one-hundred percent of the population against one-hundred percent

of the hazards and vicissitudes of life. But we have tried to frame a law which will give some measure of protection to the average citizen and to his family against the loss of a job and against poverty-ridden old age." Franklin D. Roosevelt’s Presidential Address, August 14, 1935.

“Dear Ann Landers: You would do your readers a great service if you called to their

attention the new Social Security law (effective Jan. 1, 1979) which …reduces from 20 years to 10 years the time that a divorced person need have been married in order to claim benefits based on the ex-spouse's earnings.” Landers, Ann. 1978. “Dear Ann Landers.” The Washington Post, April 7. Style Section: C16. “Imagine a Retirement Program that…grants people who signed divorce papers after being married nine years and eleven months hundreds of thousands of dollars less than those who waited another month to divorce.” Steuerle and Favreault, November 2002, Social Security For Yesterday’s Family.

1. INTRODUCTION

As the divorce rate increased through the 1970s, the policy climate for divorce also

changed dramatically, raising the possibility that the secular trends in divorce were influenced

by these policy changes or vice versa. One such policy change, passed in late 1977 in

recognition of falling marriage durations, entitled divorced persons to claim auxiliary Social

Security benefits on their ex-spouse’s record if the marriage lasted at least ten years, where

Social Security previously required 20 years of marriage. The law change intended to provide

a safety net for the increasing numbers of divorced women entering retirement during a time

when the divorce rate in the United States was reaching its peak. The possibility of

increasing moral hazard accompanies the expansion of any insurance program, which, in this

case, might include increased divorce rates for couples with at least ten years of marriage or

lower remarriage rates among divorced women. It is also possible that the arbitrary choice of

a marriage duration that entitles a spouse to Social Security benefits distorts behavior by

altering the timing of divorce, resulting in potential inefficiencies. This study investigates

whether lowering the duration of an eligible marriage altered divorce and separation behavior

2

and whether granting divorced women with at least a ten-year marriage a new claim on future

Social Security benefits changed their remarriage behavior.

With the evolution of divorce as a common family event in the United States,

understanding the interaction between Social Security and divorce decisions is particularly

relevant.1 Butrica and Iams (1999) find, for the 1931 to 1935 birth cohort, that 11 percent of

the women entering retirement are divorced, and they project an increase to 19 percent over

the next 20 years. In addition, divorced women are a particularly economically vulnerable

group, facing a high risk of poverty in their later years. Anzick and Weaver (2001) show that

20.4 percent of divorced elderly women are poor as compared to 4.3 percent for married

women.

There is growing empirical evidence that implicit financial incentives provided by

government programs affect propensities and timing of divorce and marriage, consistent with

economic theory (Becker, Landes, and Michael, 1977). In survey articles, Alm, Dickert-Conlin

and Whittington (1999) and Moffitt (1998) find that taxes and transfers are correlated with

marriage and divorce propensities. On the timing margin, Alm and Whittington (1997),

Sjoquist and Walker (1995), and Gelardi (1996) find that couples with high marriage income

tax penalties are more likely to delay marriage into the following tax year. However, Alm

and Whittington (1997) find no evidence that taxes influence the timing of divorce.

The literature on Social Security and marriage is more scarce, although at least two

papers find that widow(er)s respond to incentives in the Social Security (Brien, Dickert-

Conlin and Weaver, 2004) and Canadian Income Security (Baker, Hanna and Kantarevic,

2004) systems by delaying or avoiding marriage to prevent losing a claim on widow benefits.

3

Closely related to this paper, Goda, Shoven, and Slavov (2007) examine whether the

Social Security ten-year marriage rule influences the decision of married couples to delay

divorce from year nine to year ten in order to collect spousal benefits. Using the 1985

through 2003 Marital History File of the Panel Study of Income Dynamics, they find that

“vulnerable” couples, who have a large disparity in earnings between spouses and therefore

for whom the spousal benefits are more valuable, are more likely to delay divorce from year

nine to year ten compared to the control group. However, the effect is small in magnitude and

statistically insignificant.

This paper uses a different source of identification and, more importantly, focuses on

the large number of divorces occurring in the time period surrounding the 1977 law change.

Using the 1977 law change as exogenous variation in divorce incentives and a very large

sample of vital statistics divorce records, we consider the possibility that the financial

incentives associated with a claim on auxiliary Social Security benefits discouraged divorces

from occurring immediately before the ten year duration was reached and encouraged

divorces that reached the ten year duration. We find no evidence that the divorce timing or

propensities reacted in this way or that divorces ending after ten years followed longer

periods of separation to reach the ten year duration. This is consistent with existing research

that divorce timing does not respond to financial incentives. However, this is not consistent

with evidence that the timing of other demographic behaviors such as marriage, births and

1 The ratio of divorces to marriages in a given year has increased from 0.23 in 1950 to 0.49 in 1996 (2001 Statistical Abstract of the U.S., Table 68) and the number of divorced persons quadrupled between 1970 and 1996 to 18.3 million (Saluter and Lagailia, 1998).

4

even deaths respond to financial incentives, suggesting that the costs of delaying divorce are

prohibitively high.2

Finding no evidence that the law affected divorce propensities or timing, we consider

whether the law providing a new claim on future Social Security benefits to all women with at

least a ten-year marriage lowered remarriage probabilities of these women relative to women

with only a nine-year marriage. We use marital histories from the June Current Population

Survey and, again, the incentives implicit in the law do not appear to have affected remarriage

behavior. This evidence suggests that the effect of the law change was mainly redistributive

and did not have the unintended consequence of changing divorce or remarriage behavior.

Section 2 of the paper describes the divorce and remarriage rules of the Social

Security program and their predicted effects on divorce and remarriage behavior. Section 3

presents empirical results on divorce behavior; and Section 4 estimates the effect of the Social

Security rules on remarriage behavior. Section 5 concludes our study.

2. Institutional Details of the Social Security Program: Divorce and Remarriage

Retired workers receive Social Security benefits if they accumulated at least ten years

of covered earnings over their work lives (U.S. Social Security Administration, 2006).

Benefits are based upon a primary insurance amount (PIA) computed from their average

indexed monthly earnings. One may also be eligible for Social Security “auxiliary benefits”

as the spouse or ex-spouse of a covered worker. In practice, these benefits accrue almost

entirely to women (see Table 5.F1 in U.S. Social Security Administration, 2006). To be

2 A growing number of studies conclude that government incentives influence the timing of other demographic decisions such as childbirth (Dickert-Conlin and Chandra, 1999; Gans and Leigh, 2006a) and death (Gans and Leigh, 2006b; Kopczuk and Slemrod, 2003).

5

eligible as an ex-spouse, the law requires that the marriage lasted at least ten years. A

marriage lasting any amount less than ten years does not generate auxiliary benefits for the

divorced spouse. If the marriage qualifies under the ten-year rule, the benefits to the ex-

spouse are identical to spousal benefits, which are one-half of the husband’s PIA upon

reaching normal retirement age or an actuarially adjusted benefit if the claim is made at a

different age.3 Since 1983, spousal benefits become survivor benefits and are more generous

if the spouse or ex-spouse dies. Survivor benefits can be claimed at an earlier age and are

paid at a higher rate than the spousal benefits (potentially equal to 100 percent of the deceased

spouse or ex-spouse’s PIA). Eligibility for divorced spouse benefits terminate if the woman

remarries (and remains remarried), but surviving divorced spouse benefits are payable if the

remarriage occurs at age 60 or later.4 A divorced woman’s claim is unaffected by the

remarriage of her ex-husband.

Auxiliary benefits are also relevant for women who are entitled to retired-worker

benefits based on their own covered employment. After computing auxiliary benefits based

on each eligible marriage, SSA compares the woman’s retired-worker benefits with the

highest spousal or survivor benefit available. If these auxiliary benefits are higher than her

retired-worker benefits, she is considered dually entitled, and SSA supplements her worker

benefits to make them equal to the auxiliary benefits she is entitled to. If the retired-worker

benefits are larger, the woman will not receive any auxiliary benefits.

The current requirement that a marriage last at least ten years for a woman to be

eligible was passed as part of the 1977 Amendments to the Social Security Act, which were

enacted in January 1979. The main provisions of the 1977 bill addressed the financing

3 The Normal Retirement Age is 65 for individuals born in 1937 or earlier and is increasing to 67 for individuals

6

problems of the program (Myers, 1993). However, rising divorce rates in the 1960s and

1970s prompted the switch from a 20 to a ten-year minimum marriage. The change in the

Social Security benefits was designed to help the increasing number of people divorcing after

fewer years of marriage. The law applied to all existing divorced persons, not just those

divorces that occurred after the law, so all persons filing for Social Security benefits after

January 1, 1979 are eligible for the spousal benefits.

The effect of the 1977 Social Security law change on divorce may work through either

the propensity to divorce or the timing of the divorce. We argue that both whether and when

to divorce should manifest themselves similarly in observable outcomes around the arbitrary

marriage duration thresholds.5 We assume that the costs of postponing a divorce, such as

living in an unhappy marriage and deferring the search for or commitment to a new partner,

increase the further a couple is below the ten-year threshold duration. In addition, we assume

that couples waiting to reach the threshold are likely to divorce quickly upon reaching it. We

also assume that couples just below the threshold duration and considering divorce may be

less likely to turn the consideration into an actual divorce than those above the threshold

marriage duration, all else equal. Therefore, we might expect the biggest effect of the law, if

any, would be on marriages ending near the ten and 20 year thresholds. Specifically, relative

to before the law change, the divorce probability of couples with fewer than but close to ten

years of marriage may decline with a corresponding increase at ten years of marriage.

Likewise, relative to before the law change, divorce probability of couples with fewer than

born in 1960 or later. http://www.ssa.gov/OACT/ProgData/nra.html 4 Prior to 1984, the remarriage of a surviving divorced spouse terminated her benefits. 5 Clearly there are different welfare implications if a couple is deciding whether or not to divorce or simply when to divorce. With respect to behavioral responses to taxation, Slemrod (1990) suggests that timing is the highest order effect.

7

but close to 20 years of marriage may increase with a corresponding decrease at 20 years of

marriage.

The change in the law also has implications for remarriage behavior. The law gives

divorced women married at least ten years a claim on guaranteed future Social Security

benefits and therefore increases their expected gains from being single. We predict that, all

else equal, the higher future wealth makes a divorced woman in this group less likely to

remarry under the new law.

3. DIVORCE BEHAVIOR

3.1 Data and Identification Strategy

Empirically, we use the change in the 1977 law to identify an effect of changing

financial incentives on divorce behavior. We define the period after the law change as any

year after 1977. Although the law was not effective until 1979, the law passed in 1977 and

marriages that ended in 1978 would be eligible for the full benefits described in the law,

beginning in 1979. In addition, the law was publicized in 1978 in places like Ann Lander’s

(1978) newspaper column and a front-page article of the Washington Post (Rich, 1978), so we

assume that the “after” period includes the year 1978.

To consider the question of whether the law affects divorce decisions, we use a large

sample of divorce records from the Vital Statistics (VS) data for the United States. The VS,

compiled by the National Center for Health Statistics, contains information from divorce

certificates collected at the state level in the United States for every year from 1968 to 1995.6

About half of all states participated in the divorce reporting. Some states provide a random

6 Funding ended for this project after 1995.

8

sample of records and other states include the entire universe of records.7 A typical record

contains the date of divorce (month and year), the date of marriage (month and year), state of

divorce, and information on both spouses, including their age, marital, and fertility history.

We only use divorces from first marriages to avoid the issue of prior marriages generating

Social Security eligibility.

In Figures 1, 2 and 3, we show divorce counts from the 25 states that reported their

divorce records continuously from 1968 until 1995. Figure 1 shows the average divorce

counts by marriage duration in multi-year, calendar-time intervals. Figures 2 and 3 highlight

marriage durations of particular interest from Figure 1. We include 1978, the year after the

law was passed, separately, in case there is a one time effect of the law change. These figures

show that the counts of divorces rose dramatically after the 1969 to 1973 period. The count

numbers peaked in the 1979 to 1982 span and then fell slightly after that. Over time, the

distribution shifted toward shorter marriage durations.

With respect to our hypotheses, Figures 1, 2, and 3 illustrate a number of things.

There are no dramatic declines in the number of divorces at the marriage duration of nine

years, with no corresponding increase at ten years after the 1977 law change. In 1978, there

is some evidence of a relatively low number of divorces after nine years of marriage, but there

is no obvious increase in the number of divorces after ten years in subsequent years. There is

also no obvious increase in the number of 19-year marriages and a corresponding decrease the

number of 20-year marriages.

While providing a first look at the trends, the count data in the figures fail to account

for other secular trends and demographic characteristics. To account for these, we rely on

7 Each record represents one divorce and is weighted based on the sampling rate of the reporting area. The

9

regression analysis to identify any effect of the law, considering differences in the timing and

propensity of divorces before and after the law change. We continue to use the VS data,

because they provide very large sample sizes for single year marriage durations in single

calendar years and we hypothesize that the largest changes are most likely to happen very near the

marriage duration thresholds. These data also allow us to control for relevant covariates, such as

race, age, education, presence of children, and which spouse filed for divorce. We believe the

large sample sizes at single year marriage durations outweigh the disadvantage that the VS has no

information on Social Security eligibility and the sample includes only marriages that ended in

divorce, which precludes us from considering the sample of married couples (at risk for divorce)

who never divorce.8

3.2 Results

We begin by restricting our analysis to the years from 1975 to 1980 to minimize the

effect of unobservable changes in the divorce environment over time and the possibility of

dynamic selection. By dynamic selection we mean that if couples react when the law is

initially passed, there may be a selection into who is in the data later and therefore at risk to

change behavior. For example, if the law lowers the probability that individuals in nine-year

marriages ever get divorced, then they will not show up in our data as ten-year marriages at

risk to end in divorce. It is not clear which way this bias might go, but shortening the time

frame minimizes the concern. We also run separate regressions on marriage of fewer than 15

years to consider effects around the ten year threshold and on marriages of 15 to 25 years to

information regarding the divorce and the spouses is obtained from the certificate of the granted divorce. Marital history includes the number of marriages for both spouses. 8 We experimented with the June CPS, which has retrospective marital histories, but the data were very noisy at the single calendar year, single marital duration level. In addition, results are available from the authors where the dependent variable is divorce rates. However, the Census is the only data that is large enough to give reliable counts of at risk, married, women by single durations of marriage and these data are very limiting

10

consider the effects of the 20 year threshold. We run these separately because of our

assumption that marriage durations “close” to the threshold are most likely to be affected and

most equivalent in non-observable ways. To allow flexibility in our analysis around each

single marriage duration year, we expand each divorce into an observation for every marriage

year. We create the variable divorce that takes the value 0 for each year of the marriage spell

and 1 for the year when the spell ends in a divorce.

When considering marriages of fewer than 15 years of marriage, we begin with

1,035,325 divorces, which become 3,806,569 marriage-years when expanded. The results in

Figure 4 show that conditional on reaching each marriage duration, the probability of

divorcing at that duration is higher after the law change. This reinforces Figure 1 that

divorces are occurring after fewer years of marriage after the law change. However, the

pattern around ten years of marriage duration after the law change is not consistent with the

prediction that the new financial incentives discourage divorces at just under ten years of

marriage or encouraged divorces at ten or more years of marriage.

For ease of interpretation, we estimate the following linear probability model with

robust standard errors to account for observable differences between the treatment and

comparison groups9:

divorceits= β0+ β1(duration dummiesit) + β2(duration dummiesit*afterit) + γs+ δt +εist

where divorce is equal to one if the divorce ends at that duration for individual i and 0

otherwise. The vector duration dummies is a set of dummy variables for each marriage

because they are only available every 10 years and the sample sizes remain small within demographic categories that would be useful for the analysis. 9 We find similar results with probit and logistic models.

11

duration.10 The variable after is equal to one for all years after 1977, when the law changed.

The coefficients of interest are in vector β2, which captures any differential timing in divorce

after the law change. In our baseline analysis, we also include a set of state fixed effects, s ,

to capture any state characteristics that are fixed over time; one caveat with this variable is

that this is the state the divorce was filed in, so it may not represent the state the person lived

in for the duration of their marriage. Year fixed effects, t , capture the year specific effects

that are constant across individuals.11 The error term is its.

The results for this baseline specification are shown in Column 1 of Table 1 and are

consistent with the raw data. Conditional on reaching any one of the first three years of

marriage, a divorce is less likely to occur in that period than for a couple reaching the seventh

year of marriage (the omitted marriage duration dummy), but more likely to occur in other

years of marriage, confirming that the probability of divorce peaks between three and seven

years in the period before the law change.

Focusing on the interaction terms between the marriage duration and the period after

the law change, we see that the probability of divorcing at each marriage duration, conditional

on reaching it, is higher after the law change. Near the ten year marriage duration after the

law change, we see that the probability of divorcing at nine years of marriage (coefficient =

0.2303, s.e.=0.0020) increased more than the probability of divorcing at ten years of marriage

(coefficient = 0.2040, s.e.=0.0022) and this difference is significant at the one percent level (F

statistic (1,3782661) = 88.96). This result is counter to our expectation that the law would

discourage nine-year marriages and encourage ten-year marriages. Even considering the

10 For the regression analysis, we drop the observations from the expanded data where the duration is 14 years (and divorce always equals 1) because this is a perfect predictor of the dependent variable, given our selection on marriages that end in 14 or fewer years.

12

possibility that finalizing a divorce takes time after the process begins and perhaps the effect

does not show up until after the ten-year duration has passed, we see that the probability of

divorcing after nine years of marriage also increased more after the law change than the

probability of divorcing after eleven, twelve, thirteen or fourteen years of marriage.

It is possible that the observable characteristics of divorced persons are changing over

time and that these changes are masking an effect of the law. To test this, we use information

such as age, education, and number of children, which are available on some divorce

records12 to run another set of regressions. These characteristics include race of each spouse,

age of each spouse and age squared in year t, an indicator for the member(s) of the couple

who filed for divorce, how many children the couple had at the time of divorce, and the wife’s

level of education at the time of divorce. In addition, we include two variables that capture

the state laws about divorce during year t: (1) whether the state had a unilateral divorce law

in year t so that either couple could file for divorce without the other’s consent and (2)

whether the state allowed no fault divorces in year t (see Friedberg, 1998; Peters, 1986; and

Stevenson and Wolfers, 2007 for a well-developed literature on whether these laws affected

divorce propensity).13 Column (2) of Table 1 shows that the qualitative results are

unchanged. Although adding the covariates reduces the magnitude of the interaction terms,

the probability of a marriage ending after ten years does not increase relative to the

probability of a marriage ending after nine years, following the law change.

Our identification relies on whether we see differential patterns in the timing of

divorce before and after the law change. It is also possible that specific groups were more

11 We exclude after as a separate dummy because it allows us to include all single marriage duration years. 12 The missing observations are most often due to states choosing not to record these data on divorce records in all years. 13 We follow Gruber’s (2004) coding of the unilateral and no fault legislation.

13

likely to be affected by the law change than others. One hypothesis is that women who are

older are more likely to respond to the Social Security law because they are closer to filing for

Social Security benefits. Or, maybe women with children are more likely to respond because

the wives have lower labor force attachments and therefore expect to rely more on auxiliary

Social Security benefits (Hayghe, 1997). In their related study, Goda et al. (2007) directly

identify couples for whom the wife has had little labor force attachment, but our divorce

record data do not have that level of detail. To account for potentially differential effects, we

interact the duration dummies*after variables with a dummy for whether the couple has

children and, separately, a dummy for women who are over age 44. These results are in

Columns (3) and (4) of Table 1.

Column (3) shows that there is a differential effect for couples with children after the

law change. Among couples with children, the probability of divorce is higher after the law

change at the ten and higher years’ durations, compared to those without children, and the

differential is increasing as marriages pass the ten-year threshold. While this is certainly

broadly consistent with the predictions from the law, there are no economically large

deviations specifically around the ten-year threshold to suggest the law change is solely

responsible for these trends.

Column (4) shows that, overall, older women are less likely to divorce after the law

change at durations around the ten-year threshold. Noticeably, older women with only nine

years of marriage duration have significantly lower probabilities of divorcing after the law

change, relative to older women with ten-year marriages. Consistent with our predictions,

14

these results suggest that there may be a response to the law change among couples in which

the women would be claiming Social Security benefits in fewer years than younger women.14

We conduct a similar analysis for marriages around the 20-year duration and find

similar results: for the overall population of divorces, the timing of divorce is not different

after the law change in a way that is consistent with the incentives provided by the law

change.15 Selected results are in Table 2. Note that among marriages ending between 15 and

25 years, there is a marked increase in the probability of divorce at the 20-year threshold.

Like the shorter marriages we considered earlier, we see that the probability of divorcing at

each marriage duration, conditional on reaching it, is higher after the law change. After the

law change we see that the probability of divorcing at 19 years of marriage (coefficient =

0.2434, s.e.=0.0033) increased more than the probability of divorcing at 20 years of marriage

(coefficient = 0.2293, s.e.=0.0035) and this difference is significant at the one percent level (F

statistic=11.27), which is consistent with the predictions from the law change. However,

given the changes in the probability of divorce at each marriage duration after the law, these

results do not appear different from an overall shift in the timing of divorce: marriages are

ending earlier after the law change. For example, the probabilities of a divorce ending before

19 years of duration increased similarly to the probability of a divorce ending at 19 years.

With large sample sizes, the coefficients are almost all statistically different from one another,

but economically, it does not appear that there was a big shift from people waiting to attain 20

years of marriage to ending their marriages at 19 years of marriage.

14 Given that our sample only includes divorces that ended a 1st marriage, we acknowledge the fact that the wife’s age may be a proxy for other factors potentially influencing the divorce decision, such as labor force participation. 15 For the regression analysis, we drop the observations from the expanded data where the duration is 25 years (and divorce always equals 1) because this is a perfect predictor of the dependent variable, given our selection on marriages that end in 25 or fewer years.

15

3.3 Alternative outcome: months separated before divorce

Although the previous results show little effect on divorce timing, the law change may

have increased separation durations in order to reach the ten-year marriage duration. To test

this possibility, we use the self-reported information on divorce records of how many months

a couple was separated before the divorce. Using marriages of fewer than 15 years of

duration for 1975 through 1980, Figure 4 shows the mean number of months separated by

marriage duration before and after the 1977 law. The mean of months separated is higher

after the law change for all marital durations, but relative to other marriage duration, the mean

is not significantly higher among marriages that lasted exactly ten years.

To account for other characteristics that may affect separation periods, we run a

generalized linear Poisson model, where the unit of observation is the divorce. The

specification is the following:

months separatedits= β0+ β1(duration dummiesit) + β2(duration dummiesit***afterit)+ γs+ δt +εist

The dependent variable is the separation period in months. We include a full set of state

dummies, s and year dummies, t , and estimate robust standard errors ist. We are looking

for differences in marital separation periods among the marriage durations that are most likely

to be affected by the change in law so 2 is our coefficient vector of interest.

Table 3 shows the results of this regression.16 In Column (1), we have the baseline

regression with no covariates for the years 1975 through 1980. Both columns show that

separation periods increase with the length of the marriage (the omitted category of marriage

duration is marriages ending after seven years). The interaction terms of marriage duration

16 The results are not qualitatively different when we expand the sample to include marriages of longer duration, including those around the 20 year duration, or those over a longer timeframe, 1973-1982. These results are available from the authors.

16

with after the law change are all positive and statistically significant, suggesting that

separation periods increased after the law change for all marriage durations. However, the

changes are not consistent with a prediction that couples choose longer separation in order to

reach the ten year marriage duration. For example, the separation periods went up by a

higher percent for marriages ending after nine years (7.23 percent) rather than ten years (5.63

percent), which is counter to the prediction, and we reject the coefficients being equal at

standard levels ( 80.512 X ). The percent increase in separation periods for marriages

ending after 11 years is statistically and economically larger than for nine years. If the

divorce process takes time to complete, the law change might explain this. However, given

that the coefficients on marriage durations of 12 and 13 years are also much larger than those

for marriages ending after nine years, it seems more likely that in the later years of the sample

there is something besides the law change affecting all these divorces. While this evidence is

only suggestive, there is certainly no “smoking gun” that suggests couples separated but

waited to attain the ten year marriage duration.

The results on key interaction terms are similar in Columns (2), which limits the data

to 1977 and 1979 to test for an immediate reaction and Columns (3) and (4), which replicate

columns (1) and (2) with the addition of covariates. The other covariates in Columns (3) and

(4) provide some insights into the divorce process during this time period. In our sample of

divorces ending before 15 years of marriage, we find that African Americans remain

separated for 30 percent longer than white couples. The separation period is shorter when

both parties file for divorce, relative to when the wife only files. With respect to government

policy, those states with no fault divorce as an option have shorter separation period than

those that require establishing fault. Having more children and more education are negatively

17

correlated with the separation period. Many of the state dummy variables (not shown in the

table) are also statistically significant, which likely reflects variation in states’ laws about

mandated separation periods or the legal divorce environment in general.

3.4. Discussion

One possible explanation for finding little response to the law is that the financial

incentives in Social Security are too small to matter much in the divorce decision. Much of

the previous literature on the effect of taxes and transfers finds small effects of these financial

incentives on marriage behavior and, in the case of Social Security, the financial benefits of

waiting may be decades from the time the decision to divorce is being made. Perhaps the fact

that divorced women start receiving the auxiliary benefits only at retirement places these

financial benefits too far into their future. To get a sense of how large the spousal benefits

could be, consider a woman born in the cohort including 1941 through 1945, approaching the 10th

anniversary of her marriage in 1977. If she retires at the normal retirement age and receives the

predicted benefits for a divorced women of her age, her monthly benefits in 1998 dollars will be

between $659 and $1000, depending on whether her ex-spouse is alive or not (Butrica and Iams,

2000, Table 3). If these women live 17 additional years beyond the time when they begin

collecting Social Security,17 the present discounted value of their stream of spousal benefits at the

time of her divorce is between $21,370 and $28,475 in 2004 dollars. While this estimate

represents only a back of the envelope calculation, these numbers would seem to be a non-trivial

amount. We expect the dollar amounts to be more significant for those closer to retirement age,

yet we find no statistically significant differences in responses to the law change among older age

groups.

17 Life expectancy for a typical 65 year old of this generation is 19 years (Butrica and Iams, 2000), although divorced women typically have lower life expectancies.

18

The back of the envelope calculation obviously ignores uncertainty. One source of the

uncertainty may arise from the probability that women rely on these spousal benefits. If

women expect to remarry and therefore acquire a claim to new auxiliary benefits or if women

expect to rely on their own worker benefits, the expected value of the spousal benefits from

an ex-husband decrease. Both remarriage and own worker benefits are common, although it

is beyond the scope of this project to value the expectations at the time of divorce.

Descriptively, our calculation using the Current Population Survey data show that the

majority of divorced women remarry eventually and almost two thirds of the remarriages

occur within three years from divorce. Butrica and Iams (2000, Table 3) estimate that 51

percent of the divorced women retiring between 1996 and 2000 receive only their own

retired-worker benefits (another 33% are dually entitled to their own benefits and auxiliary

benefits). An additional related uncertainty is that, at the time of the divorce, it is not possible

to know the relevant dollar amounts of the potentially forgone benefit, because they are

determined by lifetime earnings of the ex-spouse.

There may also be uncertainty over the Social Security laws, which results in no

obvious responses to the law change. Casting some skepticism on the possibility that couples

didn’t know the law existed is the previously mentioned media attention it received at the

time: a nationally syndicated letter to Ann Landers (Landers, 1978) appeared in April 1978

and a front page Washington Post article (Rich, 1978) in January of the same year detailed the

law change.18

18 Most current books or online sources that advise women on the financial implications of a divorce give a clear description of the ten-year-marriage rule governing Social Security benefits. For example, “The American Bar Association Guide to Family Law” (The American Bar Association, 1996), “Divorce for Dummies” (Ventura, Reed and Reed, 1998), and “Divorce Made E-Z” (Made E-Z, 2001), all have a clear and easy to find description of the eligibility requirements for Social Security auxiliary benefits. The following web pages are a few that directly address the 10-year rule: “A Divorced Woman's Guide to Social Security” by Kerry Hannon.

19

Finally, Alm and Whittington (1997) offer a number of explanations for why they find

that the timing of marriage is related to income taxes, but the timing of divorce is not, such as

small sample sizes in the divorces and difficulty in measuring the financial costs. In our

approach, these issues are not relevant, yet our divorce results are consistent with theirs.

They propose that the most important reason for not finding a correlation between financial

incentives and divorce is that “the psychic costs of remaining married may well outweigh any

economic factors (page 237-8)”.

4. REMARRIAGE BEHAVIOR

4.1 Identification strategy

The divorced women previously married for ten years received a claim to a future

stream of income when the Social Security law changed in 1977. We use this group as our

treatment group for investigating the effect of a future claim to Social Security on remarriage.

Women married only nine years did not receive this claim to a future stream of income and so

they serve as our comparison group. We restrict our initial analysis to just nine-year versus

ten-year old marriages because we believe that these two groups are similar, conditional on

observable and unobservable characteristics. In a sensitivity test, we expand our comparison

group to marriages of eight or nine years duration and the treatment group to ten or 11 years

duration. To avoid the possibility that the 1977 amendment affected the decision of when to

divorce, and therefore determined who is in the treatment and comparison groups after the

law change, we restrict our analysis to women whose divorce occurred in 1976 and earlier.

http://parenting.ivillage.com/mom/structure/0,,36r4,00.html “Ask Kim: Social Security and Divorce” by Kimberly Lankford http://www.divorcenet.com/states/nationwide/social_security_and_divorce. Many specifically caution women who approach the 10th marital anniversary to make sure the divorce becomes final after the ten-year mark or lose substantial benefits.

20

Our prediction is that the divorced women previously married ten years remarry less under

the new SSA rules.

4.2 Data

For this analysis we pool the 1980, 1985, 1990, and 1995 June CPSs. A CPS record

has demographic information on the respondent and the spouse, and, most importantly for our

purpose, the marital and fertility history of the respondent. The CPS questionnaire asks

details on the respondent’s first, second and last marriage; the individual reports the date of

marriage, dates of separation and of divorce, as well as the reason of ending each marriage.19

For reasons of consistency across the four waves of the survey, we keep only the

women between 15 and 65 years of age. Because the law is slightly more complicated for

women married more than once, we restrict our sample to divorce spells following a first

marriage. We expand each divorce spell observation and create the variable remarried that

takes the value 0 for each year of the divorce spell and 1 for the year when the spell ends in a

remarriage, if any.

Table 4 shows that the treatment and comparison groups have relatively similar

characteristics. However, small but statistically significant demographic differences exist.

For example, marriages that ended after nine years, relative to those that ended after ten years,

are more likely to be women who are white, slightly younger, have younger children and have

lower education. Surprisingly, the average durations of the divorce spells are not statistically

different from one another.

19 One of the concerns with the Current Population Surveys is the data quality because of their retrospective nature. Goldstein (1999) argues that both marriages and divorces are underreported and there are many missing responses imputed, yet using CPS for both numerator and denominator alleviates the severity of the problem.

21

To explore the remarriage patterns around the time of the law change, Figure 5

presents the remarriage rates (hazards) for the period 1972 to 1985 (we do not restrict our

regression analysis to these years). Examining Figure 5, the law does not seem to have the

predicted effect on remarriage. If anything, the remarriage probability of a woman with a ten-

or eleven-year marriage is higher immediately following the law change, relative to the

remarriage rate of nine-year marriages.

4.3 Results

We see from Table 5 that the probability of remarriage falls after 1977. In the

comparison group of first marriages that lasted nine years, the probability falls by 7.4

percentage points. And, in the treatment group of first marriages that lasted ten years, the

probability of remarriage falls by 6.5 percentage points. In contrast to our predictions, the

probability of remarriage after the law change fell by more for the comparison group.

However, the difference in difference is only 0.9 percentage points and is not statistically

significant at standard levels. We find similar results when we expand our treatment and

comparison groups to include 11- and eight-year marriages.

Because the presence of observable differences between the treatment and control groups

might be masking any effect of the law, we estimate the probability of remarriage with the

following linear probability regression model for ease of interpretation.20

ittititit

itititit

durationdurationX

afteryearafteryearremarried

2

654'

321 *1010

20 We find similar results with a maximum-likelihood complementary log-log model. Complementary log-log analysis is an alternative to probit and logit analysis but is unlike these other estimators in that the transformation is not symmetric. This model is typically used when the non-zero outcome is rare.

22

where remarried takes the value 0 for each year of the divorce spell that the woman remains

unmarried and 1 for the year when the spell ends in a remarriage, if any; 10year indicates that

the observation is in the treatment group; after is a dummy that takes the value 1 if the year of

the observation is 1978 or later; duration measures the length of the divorce spell in years; i

indexes the divorce spells; t are year fixed effects; and it is the error term. The other

covariates in the vector X include the respondent’s age in years, a dummy for whether or not

the woman is white, the number of children six and younger, the number of children 18 and

younger, education controls, and dummies for the specific June CPS supplement the

observation was drawn from. The coefficient of interest is on the interaction term between

10year *** after, 3 , which we predict will be negative if the claim to a future stream of

income reduces the remarriage probability for eligible women.

The first column of Table 6 reports the results from the regression comparing nine-

year marriages with ten-year marriages that ended in divorce. The coefficient of interest is a

positive 0.009, which does not have the expected sign, but is not statistically significant.

Next, we expand our sample: the treatment group now contains divorced women previously

married for ten or 11 years while divorced women previously married eight or nine years

comprise the comparison group. The treatment and comparison groups are less likely to be

identical on unobservable characteristics, but we gain from an increased sample size. The key

coefficient is now 0.013 and not statistically significant.

Most of the estimates of the covariates have the expected sign in all specifications.

Older women are less likely to remarry, and the presence of minor children decreases the

probability of remarriage. Women with some college education, relative to women with a

23

high school education, also remarry less, possibly because the education gives them more

financial independence.

As specification checks, we replicated the remarriage analysis restricting the sample to

only divorces that ended on or after 1965, to minimize the effects of unobservable changes

over time and these are available in Columns 3 and 4 of Table 6. Comparing nine-year

marriages with ten-year marriages that ended in divorce, the coefficient of interest is a

positive 0.021, which does not have the expected sign, and is statistically significant at the ten

percent level. Comparing eight- or nine-year marriages with ten- or 11 -year marriages that

ended in divorce, the coefficient of interest is a positive 0.022, and is statistically significant

at the five percent level. That is, women who received a claim on a future stream of social

security as a result of the law change, are more likely to end their divorce spell.

4.4 Discussion

Our remarriage analysis suggests that a claim to future Social Security benefits did not

decrease the probability that women remarry. As in the divorce case, we expect that the

reasons why people do not respond to these incentives in the expected way may be that

women are not familiar with the law (unlikely, due to the media attention it received at the

time), the present discounted value of the benefit is low at the time of remarriage decisions, or

women are more likely to rely on their own work history than that of their ex-spouse.

5. Conclusion

Despite a change in the marriage duration necessary to qualify for the Social Security

“divorce subsidy”, from 20 to ten years of marriage, we find little evidence of behavioral

effects on divorce. Specifically, while we see no overall effects consistent with the

24

predictions from the law, older women seem to have responded as expected to the law

change, maybe because they are closer to filing for Social Security benefits. The overall

finding is consistent with work by Goda, Shoven and Slavov (2007), who ask the same

question as us, using a different source of identification and a different time period.21 It is

also consistent with the literature on taxes and the timing of divorce (Alm and Whittington,

1997). Given the increasing amount of evidence that financial incentives affect the timing of

other decisions, including the entrance to marriage, childbirth and death, it may simply be that

the timing of exiting a marriage is not sensitive to financial incentives. In addition, we find

no evidence that a future claim to Social Security benefits, as a result of a ten-year marriage,

influences the remarriage decision in the predicted way. These results suggest that the cost of

timing a divorce is prohibitive and changing the duration of marriage requirement is unlikely

to encourage divorce or discourage remarriage.

Despite lack of strong evidence on a behavior response to the arbitrary ten-year rule,

the policy is not moot. There are clearly equity issues that are difficult to reconcile with the

fact that an additional month of marriage for someone married nine years and 11 months is

the difference between a claim on Social Security and none. Work by Harrington Meyer et al.

(2006) suggests the equity issues may be even more striking: they estimate that while more

than 80 percent of white women born in the 1960s will reach retirement age having had a ten-

year marriage, only 50 percent of African American women will. This ten-year marriage

cutoff is arbitrary and, together with the increasing prevalence of women earning their own

worker benefits in Social Security, leads to question about whether the calculation of spousal

21 In their more recent data, information on the benefits of waiting until 10 years of marriage to finalize a divorce may be more readily available, but women are less likely to rely on a spouse’s income for benefits.

25

benefits is out of date in the current marriage market environment (Favreault and Steuerle,

2007).

26

References Alm, James, Stacy Dickert-Conlin and Leslie A. Whittington. 1999. “The Marriage Penalty.” Journal of Economic Perspectives. 13(3). 193-204. Alm, James and Leslie A. Whittington. 1997. “Income Taxes and The Timing of Marital

Decisions.” Journal of Public Economics. 64(2): 219-40.

The American Bar Association. 1996. “The American Bar Association Guide to Family Law: The Complete and Easy Guide to All the Laws of Marriage, Parenthood, Separation and Divorce.” Random House, Incorporated.

Anzick, Michael A. and David A. Weaver. 2001. “Reducing Poverty Among Elderly Women.” ORES Working Paper Series. Social Security Administration, Washington, D.C.

Auerbach, Alan J. 1992. “On the Design and Reform of Capital-Gains Taxation.” American

Economic Review. 82(2). 263-67. Auerbach, Alan J. and Joel Slemrod. 1997. “The Economic Effect of the Tax Reform Act of

1986.” Journal of Economic Literature. 35(2). 589-632. Baker, Michael, Emily Hanna and Jasmin Kantarevic. 2004. “The Married Widow: Marriage

Penalties Matter!” Journal of the European Economic Association. 2: 634-664. Becker, Gary S. 1973. “A Theory of Marriage: Part I.” The Journal of Political Economy.

81(4). 813-846.

Becker, Gary S. 1974. “A Theory of Marriage: Part II.” The Journal of Political Economy. 82(2). S11-S26.

Becker, Gary S., Elisabeth M. Landes and Robert T. Michael. 1977. “An Economic Analysis of

Marital Instability,” Journal of Political Economy, 85(6): 1141-87. Brien, Michael J., Stacy Dickert-Conlin and David A. Weaver. 2004. “Widows Waiting to

Wed? (Re)Marriage and Economic Incentives in Social Security Widow Benefits.” Journal of Human Resources, 39(3): 585-623.

Butrica, Barbara A. and Howard M. Iams. 2000. “Divorced Women at Retirement: Projections

of Economic Well-Being in the Near Future.” Social Security Bulletin. 63(3). 3-12. Congressional Research Service. 2001. Welfare Reform: TANF Provisions Related to Marriage

and Two-Parent Families. Washington, D.C.: Congressional Research Service. Costa, Dora L. 1997. “Displacing the Family: Union Army Pensions and Elderly Living

Arrangements.” Journal of Political Economy. 105(6): 1269-92.

27

Costa, Dora L. 1999. “A House of Her Own: Old Age Assistance and the Living Arrangements of Older Nonmarried Women.” Journal of Public Economics. 72(1). 39-59.

Dickert-Conlin, Stacy. 1999. “Taxes and Transfers: Their Effects on the Decision to End a

Marriage.” Journal of Public Economics. 73(2). 217-40. Engelhardt, Gary V., Jonathan Gruber and Cynthia D. Perry. 2002. “Social Security and

Elderly Living Arrangements.” NBER Working Paper No. 8911. Favreault, Melissa, and C. Eugene Steuerle 2007. “Social Security Spouse and Survivor

Benefits for the Modern Family.” Discussion Paper 07-01 in The Retirement Project series http://www.urban.org/url.cfm?ID=311436, March 27.

Friedberg, Leora. 1998. “Did Unilateral Divorce Raise Divorce Rates? Evidence from Panel

Data.” The American Economic Review 88(3): 608-28 Gans, Joshua and Andrew Leigh. 2006a. “Born on the First of July: An (Un)natural

Experiment in Birth Timing.” Working paper. Gans, Joshua and Andrew Leigh. 2006b. “Did the Death of Australian Inheritance Taxes

Affect Deaths?” Berkeley Electronic Press: Topics in Economic Analysis & Policy, Volume 6, Issue 1, Article 6.

Gelardi, Alexander M.G. 1996. “The Influence of Tax Law Changes on the Timing of

Marriages: A Two-Country Analysis.” National Tax Journal. 49(1). 17-30. Goldstein, Joshua R. 1999. “The Leveling of Divorce in the United States.” Demography.

36(3). 409-414. Goda, Gopi Shah, John B. Shoven, and Sita Nataraj Slavov. 2007. “Social Security and the

Timing of Divorce.” NBER Working Paper No. 13382. Gruber, Jonathan. 2004. “Is Making Divorce Easier Bad for Children? The Long Run

Implications of Unilateral Divorce.” Journal of Labor Economics, 22 (4): 799-833. Harrington Meyer, Madonna; Douglas Wolf and Christine Himes. 2006. “Declining

Eligibility for Social Security Spouse and Widow Benefits in the United States? ” Research on Aging 28: 240-60.

Hayghe, Howard. 1997. “Developments in Women’s Labor force Participation.” Monthly Labor

Review. September: 41-46. http://www.bls.gov/opub/mlr/1997/09/art6full.pdf Accessed 3/1/08.

Hoynes, Hilary. 1997. “Does Welfare Play Any Role in Female Headship Decisions?” Journal

of Public Economics. 65(2). 89-117.

28

Kopczuk, Wojciech and Joel Slemrod. 2003. “Dying to Save Taxes: Evidence from Estate Tax Returns on the Death Elasticity.” Review of Economics and Statistics, 2003, 85(2), 256-65.

Landers, Ann. 1978. “Dear Ann Landers.” The Washington Post, April 7. Style Section: C16. Made E-Z. 2001. “Divorce Made E-Z.” Made E-Z Products. McGarry, Kathleen and Robert F. Schoeni. 2000. “Social Security, Economic Growth, and the

Rise in Elderly Widows’ Independence in the Twentieth Century.” Demography. 37(2). 221-36.

Milligan, Kevin. 2002. “Subsidizing the Stork: New Evidence on Tax Incentives and Fertility.”

NBER Working Paper No. 8845. Moffitt, Robert. 1994. “Welfare Effects on Female Headship with Area Effects.” Journal of

Human Resources. 29(2). 621-36. Moffitt, Robert A. 1998. “The Effect of Welfare on Marriage and Fertility.” Robert A. Moffitt, ed.

Welfare, The Family, and Reproductive Behavior: Research Perspectives. Washington, D.C.: National Academy Press, 1998.

Myers, Robert J. 1993. Social Security. Philadelphia: University of Pennsylvania Press. Norton, Arthur J. and Louisa F. Miller. 1992. “Marriage, Divorce, and Remarriage in the 1990s.”

Current Population Reports: Special Studies. Series P-23, No. 180. Washington, D.C.: U.S. Census Bureau.

H. Elizabeth Peters. 1986. “Marriage and Divorce: Informational Constraints and Private

Contracting.” The American Economic Review 76(3): 437-454. Phillips, Patricia and George Mair. 1995. “Divorce: A Woman's Guide to Getting a Fair Share.”

Peterson's. Powers, Elizabeth T. and David Neumark. 2001. The Supplemental Security Income Program

and Incentives to Take Up Social Security Early Retirement: Empirical Evidence from Matched SIPP and Social Security Administration Files.” NBER Working Paper No. 8670.

Saluter, Arlene and Terry Lagaila. “Marital Status and Living Arrangements: March 1996”

United States Census Bureau Current Population Report P20-426 http://www.census.gov/prod/3/98pubs/p20-496.pdf.

Rich, Spencer. 1978. “Hill Widens Old-Age Aid For Women; Social Security Changes Reflect

Women's Wider Role.” Washington Post January 10, First Section; A1. Sjoquist, David L. and Mary Beth Walker. 1995. “The Marriage Tax and the Rate and Timing of

Marriage.” National Tax Journal. 48(4). 547-58.

29

Slemrod, Joel. 1990. “The Economic Impact of the Tax Reform Act of 1986.” Joel Slemrod, ed.

Do Taxes Matter? The Impact of the Tax Reform Act of 1986. Cambridge, Mass.: MIT Press, 1-12.

Stevenson, Betsey and Justin Wolfers, 2007. “Marriage and Divorce: Changes and their

Driving Forces.” Journal of Economic Perspectives, 21(2) 27-52. U.S. Census Bureau. 2000. Statistical Abstract.

http://www.census.gov/prod/2001pubs/statab/sec20.pdf U.S. Census Bureau. 2001. Statistical Abstract.

http://www.census.gov/prod/2002pubs/01statab/stat-ab01.html U.S. Congress, Committee on Ways and Means. 2000. 2000 Green Book. Washington, D.C.: U.S.

Government Printing Office. U.S. Social Security Administration. 2006. Annual Statistical Supplement to the Social Security

Bulletin,2005 . Washington, D.C.: U.S. Government Printing Office. http://www.ssa.gov/policy/docs/statcomps/supplement/2005/5f.pdf

Ventura, John and Mary Reed Wiley. 1998. Divorce for Dummies. John & Sons, Inc. Whittington, Leslie A. and James Alm. 1997. “Till Death or Taxes Do Us Part: The Effect of

Income Taxation on Divorce.” Journal of Human Resources. 32(2). 388-412.

30

Table 1 Linear Probability Model for Marriages lasting fewer than 15 years

Dependent Variable: 1 if marriage ends in Divorce

(1) (2) (3) (4) Duration of <1 year -0.1081*** -0.1774*** -0.1778*** -0.1788*** (0.0010) (0.0011) (0.0011) (0.0011) Duration of 1 year -0.0532*** -0.1162*** -0.1165*** -0.1170*** (0.0011) (0.0011) (0.0011) (0.0011) Duration of 2 years -0.0205*** -0.0749*** -0.0751*** -0.0751*** (0.0011) (0.0011) (0.0011) (0.0011) Duration of 3 years -0.0020* -0.0461*** -0.0463*** -0.0459*** (0.0011) (0.0011) (0.0011) (0.0011) Duration of 4 years 0.0035*** -0.0294*** -0.0296*** -0.0289*** (0.0012) (0.0012) (0.0012) (0.0012) Duration of 5 years 0.0033*** -0.0178*** -0.0180*** -0.0173*** (0.0012) (0.0012) (0.0012) (0.0012) Duration of 6 years 0.0022* -0.0079*** -0.0080*** -0.0075*** (0.0012) (0.0012) (0.0012) (0.0012) Duration of 8 years 0.0035*** 0.0120*** 0.0121*** 0.0114*** (0.0013) (0.0013) (0.0013) (0.0013) Duration of 9 years 0.0111*** 0.0266*** 0.0268*** 0.0252*** (0.0014) (0.0014) (0.0014) (0.0014) Duration of 10 years 0.0380*** 0.0594*** 0.0596*** 0.0570*** (0.0017) (0.0017) (0.0017) (0.0017) Duration of 11 years 0.0799*** 0.1066*** 0.1069*** 0.1030*** (0.0020) (0.0020) (0.0020) (0.0020) Duration of 12 years 0.1540*** 0.1856*** 0.1859*** 0.1807*** (0.0025) (0.0025) (0.0025) (0.0025) Duration of 13 years 0.3138*** 0.3501*** 0.3504*** 0.3437*** (0.0035) (0.0035) (0.0035) (0.0035) Duration of <1 year*after1977 0.1582*** 0.1433*** 0.1536*** 0.1422*** (0.0011) (0.0011) (0.0014) (0.0011) Duration of 1 years*after1977 0.2120*** 0.1982*** 0.2248*** 0.1975*** (0.0013) (0.0013) (0.0017) (0.0013) Duration of 2 years*after1977 0.2361*** 0.2226*** 0.2477*** 0.2222*** (0.0014) (0.0014) (0.0018) (0.0014) Duration of 3 years*after1977 0.2403*** 0.2263*** 0.2438*** 0.2258*** (0.0015) (0.0015) (0.0020) (0.0015) Duration of 4 years*after1977 0.2494*** 0.2337*** 0.2462*** 0.2331*** (0.0015) (0.0015) (0.0021) (0.0015) Duration of 5 years*after1977 0.2517*** 0.2344*** 0.2430*** 0.2339*** (0.0016) (0.0016) (0.0023) (0.0016) Duration of 6 years*after1977 0.2489*** 0.2310*** 0.2331*** 0.2304*** (0.0017) (0.0017) (0.0026) (0.0017) Duration of 7 years*after1977 0.2511*** 0.2336*** 0.2222*** 0.2331*** (0.0017) (0.0017) (0.0029) (0.0017) Duration of 8 years*after1977 0.2419*** 0.2261*** 0.2066*** 0.2257*** (0.0018) (0.0018) (0.0032) (0.0018) Duration of 9 years*after1977 0.2303*** 0.2171*** 0.1895*** 0.2172*** (0.0020) (0.0020) (0.0037) (0.0020) Duration of 10 years*after1977 0.2040*** 0.1937*** 0.1534*** 0.1935***

31

(0.0022) (0.0022) (0.0043) (0.0022) Duration of 11 years*after1977 0.1751*** 0.1662*** 0.1010*** 0.1660*** (0.0026) (0.0026) (0.0051) (0.0026) Duration of 12 years*after1977 0.1495*** 0.1416*** 0.0729*** 0.1415*** (0.0033) (0.0033) (0.0065) (0.0033) Duration of 13 years*after1977 0.1498*** 0.1419*** 0.0488*** 0.1419*** (0.0045) (0.0045) (0.0091) (0.0045) Age of Wife (years) -0.0009*** -0.0018*** -0.0065*** (0.0003) (0.0003) (0.0003) Age of Wife2 (years) -0.0000*** -0.0000 0.0001*** (0.0000) (0.0000) (0.0000) Age of Wife missing -0.0108** -0.0237*** -0.0768*** (0.0042) (0.0042) (0.0045) Age of Husband (years) -0.0016*** -0.0019*** -0.0017*** (0.0002) (0.0002) (0.0002) Age of Husband2 0.0000*** 0.0000*** 0.0000*** (0.0000) (0.0000) (0.0000) Age of Husband missing 0.0052 -0.0001 0.0030 (0.0039) (0.0039) (0.0039) Wife is African American -0.0200*** -0.0199*** -0.0199*** (0.0021) (0.0021) (0.0021) Wife is Other Race -0.0162*** -0.0161*** -0.0167*** (0.0021) (0.0021) (0.0021) Race of wife is missing -0.0025 -0.0025 -0.0025 (0.0042) (0.0042) (0.0042) Husband is African American 0.0043** 0.0048** 0.0044** (0.0021) (0.0021) (0.0021) Husband is Other Race 0.0114*** 0.0112*** 0.0116*** (0.0023) (0.0023) (0.0023) Race of husband is missing -0.0043 -0.0041 -0.0043 (0.0042) (0.0042) (0.0042) Husband filed for divorce -0.0176*** -0.0178*** -0.0176*** (0.0004) (0.0004) (0.0004) Both filed for divorce 0.0096*** 0.0092*** 0.0097*** (0.0017) (0.0017) (0.0017) Other than husband or wife filed for divorce -0.0168 -0.0171 -0.0174 (0.0171) (0.0171) (0.0171) Missing who filed for divorce 0.0089*** 0.0086*** 0.0089*** (0.0021) (0.0021) (0.0021) One child -0.0598*** -0.0516*** -0.0605*** (0.0005) (0.0006) (0.0005) Two children -0.1271*** -0.1225*** -0.1276*** (0.0006) (0.0006) (0.0006) More than two children -0.1575*** -0.1546*** -0.1579*** (0.0008) (0.0008) (0.0008) Number of Children Missing -0.0221*** -0.0188*** -0.0224*** (0.0022) (0.0022) (0.0022) High school degree -0.0271*** -0.0278*** -0.0270*** (0.0006) (0.0006) (0.0006) Some College -0.0437*** -0.0447*** -0.0437*** (0.0007) (0.0007) (0.0007) College Degree or more -0.0680*** -0.0682*** -0.0690*** (0.0008) (0.0008) (0.0008)

32

Missing Education 0.0515*** 0.0510*** 0.0513*** (0.0011) (0.0011) (0.0011) State has no fault divorce law -0.0335*** -0.0328*** -0.0334*** (0.0044) (0.0044) (0.0044) State has unilateral divorce law 0.0024 0.0033 0.0025 (0.0031) (0.0031) (0.0031) wife>44 years old -0.1131*** (0.0030) Duration of <1 year*after 1977*wife>44 years old 0.1170*** (0.0112) Duration of 1 years*after 1977*wife>44 years old 0.0486*** (0.0120) Duration of 2 years*after 1977*wife>44 years old 0.0037 (0.0125) Duration of 3 years*after 1977*wife>44 years old -0.0078 (0.0127) Duration of 4 years*after 1977*wife>44 years old -0.0007 (0.0127) Duration of 5 years*after 1977*wife>44 years old -0.0218* (0.0118) Duration of 6 years*after 1977*wife>44 years old -0.0079 (0.0119) Duration of 7 years*after 1977*wife>44 years old -0.0214* (0.0121) Duration of 8 years*after 1977*wife>44 years old -0.0322*** (0.0121) Duration of 9 years*after 1977*wife>44 years old -0.0607*** (0.0118) Duration of 10 years*after 1977*wife>44 years old -0.0357*** (0.0122) Duration of 11 years*after 1977*wife>44 years old -0.0320** (0.0137) Duration of 12 years*after 1977*wife>44 years old -0.0413*** (0.0159) Duration of 13 years*after 1977*wife>44 years old -0.0438** (0.0216) Duration of <1 year*after 1977*couple has children -0.0289*** (0.0015) Duration of 1 years*after 1977*couple has children -0.0658*** (0.0019) Duration of 2 years*after 1977*couple has children -0.0537*** (0.0021) Duration of 3 years*after 1977*couple has children -0.0337*** (0.0022) Duration of 4 years*after 1977*couple has children -0.0219*** (0.0024) Duration of 5 years*after 1977*couple has children -0.0139*** (0.0025) Duration of 6 years*after 1977*couple has children -0.0031 (0.0028) Duration of 7 years*after 1977*couple has children 0.0159*** (0.0031) Duration of 8 years*after 1977*couple has children 0.0254***

33

(0.0034) Duration of 9 years*after 1977*couple has children

0.0342*** (0.0038)

Duration of 10 years*after 1977*couple has children

0.0481*** (0.0043)

Duration of 11 years*after 1977*couple has children 0.0759*** (0.0051) Duration of 12 years*after 1977*couple has children 0.0787*** (0.0065) Duration of 13 years*after 1977*couple has children 0.1050*** (0.0090) Constant 0.4233*** 0.5124*** 0.5272*** 0.5821*** (0.0076) (0.0087) (0.0087) (0.0089) Observations 3782725 3782725 3782725 3782725 R-squared 0.0751 0.0945 0.0954 0.0949 Notes: Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. The regressions include state and year dummies. The omitted marriage duration is 7 year marriages. The omitted race categories are white. The omitted category for filer is that the wife filed for divorce. No children and education less than high school are the omitted categories.

Source: Vital Statistics various years. Standard errors in parentheses.

34

Table 2 Selected Coefficients in Linear Probability Model for marriages ending in 15 to 24 Years

Dependent Variable: 1 if marriage ends in Divorce (1) (2) Duration of 16 years -0.0010 0.0029 (0.0019) (0.0019) Duration of 17 years 0.0025 0.0119*** (0.0019) (0.0019) Duration 18 of years -0.0013 0.0154*** (0.0020) (0.0020) Duration 19 of years -0.0002 0.0221*** (0.0020) (0.0021) Duration of 20 years 0.0281*** 0.0536*** (0.0023) (0.0024) Duration of 21 years 0.0537*** 0.0803*** (0.0026) (0.0027) Duration of 22 years 0.0982*** 0.1253*** (0.0031) (0.0033) Duration of 23 years 0.1675*** 0.1948*** (0.0040) (0.0041) Duration of 24 years 0.3302*** 0.3565*** (0.0053) (0.0055) Duration of 15 years*after1977 0.2480*** 0.2435*** (0.0029) (0.0029) Duration of 16 years*after1977 0.2450*** 0.2351*** (0.0030) (0.0030) Duration of 17 years*after1977 0.2364*** 0.2242*** (0.0030) (0.0030) Duration 18 of years*after1977 0.2449*** 0.2326*** (0.0031) (0.0031) Duration 19 of years*after1977 0.2434*** 0.2353*** (0.0033) (0.0032) Duration of 20 years*after1977 0.2293*** 0.2261*** (0.0035) (0.0034) Duration of 21 years*after1977 0.2040*** 0.2072*** (0.0038) (0.0038) Duration of 22 years*after1977 0.1760*** 0.1830*** (0.0044) (0.0043) Duration of 23 years*after1977 0.1544*** 0.1634*** (0.0054) (0.0053) Duration of 24 years*after1977 0.1544*** 0.1641*** (0.0071) (0.0071) Constant 0.4013*** 0.1726*** (0.0185) (0.0270) Observations 721728 721728 R-squared 0.0653 0.0909 Notes: Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. The regressions include state and year dummies. The omitted marriage duration is 15 year marriages. The omitted race categories are white. The omitted category for filer is that the wife filed for divorce. No children and education less than high school are the omitted categories. Column (2) includes the covariates shown in Column (2) of Table 1.

Source: Vital Statistics

35

Table 3 GLM Poisson Model

Dependent Variable: Months of Separation (1) (2) (3) (4) Duration of <1 year -1.4748*** -1.5210*** -1.6366*** -1.6696*** (0.0100) (0.0160) (0.0105) (0.0169) Duration of 1 year -0.8541*** -0.8835*** -1.0111*** -1.0188*** (0.0078) (0.0125) (0.0080) (0.0129) Duration of 2 years -0.5595*** -0.5872*** -0.6957*** -0.6963*** (0.0075) (0.0120) (0.0075) (0.0120) Duration of 3 years -0.3810*** -0.3955*** -0.5000*** -0.4956*** (0.0076) (0.0120) (0.0074) (0.0117) Duration of 4 years -0.2538*** -0.2550*** -0.3470*** -0.3473*** (0.0078) (0.0123) (0.0074) (0.0117) Duration of 5 years -0.1601*** -0.1699*** -0.2151*** -0.2225*** (0.0081) (0.0128) (0.0076) (0.0120) Duration of 6 years -0.0691*** -0.0779*** -0.0929*** -0.0962*** (0.0085) (0.0137) (0.0079) (0.0126) Duration of 8 years 0.0597*** 0.0617*** 0.0827*** 0.0802*** (0.0098) (0.0154) (0.0089) (0.0140) Duration of 9 years 0.0958*** 0.1136*** 0.1416*** 0.1495*** (0.0106) (0.0167) (0.0096) (0.0151) Duration of 10 years 0.1644*** 0.2040*** 0.2292*** 0.2670*** (0.0114) (0.0181) (0.0102) (0.0162) Duration of 11 years 0.1777*** 0.1691*** 0.2647*** 0.2631*** (0.0122) (0.0195) (0.0110) (0.0175) Duration of 12 years 0.1906*** 0.2102*** 0.3017*** 0.3205*** (0.0130) (0.0206) (0.0119) (0.0189) Duration of 13 years 0.2412*** 0.2195*** 0.3571*** 0.3406*** (0.0142) (0.0227) (0.0128) (0.0203) Duration of 14 years 0.2986*** 0.2971*** 0.4205*** 0.4212*** (0.0152) (0.0241) (0.0137) (0.0220) Duration of <1 year*after1977 0.0061 0.0200 0.0000 0.0155 (0.0109) (0.0169) (0.0113) (0.0180) Duration of 1 year ***after 1977 0.0179** 0.0276*** 0.0286*** 0.0321*** (0.0072) (0.0098) (0.0071) (0.0100) Duration of 2 years*after1977 0.0141** 0.0239*** 0.0298*** 0.0281*** (0.0067) (0.0087) (0.0065) (0.0087) Duration of 3 years*after1977 0.0191*** 0.0023 0.0391*** 0.0192** (0.0069) (0.0090) (0.0066) (0.0088) Duration of 4 years*after1977 0.0368*** 0.0120 0.0490*** 0.0351*** (0.0073) (0.0097) (0.0069) (0.0092) Duration of 5 years*after1977 0.0588*** 0.0469*** 0.0471*** 0.0438*** (0.0079) (0.0109) (0.0074) (0.0102) Duration of 6 years*after1977 0.0690*** 0.0711*** 0.0393*** 0.0307*** (0.0086) (0.0123) (0.0079) (0.0114) Duration of 7 years*after1977 0.0559*** 0.0403*** 0.0315*** 0.0225* (0.0095) (0.0138) (0.0087) (0.0127) Duration of 8 years*after1977 0.0524*** 0.0317** 0.0301*** 0.0218 (0.0105) (0.0157) (0.0095) (0.0142) Duration of 9 years*after1977 0.0723*** 0.0347** 0.0522*** 0.0306** (0.0117) (0.0176) (0.0105) (0.0156) Duration of 10 years*after1977 0.0563*** 0.0087 0.0438*** 0.0018

36

(0.0128) (0.0196) (0.0112) (0.0172) Duration of 11 years*after1977 0.0782*** 0.0897*** 0.0652*** 0.0745*** (0.0142) (0.0221) (0.0125) (0.0193) Duration of 12 years*after1977 0.0947*** 0.0744*** 0.0788*** 0.0715*** (0.0153) (0.0243) (0.0137) (0.0215) Duration of 13 years*after1977 0.0842*** 0.0944*** 0.0678*** 0.0730*** (0.0170) (0.0271) (0.0149) (0.0234) Duration of 14 years*after1977 0.0612*** 0.0520* 0.0535*** 0.0501* (0.0184) (0.0292) (0.0161) (0.0257) Age of Wife (years) -0.0101*** -0.0093*** (0.0019) (0.0032) Age of Wife2 (years) 0.0002*** 0.0001*** (0.0000) (0.0000) Age of Wife missing -0.5781*** -0.5587*** (0.0421) (0.0686) Age of Husband (years) -0.0027** 0.0000 (0.0012) (0.0020) Age of Husband2 0.0000*** 0.0000 (0.0000) (0.0000) Age of Husband missing 0.3731*** 0.4439*** (0.0340) (0.0539) Wife is African American 0.3218*** 0.3349*** (0.0113) (0.0186) Wife is Other Race 0.1357*** 0.1260*** (0.0121) (0.0203) Race of wife is missing 0.1527*** 0.1427*** (0.0266) (0.0435) Husband is African American 0.3412*** 0.3560*** (0.0112) (0.0184) Husband is Other Race 0.2119*** 0.2497*** (0.0127) (0.0211) Race of husband is missing 0.2570*** 0.2781*** (0.0265) (0.0433) Husband filed for divorce 0.0497*** 0.0451*** (0.0026) (0.0042) Both filed for divorce -0.3503*** -0.3130*** (0.0123) (0.0219) Other than husband or wife filed for divorce

-0.1672 -0.6334***

(0.1087) (0.2058) Missing who filed for divorce -0.0260** -0.0315 (0.0126) (0.0204) One child -0.0978*** -0.1036*** (0.0030) (0.0050) Two children -0.3808*** -0.3914*** (0.0039) (0.0064) More than two children -0.4568*** -0.4626*** (0.0056) (0.0094) Number of Children Missing -0.0505*** -0.0504** (0.0114) (0.0204) High school degree -0.2855*** -0.2951*** (0.0033) (0.0054) Some College -0.3640*** -0.3575*** (0.0041) (0.0068)

37

College Degree or more -0.4303*** -0.4313*** (0.0045) (0.0074) Missing Education -0.1662*** -0.1844*** (0.0061) (0.0100) State has no fault divorce law -0.1161* -0.1722** (0.0639) (0.0852) State has unilateral divorce law 0.0835*** 0.0270 (0.0180) (0.0312) Constant 0.5997*** 2.2039*** 3.1941*** 3.1842*** (0.0120) (0.0777) (0.0481) (0.0824) Observations 603163 225273 603163 225273 Years of Data 1975-1980 1977 and 1979 1975-1980 1977 and 1979 Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%, The Regression also includes state and year dummies. The omitted marriage duration is 7 year marriages. The omitted race categories are white. The omitted category for filer is that the wife filed for divorce. No children and education less than high school are the omitted categories.

38

Table 4 Means and Standard Deviations of First Divorces beginning before 1977

9 year

marriages 10 year

marriages 8 and 9 year marriages

10 and 11 year

marriages Percent White (1 if white, 0 otherwise) 0.822

(0.382) 0.804

(0.397) 0.810

(0.392) 0.797

(0.403) Age (in years) 37.042

(7.653) 38.029 (7.678)

36.418 (7.773)

38.411 (7.512)

Number of Children younger than 6 0.671 (1.114)

0.613 (1.114)

0.718 (1.149)

0.577 (1.079)

Number of Children younger than 18 2.555 (1.878)

2.631 (2.035)

2.530 (1.888)

2.652 (2.023)

Percent with less than high-school 0.214 (0.410)

0.198 (0.398)

0.204 (0.403)

0.227 (0.419)

Percent with more than high-school 0.550 (0.498)

0.606 (0.489)

0.569 (0.495)

0.584 (0.493)

Duration of divorce spell (in years) 7.693 (6.572)

7.736 (6.636)

7.793 (6.800)

7.571 (6.526)

Percent in 1985 CPS 0.244 (0.429)

0.293 (0.455)

0.258 (0.438)

0.275 (0.447)

Percent in 1990 CPS 0.231 (0.422)

0.250 (0.433)

0.231 (0.422)

0.245 (0.430)

Percent in 1995 CPS 0.214 (0.410)

0.191 (0.393)

0.207 (0.406)

0.221 (0.415)

N 6836 6489 14107 11950

Source: Authors’ calculations from the 1980, 1985, 1990 and 1995 June CPSs.

39

Table 5 Remarriage Probability for First Divorces beginning before 1977

Pre-1977

amendment (1)

Post-1977 amendment

(2)

Difference

(2)-(1)

Difference-in-differences (Treatment-Comparison)

Panel A

A. Comparison group 9- year marriages (n=6836)

0.128 (0.005)

0.054 (0.005)

-0.074*

(0.008)

B. Treatment group 10-year marriages (n=6489)

0.123 (0.005)

0.058 (0.005)

-0.065*

(0.008) 0.009

(0.011)

Panel B

C. Comparison group 8 and 9- year marriages (n=14107)

0.129 (0.003)

0.059 (0.003)

-0.070* (0.006)

D. Treatment group 10 and 11-year old marriages (n=11950)

0.122 (0.004)

0.065 (0.004)

-0.057*

(0.006) 0.013

(0.008)

Source: Authors’ calculations from the 1980, 1985, 1990 and 1995 June CPSs. Standard errors in parentheses.

40

Table 6 Linear Probability

Dependent variable: Remarried = 1 First Divorces beginning before 1977

Treatment: 10 years

of marriage Control: 9 years of

marriage (all divorces)

Treatment: 10 and 11 years of marriage

Control: 8 and 9 years of marriage

(all divorces)

Treatment: 10 years of marriage Control: 9 years

of marriage (divorces after

1964)

Treatmen11 yemar

Control:years of

(divorc19

10-year marriages 0.005 (0.007)

0.008 (0.005)

-0.005 (0.010)

-0.(0.0

After law change 0.184* (0.0018)

0.189* (0.013)

-0.086 (0.075)

-0.(0.0

(10-year marriages)***(After law change)

0.009 (0.010)

0.011 (0.007)

0.021*** (0.013)

0.0(0.0

White (1 if white, 0 otherwise) 0.036* (0.006)

0.037* (0.004)

0.033* (0.007)

0.(0.0

Age (in years) -0.007* (0.001)

-0.007* (0.001)

-0.007* (0.001)

-0(0.0

Number of Children younger than 6 0.004 (0.003)

0.002 (0.002)

0.004 (0.004)

0.0(0.0

Number of Children younger than 18 -0.006* (0.002)

-0.005* (0.001)

-0.005** (0.002)

-0.(0.0

Less than high-school (dummy)

-0.001 (0.009)

-0.0001 (0.006)

-0.006 (0.012)

0.0(0.0

More than high-school (dummy) -0.016** (0.008)

-0.012*** (0.005)

-0.021** (0.009)

-0.0(0.0

Duration of divorce spell (in years) -0.004* (0.001)

-0.005* (0.001)

-0.002 (0.003)

-0.0(0.0

Duration of divorce spell squared 0.0002* (0.00004)

0.0002* (0.00003)

0.0001 (0.0001)

0.0(0.00

1985 June CPS Observation (dummy) -0.002 (0.007)

-0.009 (0.006)

-0.006 (0.011)

-0.0(0.0

1990 June CPS Observation (dummy) -0.007 (0.009)

-0.012*** (0.006)

-0.018 (0.012)

-0(0.0

1995 June CPS Observation (dummy) 0.003 (0.009)

-0.001 (0.006)

-0.007 (0.012)

-0.(0.0

Constant 0.161* (0.021)

0.144* (0.014)

0.444* (0.076)

0.(0.0

Source: Authors’ calculations from the 1980, 1985, 1990 and 1995 June CPSs. Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. We cluster on individuals present in the data more than once. The regression also includes year dummies. 1980 June CPS, no children, and high school education are the omitted categories.

41

Figure 1 Average Counts of divorces by marriage duration and calendar years

0

5000

10000

15000

20000

25000

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Marriage Duration in Years

Ave

rage

Num

ber o

f Div

orce

s

69-73 74-77 78 79-82 83-87 88-92 93-95

See Figure 2

See Figure 3

Sources: Vital Statistics divorce records, 25 states continuously reporting over the period 1968-1995

42

Figure 2 Average Counts of divorces by marriage duration and calendar years

8000

900010000

1100012000

13000

1400015000

1600017000

18000

7 8 9 10 11

Marriage Duration

Ave

rage

Num

ber

of D

ivor

ce

74-77 78 79-82 83-87 88-92 93-95

Sources: Vital Statistics divorce records, 25 states continuously reporting over the period 1968-1995

Figure 3

Average Counts of divorces by marriage duration and calendar years

3000

3500

4000

4500

5000

5500

6000

6500

18 19 20 21 22

Duration of Marriage

Ave

rage

Num

ber

of D

ivor

ce

74-77 78 79-82 83-87 88-92 93-95

Sources: Vital Statistics divorce records, 25 states continuously reporting over the period 1968-1995

43

Figure 3

0.1

.2.3

.4.5

.6.7

.8.9

1pr

oba

bilit

y of

div

orc

e

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Source: U.S. Vital Statistics

by marriage duration and time periodProbability of Divorce

1977 and Before After 1978

Figure 4

05

1015

2025

30A

vera

ge M

ont

hs o

f Sep

ara

tion

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Source: U.S. Vital Statistics

by marriage duration and time periodAverage Months of Separation

1977 and Before After 1978

44

Figure 5 Remarriage rates by duration of previous marriage

First marriages that ended in divorce in 1976 or earlier

Source: 1980, 1985, 1990, and 1995 June supplements of the Current Population Survey.

0

0.05

0.1

0.15

0.2

0.25

72 73 74 75 76 77 78 79 80 81 82 83 84 85

Year

Rem

arri

age

Rat

e

8 9 10 11

Law Passed 12/77 Law Enacted 1/79

Duration of Previous Marriage in Years