NBER WORKING PAPER SERIES
BARGAINING IN THE SHADOW OF THE LAW:
DIVORCE LAWS AND FAMILY DISTRESS
Betsey Stevenson
Justin Wolfers
Working Paper 10175
http://www.nber.org/papers/w10175
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
December 2003
This project has drawn on the advice of many generous friends and colleagues, including Olivier Blanchard,
Margaret Brinig, David Cutler, Tom Dee, David Ellwood, Leora Friedberg, Ed Glaeser, Claudia Goldin,
Caroline Minter Hoxby, Sandy Jencks, Alan Krueger, Steve Levitt, Jeff Miron, Katherine Newman, Robert
Putnam, David Weiman, Julie Wilson and seminar participants at Harvard, the MacArthur Network on
Inequality and Social Interactions, Stanford, Michigan, Princeton, LSE, Berkeley, Columbia, Yale,
Melbourne and the Society of Labor Economists. Special thanks goes to Larry Katz for his guidance
throughout the project. We have also benefited from the excellent research assistance of Eric Klotch, Amalia
Miller and Jason Grissom. Remaining errors are our own. We are grateful to the MacArthur Foundation and
the Social Science Research Council for funding for this project. The views expressed herein are those of the
authors and not necessarily those of the National Bureau of Economic Research.
©2003 by Betsey Stevenson and Justin Wolfers. All rights reserved. Short sections of text, not to exceed two
paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given
to the source.
Bargaining in the Shadow of the Law: Divorce Laws and Family Distress
Betsey Stevenson and Justin Wolfers
NBER Working Paper No. 10174
December 2003
JEL No. D1, I1, I3, J1, K1
ABSTRACT
Over the past thirty years changes in divorce law have significantly increased access to divorce. The
different timing of divorce law reform across states provides a useful quasi-experiment with which
to examine the effects of this change. We analyze state panel data to estimate changes in suicide,
domestic violence, and spousal murder rates arising from the change in divorce law. Suicide rates
are used as a quantifiable measure of wellbeing, albeit one that focuses on the extreme lower tail of
the distribution. We find a large, statistically significant, and econometrically robust decline in the
number of women committing suicide following the introduction of unilateral divorce. No
significant effect is found for men. Domestic violence is analyzed using data on both family conflict
resolution and intimate homicide rates. The results indicate a large decline in domestic violence for
both men and women in states that adopted unilateral divorce. We find suggestive evidence that
unilateral divorce led to a decline in females murdered by their partners, while the data revealed no
discernible effects for men murdered. In sum, we find strong evidence that legal institutions have
profound real effects on outcomes within families.
Betsey Stevenson
Justin Wolfers
Graduate School of Business
Stanford University
518 Memorial Way
Stanford, CA 94305
and NBER
1
1. Introduction
In 1969, then Governor Ronald Reagan signed a bill creating unilateral divorce in
California. This legislative change was one of the first in a series that increased access to divorce
across the nation. In the ensuing decade, most states followed California’s lead; although the
specific family law reforms varied in each state, the end result was legislation that allowed
unilateral divorce. In other words, in many states it became possible for a married person to seek
the dissolution of their marriage without the consent of their spouse.
At the time, the legal changes that occurred were thought of more as a matter of
procedural policy refinement, rather than a matter of social policy.
1
Despite the lack of intent,
this legislative change has been an important force in altering social norms and perceptions about
marriage and family. Consequently, divorce law has become an issue of social concern, with
some states in recent years revisiting their earlier reforms, asking if perhaps they went too far.
Unfortunately, the recent debate over tightening access to divorce is occurring with little
knowledge of the effects of the initial changes on adult well-being.
2
The existing work directly examining the effects of divorce law changes suggests that the
reforms begun in 1969 may have caused divorce rates to rise.
3
Further, divorced people are
known to exhibit a range of negative health and lifestyle characteristics, and the financial position
of women typically deteriorates following a divorce.
4
These two facts have led some to argue
that increasing access to divorce decreases well-being. However, there are several reasons why
these arguments are misleading. First, the estimated correlation between divorce and poor
1
Historical accounts of this legislative movement indicate that it was catalyzed by reformers who were
interested in preserving the integrity of the legal system. The courts had become filled with cases involving
fraudulent charges of adultery and abuse as spouses attempted to divorce when the state’s laws did not
provide for divorce any other way. Jacob (1988).
2
For the effects on children, see Gruber (2000).
3
Friedberg (1998), although Wolfers (2003) argues that these effects on the divorce rate were temporary.
4
Holden and Smock (1991.
2
outcomes is unlikely to reflect a purely causal relationship.
5
Second, while divorce might be
financially deleterious for women, introspection suggests that there are likely to be offsetting non-
financial benefits (for at least one of the spouses). Further, there exists an important difference
between the average divorce observed and the marginal divorces that are enabled by unilateral
divorce (the latter being potentially welfare enhancing). Finally, such partial equilibrium
assessments may be incomplete because unilateral divorce increases everyone’s access to divorce,
and even those who choose not to get divorced may be affected by the existence of this new
option.
In the literature on the economics of the family there has been growing consensus on the
need to take bargaining and distribution within marriage seriously. Such models of the family
rely on a threat point to determine allocation within the household. The switch to a unilateral
divorce regime redistributes power in a marriage, giving power to the person who wants out, and
reducing the power previously held by the partner interested in preserving the marriage.
Potentially, this may cause large changes in marital dynamics, whether or not there is an
increasing tendency to actually exercise the divorce option. For instance, in a society in which
people can leave abusive partners, spouses may be less likely to be abusive.
This paper exploits the variation occurring from the different timing of divorce law
reforms across the United States to evaluate changes in suicide, domestic violence, and spousal
murder rates in an attempt to measure some of the important effects of the “no-fault revolution.”
More specifically, we are examining suicide rates in an attempt to find a quantifiable measure of
well-being. While variation in suicide rates only reflects changes at the extremes of the
distribution, Di Tella, MacCulloch and Oswald (1997) show that aggregate suicide rates tend to
co-move with other aggregate measures of subjective well-being. Family violence surveys
conducted in the mid-1970s, and again in the mid-1980s, provide basic detail about domestic
5
In fact, the causation may run the other way. For example, although alcoholism may result from divorce,
it is also likely that alcoholics make undesirable spouses. Bedard and Deschenes (2003) present evidence
3
violence. Spousal murder rates are analyzed as a further quantifiable indicator of domestic
violence.
6
We find that states that passed unilateral divorce laws saw a large decline in both female
suicide and domestic violence rates. Total female suicide declined by around 20% in states that
adopted unilateral divorce. There is no discernable effect on male suicide. Our data on spousal
conflict suggest that a large decline in domestic violence occurred in reform states. Furthermore,
our results suggest a decline in women murdered by intimates, although the timing evidence is
less supportive of this claim. As with suicide, there is no discernable effect on males murdered.
2. Mediating Forces: Divorce Rates and Bargaining within Marriage
Our analysis is concerned with changes in adult wellbeing occurring as a result of a shift
to unilateral divorce (which permits divorce upon application by either spouse) from the pre-
existing divorce laws (which typically required either the consent of both spouses or a
demonstration of marital fault). There are two mechanisms through which a change in divorce
law regime may affect indicators of spousal well-being. The first is by affecting the divorce rate.
This direct mechanism traces the effects of easier access to divorce to higher divorce rates,
through to the sorts of deleterious effects of divorce documented in the public health and
sociological literatures.
7
If this were the only channel, then unilateral divorce laws would provide
a useful instrumental variable for analyzing the adverse effects of divorce.
The second mechanism is by changing bargaining power and behavior within marriage.
If the divorce regime affects the bargaining position of spouses in a way that changes intrafamily
distribution then we expect to observe changes in spousal relations and wellbeing. Note that this
mechanism may be important even if there is no effect of divorce regime on divorce rates.
that the deleterious financial consequences of divorce are not robust to a more careful analysis of causation.
6
Campbell (1992) provides evidence that domestic violence is a factor in most incidents of intimate
homicide. Furthermore, existing estimates suggest that between one-quarter and one-half of women
murdered are killed by their partner (Greenfeld et. al, 1998) making homicide of intrinsic interest.
7
Waite (1995) documents a range of cross-sectional correlations between divorce and poor outcomes.
4
What can theory tell us about these two mechanisms? The first mechanism is mediated
by rising divorce rates. Yet Becker (1981) argues that the Coase theorem applies to marital
bargaining, and hence the divorce rate should be invariant to divorce regime. Under this view,
unilateral divorce laws simply transfer a well-defined property right – the right to remarry – from
the spouse who wants to remain married to the partner desiring a divorce. Efficient bargaining
ensures that marriages only dissolve if continuing the marriage would be jointly sub-optimal, and
this efficient bargain will obtain irrespective of the initial assignment of property rights. Recent
empirical research suggests that liberalized divorce laws led to only a small and transitory rise in
divorce that dissipated within a decade.
8
The second channel reflects the impact of divorce laws on spousal bargaining over the
distribution of marital rents. There are three canonical models of intrafamily distribution. The
first is the common preference approach, which holds that families act as if maximizing a single
utility function. This common preference can be motivated either by love (altruism, such that
both spouses care equally about their own and their partner’s satisfaction, as in Becker, 1981), or
by the parties seeking to maximize a “social welfare function,” agreed upon in a complete
marriage contract.
9
The sharpest prediction of the common preference approach is that outcomes
are invariant to the distribution of resources between spouses. By contrast, bargaining models
hold that the presence of threat points determines intrafamily distribution. In the separate spheres
bargaining model of Lundberg and Pollak (1993), these threat points are internal to the marriage.
That is, the equilibrium distribution is maintained by the threat of reversion to a non-cooperative
equilibrium involving, for example, burnt toast or sleeping on the sofa. Finally, the exit threat
bargaining models of Manser and Brown (1980) and McElroy and Horney (1981) emphasize
external threat points - specifically each party’s best option outside the marriage. If this exit
threat is binding, then changing opportunities outside the marriage will change the equilibrium
8
See Wolfers (2003) which clarifies the interpretation of the results in Friedberg (1998).
9
In the latter case the division of marital rents is agreed upon prior to the marriage.
5
distribution within the marriage. If the internal threat is binding, then such changes do not affect
outcomes.
10
To see how divorce laws affect the external threat point, note that prior to unilateral
divorce, a partner wishing to dissolve the marriage could leave without their spouse’s consent.
However, in such a situation, a legal divorce is not granted and, as such, the right to remarry is
forfeited. Under unilateral divorce the value of the exit threat increases for the unsatisfied spouse,
as the right to remarry is retained regardless of the position of one’s spouse. Thus, the exit threat
model predicts that changes in divorce regimes will have real effects. If the divorce threat is
sufficiently credible, it may directly affect intrafamily bargaining outcomes without the option
ever being exercised. That is, there may be profound changes not mediated by higher divorce
rates, and hence, unilateral divorce laws cannot be considered a valid instrument for divorce.
Consequently, we estimate reduced form regressions that represent the effect of divorce regime
on suicide, domestic violence, and spousal homicide.
Although theory predicts that real effects can flow from divorce reform, signing these
effects is much harder. An increase in access to divorce could decrease suicide rates simply
because suicide and divorce might be substitutes. That is, while the misery of living in an abusive
or otherwise unhappy relationship may result in suicide, the option of divorce and remarriage may
avert this course of events.
11
Alternatively, rising divorce or the threat of divorce may increase
the number of unhappy or abandoned spouses, raising the temptation of suicide.
Similarly, domestic violence may decrease because the threat to leave if abused becomes
credible under the unilateral divorce framework. If the abuser wishes to continue the marriage
then this threat may be sufficient to prevent abusive behavior. Note that this change in behavior
results from the change in bargaining power and, as such, can occur without any observed change
10
As Lundberg and Pollak acknowledge both threat points may be possible, and therefore the relevant
threat point is determined by institutional frameworks and individual utility functions (1993, p. 1001).
11
For an analysis of suicide that emphasizes the option value of staying alive, see Hamermesh and Soss
(1974).
6
in divorce propensities. Finally, most spousal homicides occur in the context of abusive
relationships,
12
and hence any policy that reduces the barriers to exiting such a relationship
reduces the probability of both abuse and spousal homicide.
Countering these forces, there are several reasons why unilateral divorce may raise
spousal violence. The first is that without a legal system that enforces the marriage contract,
individuals may substitute private for public enforcement of their marriage contracts. Under the
consent divorce framework, spouses “owned” each other, and this ownership was enforced by the
state through legal sanction. Under unilateral divorce this perceived property right is threatened;
hence, there exists the possibility that private enforcement, through violence, will substitute for
state sanctions.
13
The resulting increase in domestic violence may also lead to an increase in
murder. Finally, the intense emotional distress and personal tumult associated with divorce
proceedings might provoke an increase in domestic violence and murder. That is, unilateral
divorce, in so much as it leads to an increase in the number of highly charged divorces, creates
more violent situations.
14
3. Empirical Strategy
We follow Friedberg’s (1998) coding of state divorce regimes and the dates of divorce
reforms. It should be noted that there are actually degrees of unilateral divorce, in that legislation
might allow unilateral divorce conditional upon a separation period. We code states both with
and without separation requirements as unilateral divorce regimes.
15
12
Campbell (1992).
13
Miron (1998) makes this argument in terms of drug contracts, arguing that when the state refuses to
enforce such contracts private enforcement mechanisms such as gang violence will replace court sanctions.
14
The study that is closest in spirit to the present paper is Gillis’ (1996) assessment of the effects of a
previous liberalization of divorce law which introduced divorce in mid-19th century France. The reform
allowed judicial separation to be granted in response to adultery or the threat or occurrence of serious
violence. Gillis analyses time series data on divorce and “deadly domestic quarrels,” finding that the
overall effect throughout this period was a decline in murder rates but a small increase in “spontaneous”
murders.
15
Around one-third of states have separation requirements, ranging from six months to five years.
7
Of the fifty states, five are yet to adopt any form of unilateral divorce: Arkansas,
Delaware, Mississippi, New York and Tennessee. Of the forty-six states that currently have
unilateral divorce regimes, nine had adopted some variant of unilateral divorce before the no-fault
revolution of the early 1970s. Along with the thirty-seven remaining states we include the
District of Columbia, which adopted unilateral divorce in 1977. Consequently, we effectively
have thirty-seven “experiments” of changing divorce laws. The remaining fourteen states are
included as controls. Table 1 lists the year that each state changed its divorce regime to allow
unilateral divorce.
We use the natural variation resulting from the different timing of the adoption of unilateral
divorce laws across states to estimate the effects of these laws on murder, suicide, and domestic
violence rates for men and women independently. Consequently, we use state-based panel
estimation, including both state and time fixed effects in all regressions. A dummy variable
indicating whether the state currently has a unilateral divorce regime is our variable of interest.
The dependent variable is the annual suicide, domestic violence, or murder rate. Where possible,
we report our coefficients as elasticities (evaluated at the unweighted cell mean). That is, the
reported results are interpreted as the percentage change in the relevant rate stemming from the
change to unilateral divorce. Appendix A provides summary statistics.
4. Suicide Results
Data on suicide comes from the National Center for Health Statistics (NCHS).
16
The
NCHS data are a census of death certificates, which code the cause of death for all deceased
persons. There are broad codes for suicide, as well as a more detailed coding structure that
includes data on the method of suicide. Individual data on gender, state of residence, and age of
16
Suicide data for 1964-1967 were hand entered from annual editions of the NCHS report “Vital Statistics:
Mortality, Vol.2”. Data for 1968-78 are calculated from ICPSR Study No. 8224, “Mortality Detail Files:
External Cause Extract, 1968-78”, PI: National Center for Health Statistics. Data from 1979-96 have been
downloaded from the Center for Disease Control’s Wonder system which accesses the NCHS “Compressed
8
death are also collected.
17
By examining the period from 1964 through to 1996, we can both
robustly identify suicide rates before the adoption of unilateral divorce laws, and trace their
evolution over the following years.
Note that the dependent variable is the suicide rate of all persons, not just those who have
been married. We analyze this variable both because of data limitations (the NCHS begin coding
marital status in 1978) and to avoid endogeneity problems posed by the possibility that marriage
decisions may respond to divorce regime.
We employ OLS to estimate:
tstst
t
ts
s
s
k
ts
k
k
ControlsYearStateUnilateralrateSuicide
,,,
εληβ
++++=
∑
∑
∑
where Unilateral
k
refers to a series of dummy variables set equal to one if a state had adopted
unilateral divorce k years ago. Thus the results, shown in Table 2, map out the full dynamic
response of the suicide rate to the law change.
There is a large and statistically significant reduction in the female suicide rate following
the change to unilateral divorce. Further, this effect seems to grow as the effects of divorce law
reform percolate through society, presumably reflecting couples learning about the new law,
social norms about the family adjusting, and spouses starting to understand their new rights.
Averaging the effects over the twenty years following reform suggests an aggregate decline of
5-10%. For male suicides, these estimates indicate no discernible effect.
We test the sensitivity of our specification to a range of controls, including a proxy for
the evolving economic power of women (the ratio of male-to-female employment rates),
indicators of the business cycle (state income per capita and unemployment), welfare generosity
(the maximum AFDC payment for a family of four, and the share of the state population on the
welfare rolls), the availability of abortion, and the racial and age composition of the state. While
Mortality Files” (http://wonder.cdc.gov/). Apart from minor revisions to the International Classification of
Diseases, these data are consistently coded.
17
Our population data, downloaded from www.census.gov, are not coded by gender; the evolution of
gender shares in each state are imputed from the March CPS files (for the population aged 14 or over).
9
we find that some of these controls are significant explanators of the suicide rate, column 2 shows
that they barely change the estimated effect of unilateral divorce. This column represents our
preferred estimate, suggesting an average decline in female suicide of around eight percent, and a
long run decline that is perhaps double that. Weighted least squares results are also broadly
similar (column 3). We add state-specific time trends in column 4, finding that their inclusion
causes the standard errors to increase. For women, the specification including state-specific time
trends is not precisely estimated enough to reject either a null that the pattern of coefficients
follows that shown in columns 1-3, or a null of no effect. The point estimates, while smaller, are
still large. For males, including state-specific trends is suggestive of a decline in male suicide
rates following the advent of unilateral divorce.
In further robustness testing (not shown), we ran each of our baseline regressions
omitting in turn individual states or years, finding that particular states or years do not unduly
influence our results. Robust estimation procedures, including median regression, yielded similar
results. Further, while OLS implicitly gives equal weight to each of our thirty-seven divorce
reform experiments, we also found similar results using population-weighted least squares and
generalized least squares. We also experimented with the control group, dropping those states
that did not change their laws from the estimation. We found that estimating off only the
variation due to the different timing of reform was sufficient to identify the noted large decline in
female suicide. This specification was also suggestive of a decline in male suicide.
Timing evidence might speak to a causal interpretation of these results. We are
particularly interested in whether the change in suicide post-dated the change in divorce regime,
and whether adjustment to the new regime seems plausible. In order to examine these potential
lags, we added a series of leads to our preferred specification, coding dummies for whether
unilateral divorce will become law in 1-2 years, 3-4 years and so on, with leads beyond 10 years
coded to the 9-10 year group. The leads are included to check if the timing of the decrease in
suicide rates corresponds with the change to unilateral divorce. The estimated coefficients are
10
shown in Figures 1 and 2 (the graphs are normalized so that the pre-change effects are centered
on zero).
Firstly, note that the coefficients on the dummies indicating the period prior to the
divorce law reform are all close to zero, and in no case are they (individually or jointly)
statistically distinguishable from zero. This speaks clearly to a causal interpretation of these
results. Secondly, reinforcing our baseline results, the graphs show that there was a large and
statistically significant decrease in female suicide rates, and no discernable affect on male rates.
The smooth and plausible shape of the response of female suicide suggests that our treatment
effect is reasonably well identified. The graph illustrates that the full effects of the change in
divorce laws take almost twenty years to percolate through society.
While the death certificate data do not continuously code marital status, we can
disaggregate our main results by age.
18
For brevity we only show the results for women because
even disaggregating by age, we still find no effects for men. Figure 3 shows our results by age
group for women, mapping the estimated response of suicide rates following the change in
divorce regimes, where the analysis in Figure 1 is repeated separately for women in each of
eleven different age groups. These age groups comprise unequal shares of the population, and so
in each case coefficients are scaled by their share of the US population, allowing these figures to
be added to yield the aggregate effect (shown in the bottom right panel). For teens, the effect is a
relatively precisely estimated zero, reflecting both the lack of correlation between teen suicide
and divorce laws and the relatively small number of teen suicides. The second row of Figure 3
shows that prime-age women account for the bulk of the main effect, with unilateral divorce
substantially reducing the suicide rates of women in each of the age groups from 25-65. Turning
to the elderly, it appears that unilateral divorce laws had little effect on suicide decisions,
although there may be some impact on women aged 65-74 (these estimates are sufficiently
imprecise as to be consistent with either no effect or a meaningful decline). The bottom right
11
panel shows that adding the effects across the eleven panels yields a set of estimates that are
consistent with those shown in Figure 1, suggesting that these initial aggregate results are robust
to the inclusion of age-specific state and year effects (and to the interaction of age with the other
controls). Overall the observed correlation between the adoption of unilateral divorce and the
decline in female suicide seems extremely robust, and we can be confident that neither youth nor
the elderly drive the observed correlation between female suicide and divorce regime – a result
that is broadly consistent with the causal interpretation we offer below.
5. Interpretation
It is useful to think about the role that divorce itself is playing in generating these
observed changes in suicide rates. Two interpretations seem particularly relevant. The first is
that the reduction in female suicide reflects both women escaping from bad marriages and the
redistribution of power within marriages that results from increased access to divorce. Under this
interpretation unilateral divorce has both direct effects on bargaining within marriage, and effects
that are mediated through increased divorce.
A more restrictive interpretation is that our results are simply the reduced form
representation of an instrumental variables regression in which unilateral divorce laws are an
instrument for higher divorce rates. This IV interpretation assumes that the decrease in female
suicide is solely the result of higher divorce rates and does not reflect bargaining within marriage.
With no direct evidence as to the presence or absence of such a channel, we are reluctant to
embrace this more restrictive interpretation. Further, the indirect evidence that we have speaks,
albeit weakly, against this view. Specifically, note that in Figure 1 there is no immediate spike in
18
We are grateful to an anonymous referee for suggesting this analysis.
12
suicide following the regime change. By contrast, a spike in divorce typically follows a shift in
regimes – as the courts cater to pent-up demand for unilateral divorces.
19
Finally, Wolfers (2003) casts doubt on Friedberg’s (1998) assessment of the relationship
between divorce laws and divorce rates. Friedberg’s result relies upon the inclusion of state-
specific time trends that are calculated on data that include only a few years of data prior to the
legal change and, in some cases, only one year. As a result the estimated state-specific time
trends reflect the dynamic response of the divorce rate to the regime change, rather than pre-
existing divorce trends in each state. By partialling out a downward trend in divorce in unilateral
divorce states, Friedberg finds an artificially large rise in divorces following the legal change.
20
When these trends are omitted, or are calculated so as to reflect only pre-existing trends in a
state’s divorce rate, Wolfers concludes that the adoption of unilateral divorce caused the divorce
rate to be higher for a decade, and then lower in the ensuing decade. By contrast, Figure 1
suggests that suicides decline steadily over the twenty years following divorce law reform. The
short-term rise in divorce and the steady decline in suicide seem somewhat difficult to reconcile if
divorce is the main mediating mechanism. These interpretation issues remain relevant as we now
turn our attention to the relationship between unilateral divorce and intimate homicide and
domestic violence.
19
This heuristic argument is equivalent to the type of argument in formal over-identification tests. The first
stage regression is formally over-identified in that we have separate instruments for a rise in divorce based
on dummies of whether the legal change occurred in the last two years, 3-4 years ago etc. The heuristic
argument is that the large change in divorce generated in the first two years after a regime change yields
only a small decline in suicide rates, a result that is difficult to reconcile with smaller changes in divorce
rates in ensuing years that yield much larger effects on suicide. Of course, appealing to a specific
constellation of lags between the rise in divorce and the decline in suicide can rationalize this result away.
20
The estimated trend is negative in unilateral divorce states relative to the controls because the divorce
rate spikes up immediately following the change in regime, and then declines smoothly, asymptoting
toward a smaller long-run effect.
13
6. Domestic Violence
The most credible cross-state data on domestic violence are the landmark Family
Violence Surveys undertaken by sociologists Straus and Gelles in 1976 and again in 1985.
21
These data come from household interviews that ask how couples resolve conflict.
22
This type of
survey instrument typically yields higher estimates of domestic violence than police reports or
crime victimization surveys because the victim need not perceive the act as domestic violence
and/or a crime for it to be recorded. While still an imperfect survey instrument, Markowitz
(1999, p.8) argues that this methodology is currently “the best available technique for collecting
truthful information on domestic violence.”
The two available surveys yield cross-sectional data for 1976, by which time thirty-one
states had recently changed their divorce laws, and again for 1985, by which time all thirty-seven
regime changes identified in Figure 1 had occurred. This timing is somewhat unfortunate in that
it is unclear how the differential timing of reform across states would translate into differential
changes in domestic violence rates over the 1976-85 period. Yet, although the differential cross-
state timing in reform yields little analytical leverage, we can compare changes in violence rates
among our thirty-seven states that constituted the “no-fault revolution” with two alternative
control groups: the five states that are yet to adopt unilateral divorce (AR, DE, MS, NY and TN),
and the nine states whose pre-existing regime involved unilateral divorce (AK, LA, MD, NC, OK,
UT, VA, VT and WV).
23
If there is an underlying relationship between domestic violence and
divorce regime, we would expect to observe changing violence propensities in the treatment
group relative to the controls. Because the survey universe consists only of couples living in a
conjugal unit, we are limited to analyzing rates of domestic violence within intact marriages.
21
Crime victimization survey data both lack state identifiers and are not available for the relevant time
period. Police reports suffer both from serious problems of under-reporting and, more importantly, changes
in social norms regarding reporting over the relevant time period.
22
Murray A. Straus and Richard J. Gelles “Physical Violence in American Families”. The 1976 and 1985
surveys are ICPSR studies 9211 and 7733, respectively.
14
Thus, we cannot disentangle whether the estimated effects reflect a decreasing propensity towards
spousal violence, or an increasing propensity for abused spouses to exit their marriages.
Table 3 shows illustrative differences-in-differences estimates of the effects of unilateral
divorce on domestic violence.
24
The first row of Panel A tells us that between 1976 and 1985
domestic violence towards women declined by 1.7 percentage points in reform states, while it
rose 2.5 percentage points in the control states. Thus, the difference-in-difference estimate
suggests that the treatment—adoption of unilateral divorce—led domestic violence rates to
decline by 4.3 percentage points, or by around one-third, over the 1976-85 period. Panel B shows
a slightly smaller, statistically insignificant decline in wife-to-husband violence. These
magnitudes are clearly important and lead us to the micro-data to probe this result more
intensively.
Table 4 analyzes household-level data in which the dependent variable, Domestic
Violence, is a dummy indicating whether the specified type of violence occurred within each
household. We estimate the micro-data analog of our differences-in-differences estimate:
t,s,iists
1985
tsts,i,
)controlseffects state( effects yearTreatment)YearTreatment(Violence Domestic
εδβ
+++++×=
where
Treatment is a dummy variable that is equal to one if the state is coded in Table 1 as a
reform state, and is zero otherwise.
The first row of Table 4 shows the mean rates of violence across households. Perhaps
surprisingly, men are as likely to be physically abused by their spouses as women are. The next
row simply reproduces the differences-in-differences estimates from Table 3, for each of the four
categories of spousal abuse. Extremely large declines in violence are found for each abuse
indicator. Adding state fixed effects in the next row sharpens these estimates somewhat, and
23
The 1976 survey did not sample from all states, and hence we are forced to omit the following states
from our analysis: AK, AR, DC, DE, HI, IA, KY, MA, ND, NH, NM, NV, RI, SD, WY.
24
The definition of domestic violence follows Gelles and Straus (1994). That is, we code domestic
violence as occurring if there has been any incident over the last year in which a person threw something at
15
these large effects are all found to be statistically significant. The following four rows show that
these results are robust to the inclusion of state fixed effects, a rich set of individual-level
controls, the set of within-state time-varying economic and social policy controls shown in
Table 2, and also the use of a probit estimator. Further, dropping specific states from the sample
did not appreciably change these results.
Comparing these declines in violence rates with their base rates, domestic violence
appears to have declined by somewhere between a quarter and a half between 1976 and 1985 in
those states that reformed their divorce laws during the “no-fault revolution.” We now turn to an
alternative indicator of domestic violence—intimate homicide—to further probe the robustness of
these results.
7. Intimate Homicide
Our data on homicide come from the FBI Uniform Crime Reports (UCR).
25
The UCR
data are derived using a voluntary police agency-based reporting system. The Supplementary
Homicide Reports of the UCR provide
incident-level information on criminal homicides,
including data describing the date and location of the incident, as well as a range of information
on both the offender and the victim. The particular richness of this data is that it codes the
relationship of the victim to the murderer, where known.
Because the FBI data rely on police reporting, there are often problems of under-
reporting or downgrading of crimes. However, the nature of homicide means that both of these
problems are minimized. The FBI counts of total murders each year by state were checked
against the independently gathered NCHS murder count. Generally, these two data sources were
their partner, pushed, grabbed, shoved, slapped, kicked, bit, hit with fist, hit or tried to hit with object, beat
up, or threatened or used a gun or knife against their partner.
25
Data for 1968-75 are from ICPSR Study No. 8676, “American Homicide, 1968-1978: Victim-Level
Supplementary Homicide Reports”, Marc Riedel and Margaret Zahn (1994). Data for 1976-94 are
extracted from ICPSR Study No. 6754, “Uniform Crime Reports [United States]: Supplementary Homicide
Reports, 1976-1994”, James Alan Fox (1996). The consistency of these data is discussed in Appendix B.
16
consistent, and hence the rest of our analysis uses the FBI data,
26
which include their coding of
victim-perpetrator relationships.
Nonetheless, there remains a range of problems when working with these data. First, the
participation of agencies is not completely consistent, and when an agency fails to report in a
particular month, we cannot tell whether this reflects laxity with paperwork or that there were no
murders to report.
27
Second, there are various coding breaks arising from the changing
definitions of victim-perpetrator relationship, causing a minor break in 1972, and a more
important break in 1976. These coding breaks present a problem for our analysis because,
conceptually, we would like to capture any relationship that may be affected by changes in family
law. Such relationships include, along with spouses, domestic and non-domestic romantic
partners and other family members (particularly children). However, there are data problems
constructing such a series that is consistent across coding breaks.
28
In this section we will
examine three successively broader definitions of intimate homicide. The narrowest only
includes spousal homicide, the next group includes homicides committed by any family member
or romantic interest, and finally we expand our treatment group to our broadest categorization,
which includes all homicides committed by non-strangers.
The defect of the broader measures is that the treatment group is defined to include many
relationships that are not affected by the treatment of unilateral divorce. The defect of narrower
measures is that police classifications of victim-perpetrator relationships as “spousal” are likely to
have changed over time, in a way that is correlated with family law regimes, leading to (difficult
26
The FBI data for Illinois are quite different from the Death Certificate data. Looking closely at the data,
we find that the Chicago Police Department failed to report any murders in 1984, 1985, November 1986-
May1987, July 1987-December 1987 and July 1990-December 1990. It is implausible that there were no
murders in these periods, and hence we believe the FBI data to be wrong. Thus we omitted Illinois from our
homicide samples.
27
When there are no data for an entire state, for a whole year, this could reflect either that the state was not
participating in the reporting program, or that there were no murders in that state-year. We assume non-
participation when a zero murder count would lie outside a three-standard error confidence band for that
state, and infer a number by linear interpolation. Otherwise we assume a zero murder count. These
adjustments affect 37 of our 2754 state-year-sex observations.
28
In Appendix B we outline our attempts to construct consistent series.
17
to sign) bias issues.
29
Further, identifying intimates narrowly, such as by “spouses”, is more
likely to suffer from endogeneity problems as the legal status that people choose for their
relationships may change with changes in the legal regime.
For women murdered, Table 5 suggests a large and significant decline following the
adoption of unilateral divorce for all three definitions of intimate homicide, with column 1
suggesting declines on the order of around 10 percent. Column 2 shows that this estimate is
robust to adding a rich set of controls, including not only the economic, social policy and
demographic variables previously considered, but also a set of criminal justice variables including
a death penalty indicator, Donahue and Levitt’s Effective Abortion Rate, and the share of the
state’s population in prison population rate, lagged one year.
The results for males murdered are imprecisely estimated and would admit large effects
in either direction. The estimates change substantially across different definitions of intimate
homicide, and adding controls leads to moderate changes in the estimates.
30
As with the suicide data, timing evidence might assist us in interpreting our results.
Therefore, we once again replace the single dummy variable
Unilateral in the baseline model
with several dummy variables indicating the number of years since (or until) the law went (goes)
into effect. We run this regression for all three categories of intimate homicide. The estimated
coefficients for females murdered are shown in Figure 4. For clarity, standard error bands are not
shown, but as a rough indicator, estimated standard errors for each lead, or lag, plotted are around
twice that shown in the corresponding row of Table 5. The imprecision with which we estimate
29
While the coding of married partners as “spouses” presents no difficulty, coding of common-law
marriages, cohabiting couples, same-sex couples, romantic partners and separated spouses is likely to have
changed over time. Although these groups may be small compared to the whole population, we do not
know if this is true of the homicidal population. All that is known with certainty is that a homicidal
member from one of the above groups would not have been coded as a stranger, which is the motivation for
looking at the broadest of our definitions of the treatment group.
30
Thomas Dee (2003) has also analyzed these data, employing count data methods on a short (1968-1978)
panel. He finds a large increase in males murdered by their spouses. His paper contains a reconciliation of
his results with ours, which largely turns on his shorter sample period, coding of intimate homicide, and
functional form. As we shall see, the effects on men appear to be extremely sensitive to small changes in
specification, undermining our confidence in any estimate.
18
effects on males murdered is sufficiently large that we omit them from the rest of the analysis of
intimate homicide.
Figure 4 confirms the initial findings of a decrease in women murdered in the period
following the passage of divorce law reforms. However, the timing evidence is somewhat
worrying, and the reader is left to judge whether the decline in homicide pre-dated the law change
to an extent that undermines our results. This raises the possibility that our regression results may
be picking up the effects of some alternative phenomenon that pre-dated divorce law reform.
The fact that family law affects behavior between intimates but not between strangers
provides an opportunity to further probe these results. Specifically,
non-intimate homicide may
serve as an ideal placebo group. Column 3 of Table 5 shows the differences-in-differences
(panel) estimates for the non-intimate homicide placebo group (that is, the dependent variable is
the aggregate homicide rate, less the relevant definition of intimate homicide). These results
suggest that there is a negative correlation between non-intimate homicide and divorce laws,
albeit not a statistically significant one. These results also give us a chance to assess an
alternative counterfactual: instead of assuming that, in the absence of divorce reform, intimate
homicide would remain unchanged (as in the first two columns), the differences-in-differences-
in-differences in column four assumes that the change in non-intimate homicide is the relevant
baseline. These triple difference estimates suggest that intimate femicide declined when
compared with this counterfactual, but that this difference is not statistically significant. For men,
the estimates remain both imprecise and sensitive to changes in definition. Finally, other crime
measures provide a further set of interesting placebos, and these results (shown in the bottom
panel of Table 5) generally show little correlation between state crime trends and divorce laws.
8. Conclusion
Our analysis examines indicators of adult well-being following a regime shift to
unilateral divorce from the pre-existing divorce laws. We have attempted to measure specific
19
changes in family distress resulting from the radical changes made over the past thirty years to
divorce laws. These changes led to one spouse being able to obtain a divorce without his or her
partner’s consent. Examining state panel data on suicide, domestic violence, and murder, we find
a striking decline in female suicide and domestic violence rates arising from the advent of
unilateral divorce. Total female suicide declined by around 20% in the long run in states that
adopted unilateral divorce. We believe that this decline is a robust and well-identified result, and
timing evidence speaks clearly to this interpretation. There is no discernable effect on male
suicide.
Data on conflict resolution reveal large declines in domestic violence committed by, and
against, both men and women in states that adopted unilateral divorce. Furthermore, we find
suggestive evidence of a decline in females murdered by intimates, although the timing evidence
makes this a more suspect result. As with suicide, there is no discernable effect on males
murdered, although this reflects the imprecision and volatility of our estimates.
While our results are open to the interpretation that the large declines identified are the
result of changing divorce rates, we believe that this is only part of the story. Indeed it is difficult
to reconcile the timing of these outcomes with the response of the divorce rate to these reforms.
A more complete story takes changes in marital dynamics into account. Unilateral divorce
changed the bargaining power in marriages and therefore impacted many marriages– not simply
the extra few divorces enabled by unilateral divorce. Speculating on the policy implications of
emerging models of the family, Lundberg and Pollak (1993, p.992) argued that the possibility of
“the dependence of intrafamily distribution on the well-being of divorced individuals provides a
mechanism through which government policy can affect distribution within marriage.” The
mechanism examined in this paper is a change in divorce regime and we interpret the evidence
collected here as an empirical endorsement of the idea that family law provides a potent tool for
affecting outcomes within families.
20
References
Becker, Gary S. (1981)
A Treatise on the Family, Harvard University Press: Cambridge.
Bedard, Kelly and Olivier Deschenes (2003) “Sex Preferences, Marital Dissolution and the
Economic Status of Women”,
mimeo, UC Santa Barbara.
Campbell, J.C. (1992) “If I Can’t Have You, No One Can: Power and Control in the Homicide of
Female Partners”. In J. Radford & D.E.H. Russel (Eds.),
Femicide: The Politics of Women
Killing
. New York: Twayne.
Dee, Thomas (2003) “Until Death Do You Part: The Effects of Unilateral Divorce on Spousal
Homicides”,
mimeo, Swarthmore College.
Di Tella, Rafael, Robert MacCulloch, and Andrew Oswald (1997) "The Macroeconomics of
Happiness", CEP working Paper 19.
Fox, James Alan (1996) “Uniform Crime Reports [United States]: Supplementary Homicide
Reports, 1976-1994”, ICPSR Study 6754.
Friedberg, Leora (1998) “Did Unilateral Divorce Raise Divorce Rates? Evidence From Panel
Data”, American Economic Review, 83(3).
Gelles, Richard J. and Murray A. Straus (1994) “Physical Violence in American Families, 1985”,
ICPSR Study No. 9211.
Gills, A.R. (1996) “So Long as They Both Shall Live: Marital Dissolution and the Decline of
Domestic Homicide in France, 1852-1909”, American Journal of Sociology, 101(5).
Greenfeld, Lawrence A. and eight co-authors (1998) Violence by Intimates, Bureau of Justice
Statistics, U.S. Department of Justice: Washington D.C.
Gruber, Jonathan (2000) “Is Making Divorce Easier Bad for Children?”,
mimeo MIT.
Hamermesh, Daniel S. and Neal M. Soss(1974) “An Economic Theory of Suicide”, Journal of
Political Economy, 82(1).
Holden, K.C. and P.S. Smock (1991) “The Economic Costs of Marital Dissolution: Why do
Women Bear a Disproportionate Cost?”, Annual Review of Sociology 17.
Levitt, Steven and John J. Donahue (2001), “The Impact of Legalized Abortion on Crime”,
Quarterly Journal of Economics, 116(2).
McElroy, Marjorie D. and Mary Jean Horney (1981) “Nash Bargained Household Decisions”,
International Economic Review 22(2).
Jacob, Herbert (1988) Silent Revolution: The Transformation of Divorce Law in the United States.
University of Chicago Press: Chicago.
Markowitz, Sara (1999), “The Price of Alcohol, Wife Abuse and Husband Abuse”, NBER
Working Paper 6916.
Miron, Jeff (1998), “Violence and the U.S. Prohibition of Drugs and Alcohol”, American Law
and Economics Review, 1, 78-114.
21
National Center for Health Statistics (1985) “Mortality Detail Files: External Cause Extract,
1968-1978”, ICPSR Study 8224.
Peters, H. Elizabeth (1986) “Marriage and Divorce: Informational Constraints and Private
Contracting”, American Economic Review, 76(3).
Pollak, Robert A. (1994) “For Better or Worse: The Roles of Power in Models of Distribution
within Marriage”, AEA Papers and Proceedings, AER 84(2).
Pollak, Robert A and Shelley Lundberg (1996) “Bargaining and Distribution in Marriage”,
Journal of Economic Perspectives, 10(4).
Pollak, Robert A and Shelley Lundberg (1994) “Noncooperative Bargaining Models of
Marriage”, American Economic Review, 84(2).
Pollak, Robert A and Shelley Lundberg (1994) “Separate Spheres Bargaining and the Marriage
Market”, Journal of Political Economy, 101(6).
Riedel, Marc and Margaret Zahn (1994) “American Homicide, 1968-1978: Victim-Level
Supplementary Homicide Reports”, ICPSR Study No. 8676.
Straus, Murray A. and Richard J. Gelles (1994) “Physical Violence in American Families, 1976”,
ICPSR Study No. 7733.
Waite, Linda (1995) “Does Marriage Matter?” Demography 32(4).
Wolfers, Justin (2003) “Did Unilateral Divorce Raise Divorce Rates? A Reconciliation and New
Results”, NBER Working Paper #10014.
Appendices - 1
Appendix A: Summary Statistics
Mean Standard
Deviation
Min. Max. n
Suicide Rates
Female Suicide Rate (Suicides per million women in the state per year)
Total 54.4 18.8 9.2 183.4 1683 state-years
within states 12.0 33 years (1964-96)
Male Suicide Rate (Suicides per million men in the state per year)
Total 202.2 53.5 74.5 435.4 1683 state-years
within states 28.0 33 years (1964-96)
Homicide Rates: FBI Count
by Spouses
Women killed by spouses per million women in the state each year
Total 7.3 4.7 0.0 36.9 1323 state-years
within states 3.4 27 years (1968-94)
Men killed by spouses per million men in the state each year
Total 5.5 5.4 0.0 43.9 1323 state-years
within states 3.7 27 years (1968-94)
by Family Members
Women killed by intimates per million women in the state each year
Total 14.7 7.9 0.0 63.2 1323 state-years
within states 5.2 27 years (1968-94)
Men killed by intimates per million men in the state each year
Total 18.4 13.4 0.0 95.9 1323 state-years
within states 8.0 27 years (1968-94)
by Non-Strangers
Women killed by non-strangers per million women in the state each year
Total 21.2 10.9 0.0 87.5 1323 state-years
within states 6.6 27 years (1968-94)
Men killed by non-strangers per million men in the state each year
Total 56.9 35.7 0.0 178.2 1323 state-years
within states 15.3 27 years (1968-94)
All Murders
Women Murdered per million women in each state each year
Total 32.8 16.6 0 134.9 1323 state-years
within states 9.3 27 years (1968-94)
Men Murdered per million men in each state each year
Total 92.6 56.2 0 314.4 1323 state-years
within states 22.6 27 years (1968-94)
Appendices - 2
Appendix A Continued
Mean Standard
Deviation
Min. Max. n
Domestic Violence
#
Incidence of Overall Husband-to-wife abuse per hundred couples in each state each year
Total 13.0 9.1 0.0 66.7 72 state-years
within states 6.3 2 years (1976, 1985)
Incidence of Severe Husband-to-wife abuse per hundred couples in each state each year
Total 4.2 4.2 0.0 25.0 72 state-years
within states 3.1 2 years (1976, 1985)
Incidence of Overall Wife-to-husband abuse per hundred couples in each state each year
Total 12.7 9.0 0.0 66.7 72 state-years
within states 6.6 2 years (1976, 1985)
Incidence of Severe Wife-to-husband abuse per hundred couples in each state each year
Total 5.5 6.8 0.0 50.0 72 state-years
within states 4.7 2 years (1976, 1985)
Unilateral Divorce Regime (=1 if unilateral, 0 if consent divorce)
Total .69 .46 0 1 1683 state-years
within states .38 33 years (1964-96)
*
Homicide data excludes IL (missing data), DC (outlier)
#
Domestic violence data excludes AK, AR, DC, DE, HI, IA, KY, MA, ND, NH, NM, NV, RI, SD, WY due to missing
observations in 1976 survey. For definitions of severe and overall abuse, see Table 5.
Appendices - 3
Appendix B: Coding of FBI Murder Data
Our data on homicide come from the FBI Uniform Crime Reports (UCR).
31
The UCR data are derived
using a voluntary police agency-based reporting system. The Supplementary Homicide Reports of the UCR
provide incident-level information on criminal homicides, including data describing the date and location of the
incident, and a range of information on both the offender and the victim. This data codes the relationship of the
victim to the murderer, where known. We would like to be able to look exclusively at murders that may be
affected by unilateral divorce. Therefore we are looking for cases in which the perpetrator maybe motivated by
the laws pertaining to divorce – intimate homicides. A useful definition of the treatment group should include
intimates such as spouses, ex-spouses, other partners, and other family members. However, relationships such as
common-law spouse, boyfriend/girlfriend, and even ex-spouse are not consistently coded through our sample.
While time-fixed effects will effectively difference out inconsistencies that are common across states, we are
worried that family law reform may have changed the common meanings of certain definitions of the treatment
group. (For instance the distinction between marriage, common-law marriage and live-in partners has changed
with the social meaning of these terms, which has in turn been affected by family law.
32
) As such we can think
of a bias/efficiency tradeoff. The most efficient strategy is based on a narrow definition of intimate homicide
that includes only spouses, but runs the risk that this category is not well-defined through the sample period. At
the other extreme, perhaps the safest identification strategy is to assume that the legal regime did not affect
murder between strangers, but did affect murder where the victim is known to the murderer – in any capacity.
We employ three definitions of the treatment group – ranging from that which we are most interested in (spousal
murder) to definitions which are less likely to suffer from coding-induced biases (stranger/non-stranger). The
definitions used are outlined in the following table.
Alternative definitions of the “Treatment Group”
Classification Treatment group Control group
Spouses
1968-72
“Spouse kills spouse” All other*
1972-75
“Spouse kills spouse” All other*
1976-94
“Husband”, “Wife”, “Common-law Husband”, “Common-law
Wife”
All other*
Family
1968-72
“Spouse kills spouse”, “Parent kills child”, “Child kills parent”,
“Other family situation”, “Love Triangle”
All other*
1972-75
“Spouse kills spouse”, “Parent kills child”, “Child kills parent”,
“Relative kills relative”, “Other family situation”, “Love
Triangle”
All other*
1976-94
“Husband”, “Wife”, “Common-law Husband”, “Common-law
Wife”, “Mother”, “Father”, “Son”, “Daughter”, “Brother”,
“Sister”, “In-law”, “Stepfather”, “Stepmother”, “Stepson”,
“Stepdaughter”, “Other family”, “Boyfriend”, “Girlfriend”,
“Ex-husband”, “Ex-wife”, “Homosexual relationship”
All other*
Known
1968-72
“Spouse kills spouse”, “Parent kills child”, “Child kills parent”,
“Other family situation”, “Love triangle”, “Money”, “Revenge”,
“Other argument”
All other*
1972-75
“Spouse kills spouse”, “Parent kills child”, “Child kills parent”,
“Relative kills relative”, “Other family situation”, “Love
Triangle”, “Argument/money”, “Other arguments”
All other*
1976-94
All other “Stranger”, “Unknown
Relationship”
* Note that “All other” includes “Murder reason unknown”, “Not stated”, “Not coded” and “Unknown relationship”.
31
Data for 1968-75 are from ICPSR Study No. 8676, “American Homicide, 1968-1978: Victim-Level Supplementary
Homicide Reports”, Marc Riedel and Margaret Zahn (1994). Data for 1976-94 are extracted from ICPSR Study No.
6754, “Uniform Crime Reports [United States]: Supplementary Homicide Reports, 1976-1994”, James Alan Fox
(1996). The consistency of these data is discussed in Appendix B.
32
An additional problem is that we cannot infer that the share of murderer-victim relationships that are coded are
representative of those for which no code was recorded.
Figures – p.1
Table 1: Year of Introduction of Unilateral Divorce Laws, by State
Pre-existing Unilateral Divorce statutes (predate beginning of sample in 1964):
Alaska, Louisiana, Maryland, North Carolina, Oklahoma, Utah, Virginia, Vermont, West Virginia
States adopting Unilateral Divorce Laws:
1969 Kansas, South Carolina
1970 Iowa
1971 Alabama, Colorado, Florida, Idaho, New Hampshire, New Jersey, North Dakota
1972 Kentucky, Michigan, Nebraska
1973 Arizona, Connecticut, Georgia, Hawaii, Indiana, Maine, Missouri, New Mexico, Nevada, Oregon,
Washington
1974 Minnesota, Ohio, Texas
1975 Massachusetts, Montana
1976 Rhode Island
1977 Washington DC, Wisconsin, Wyoming
1980 Pennsylvania
1984 Illinois
1985 South Dakota
Continuing Consent Divorce States (as of 1996):
Arkansas, Delaware, Mississippi, New York, Tennessee
Source: Friedberg (1998)
Figures – p.2
Table 2: Effects of Unilateral Divorce on Suicide Rates (% change)
Female Suicides
Male Suicides
Column No. (1f) (2f) (3f) (4f) (1m) (2m) (3m) (4m)
Year of Change
1.6%
(3.8)
1.3%
(3.4)
2.1%
(4.3)
2.2%
(3.2)
-0.8%
(2.2)
-1.4%
(2.1)
1.6%
(1.5)
-2.3%
(2.1)
1-2 years later
-1.5%
(3.7)
-1.4%
(3.5)
3.0%
(3.4)
-0.4%
(3.7)
1.2%
(1.5)
0.5%
(1.4)
3.0%
(1.2)
-0.8%
(1.6)
3-4 years later
-1.5%
(3.1)
-1.1%
(3.1)
0.8%
(2.5)
0.3%
(4.0)
0.0%
(1.6)
-0.9%
(1.5)
1.1%
(1.2)
-2.6%
(1.8)
5-6 years later
-3.0%
(2.9)
-2.0%
(2.9)
-1.3%
(2.5)
-0.3%
(4.2)
0.4%
(1.5)
-0.2%
(1.5)
0.9%
(1.3)
-2.3%
(2.1)
7-8 years later
-8.0%
(3.0)
-6.6%
(3.0)
-5.0%
(2.7)
-4.8%
(4.8)
-1.0%
(1.8)
-1.3%
(1.8)
0.7%
(1.3)
-4.0%
(2.5)
9-10 years later
-10.0%
(3.0)
-8.5%
(3.0)
-8.5%
(2.6)
-6.2%
(5.3)
-3.5%
(1.7)
-3.9%
(1.7)
-0.7%
(1.3)
-6.8%
(2.8)
11-12 years later
-11.9%
(3.1)
-10.2%
(3.2)
-5.6%
(2.3)
-7.0%
(6.1)
-2.2%
(2.0)
-2.6%
(2.0)
0.5%
(1.3)
-5.6%
(3.3)
13-14 years later
-12.8%
(3.2)
-11.1%
(3.1)
-9.6%
(2.9)
-7.3%
(6.9)
-3.2%
(2.0)
-3.6%
(2.0)
1.4%
(1.4)
-7.0%
(3.6)
15-16 years later
-13.3%
(3.7)
-11.7%
(3.6)
-10.1%
(2.7)
-7.3%
(7.7)
-1.6%
(2.0)
-2.0%
(1.9)
1.3%
(1.5)
-5.6%
(4.0)
17-18 years later
-16.4%
(3.6)
-13.9%
(3.6)
-14.5%
(3.0)
-9.5%
(8.7)
-1.6%
(2.1)
-1.9%
(2.0)
1.7%
(1.6)
-5.7%
(4.5)
≥19 years later
-18.7%
(3.2)
-16.4%
(3.3)
-19.3%
(2.9)
-11.2%
(9.6)
-3.9%
(2.0)
-4.3%
(2.0)
-3.2%
(1.5)
-8.5%
(5.0)
Mean Suicide Rate
54 suicides per million women
202 suicides per million men
Average effect over the
20 years following divorce
law reform
-9.7%
(2.3)
-8.3%
(2.3)
-7.0%
(1.9)
-5.4%
(5.6)
-1.5%
(1.3)
-2.0%
(1.3)
0.7%
(1.0)
-4.9%
(2.8)
F-test of joint significance
p=0.00 p=0.00 p=0.00 p=0.82 p=0.36 p=0.37 p=0.00 p=0.35
Estimation method
OLS OLS WLS OLS OLS OLS WLS OLS
Control variables
State and year fixed effects
9 9 9 9 9 9 9 9
Economic, demographic and
social policy controls
#
9 9 9
9 9 9
State-specific time trends
9
9
Sample 1964-1996, n=1683.
Dependent variable is the aggregate state suicide rate by year. Coefficients are reported as the percentage change in the suicide
rate due to the adoption of unilateral divorce laws the stated number of years ago; this elasticity is calculated using the
unweighted cell mean as the base. Robust standard errors are in parentheses.
#
Controls include the maximum AFDC rate for a family of four, the natural log of state personal income per capita, the
unemployment rate, the female-to-male employment rate, age composition variables indicating the share of states’ populations
aged 14-19, and then ten-year cohorts beginning with age 20 up to a variable for 90+, and the share of the state’s population
that is black, white and other. (Employment status, age and race data are constructed from Unicon’s March CPS files, and refer
to the population aged 14 years or greater.)
Figures – p.3
Table 3: Differences-in-Differences: Effects of Divorce Reform on Domestic Violence
Panel A: Husband to Wife Violence
1976 1985
Difference
(1985-1976)
Treatment
(Adopted Unilateral Divorce)
12.8%
(0.9)
11.1%
(0.7)
-1.7
(1.1)
Control
(No regime change)
10.0%
(1.0)
12.6%
(1.1)
+2.5
(1.5)
Difference
(Treatment-Control)
+2.8
**
(1.4)
-1.5
(1.3)
-4.3
**
(1.9)
[-36%]
Panel B: Wife to Husband Violence
1976 1985
Difference
(1985-1976)
Treatment
(Adopted Unilateral Divorce)
11.9%
(0.9)
11.9%
(0.5)
+0.0
(1.0)
Control
(No regime change)
10.2%
(1.0)
12.8%
(1.0)
+2.7
*
(1.5)
Difference
(Treatment-Control)
+1.8%
(1.3)
-0.9%
(1.2)
-2.7
(1.8)
[-24%]
Sample: n
1976
=2102; n
1985
=3874 (includes cross-section and state over-samples, excludes observations from states that are not
present in the 1976 data; sampling weights are applied).
*
,
**
,
***
denote significance at the 10%, 5%, and 1% levels respectively.
(Robust standard errors in parenthesis; standard errors corrected for clustering within 72 state-year cells)
[Estimate of percent change in violence, in square brackets. ie coefficient evaluated at cell mean]
Dependent variable is a dummy variable set equal to one if the household reports a violent incident as having occurred
between spouses over the preceding year, and zero otherwise. Following Gelles and Straus (1994), violent acts include any
incident in which one spouse threw something at partner, pushed grabbed or shoved, slapped, kicked, bit, hit with fist, hit or
tried to hit with something, beat up partner, threatened with gun or knife, or used a gun or knife.
Figures – p.4
Table 4: Effects of Unilateral Divorce on Domestic Violence
Overall Violence
(a)
Severe Violence
(a)
Husband to
Wife
Wife to
Husband
Husband to
Wife
Wife to
Husband
Average Incidence of Each Type of Violence
11.7% 11.9% 3.4% 4.6%
Estimated Change in Violence Rates in Treatment states relative to Control states
OLS (Diffs-in-diffs)
-4.3%
**
(1.9)
-2.7%
(1.8)
-1.1%
(1.3)
-2.9%
***
(1.0)
add state fixed effects
-5.5%
***
(1.8)
-3.2%
**
(1.5)
-2.0%
**
(0.9)
-3.6%
***
(0.7)
add individual
controls
b
-4.8%
***
(1.7)
-1.9%
(1.4)
-1.8%
*
(1.0)
-3.4%
***
(0.9)
add state-level
time-varying controls
c
-3.8%
**-
(1.8)
-1.8%
(1.3)
-1.8%
(1.0)
-3.0%
***
(0.7)
Probit with individual
controls
b
-4.7%
***
(1.6)
-2.0%
(1.3)
-1.2%
*
(0.7)
-2.1%
***
(0.7)
Sample: n
1976
=2102; n
1985
=3874 (includes cross-section and state over-samples, excludes observations from states that are not
present in the 1976 data; sampling weights are applied)
*
,
**
,
***
denote significance at the 10%, 5%, and 1% levels respectively.
(Robust standard errors in parentheses, corrected for clustering within 72 state-year cells).
All regressions include year fixed effects and either state fixed effects, or treatment/control fixed effects.
Dependent variable is a dummy variable set equal to one if the household reports a violent incident as having occurred
between spouses over the preceding year, and zero otherwise. Thus, reported coefficients reflect the change in the relevant
spousal violence rate in treatment relative to control states – in percentage points. To assess these changes in percentage
terms, compare the reported coefficient with the corresponding term in the first row. Each entry reflects a separate regression.
a
Severe violence is defined as kicked, bit, hit with fist, hit or tried to hit with something, beat up partner, threatened with gun
or knife, or used a gun or knife, in the past year. Overall violence also includes threw something at partner, pushed grabbed or
shoved, and slapped. (Follows Gelles and Straus, 1994.)
b
Individual controls include a saturated set of dummies for respondent’s age, race and gender, and the educational attainment
and current labor force status of both husband and wife. These regressions also include state-fixed effects.
c
State-level time-varying controls include the maximum level of AFDC for a family of four in that state-year, the proportion
of the population on welfare, the ratio of female to male employment rates, the state unemployment rate and log personal
income per capita.
Figures – p.5
Table 5: Effect of Unilateral Divorce on Intimate Homicide (% change)
No Controls Including Controls
#
Intimate
Homicide
(1)
Intimate
Homicide
(2)
Placebo
Non-Intimate
Homicide
(3)
Diffs-in-Diffs-in-Diffs
(Intimate less Non-
intimate)
(4)
Women murdered by Intimates
By Spouse
-10.5%
*
(5.9)
-12.6%
**
(6.0)
-3.7%
(3.5%)
-7.2%
(6.9)
By Family
-8.9%
**
(4.4)
-8.8%
**
(4.4)
-3.1%
(4.2)
-5.6%
(6.1)
By Known
-8.7%
**
(3.7)
-8.5%
**
(3.6)
-0.1%
(5.2)
-7.9%
**
(6.3)
Men Murdered by Intimates
By Spouse
12.3%
(9.2)
3.9%
(9.0)
-2.2%
(2.8)
10.9%
(9.6)
By Family
1.9%
(5.3)
-4.3%
(5.3)
-1.3%
(3.0)
0.2%
(5.9)
By Known
-2.0%
(3.1)
-5.0%
(3.1)
2.7%
(4.3)
-4.1%
(5.2)
Alternative Placebos: “Effects” on Other FBI Index Crimes
Forcible rape
-0.0%
(2.7)
3.3%
(2.4)
Robbery
-2.1%
(2.9)
3.1%
(2.8)
Aggravated
assault
-2.2%
(3.0)
5.9%
**
(2.7)
Burglary
-0.3%
(1.6)
0.7%
(1.6)
Larceny-Theft
-1.5%
(1.1)
-0.3%
(1.2)
Auto theft
-7.9%
***
(2.6)
-2.0%
(2.6)
Sample: 1968-94. Sample excludes Illinois due to missing observations from Chicago Police Department. Also excludes
Washington DC as an outlier: n=1323.
*
,
**
and
***
denote significance at the 10%, 5%, and 1% levels respectively. (Robust standard errors in parentheses.)
Dependent variable is the annual intimate homicide rate in each state. Each cell reports the estimated effect of unilateral
divorce laws from a separate regression. The rows focus on different definitions of “intimate homicide”, while columns report
different specifications. Reported coefficients reflect the percentage change in the relevant homicide rate attributed to
Unilateral Divorce laws; calculated using the unweighted cell mean as the base. All regressions include (significant) state and
year fixed effects.
#
Controls include an indicator variable for the death penalty, the Donahue and Levitt Effective Abortion Rate, and the state
incarceration rate, once lagged, as well as the AFDC rate for a family of four, the natural log of state personal income per
capita, the unemployment rate, the female-to-male employment rate, age composition variables indicating the share of states’
populations aged 14-19, and then ten-year cohorts beginning with age 20 up to a variable for 90+, and the share of the state’s
population that is black, white and other.
Figures – p.6
Figure 1
-20%
-10%
0%
10%
% Change in Suicide Rate
-8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18<-8 yrs >18 yrs
Years Since (until) Unilateral Divorce Laws Adopted
Coefficient Estimates +/- one standard error
Effect of Unilateral Divorce on Female Suicide
Figure 2
-20%
-10%
0%
10%
% Change in Suicide Rate
-8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18<-8 yrs >18 yrs
Years Since (until) Unilateral Divorce Laws Adopted
Coefficient Estimates +/- one standard error
Effect of Unilateral Divorce on Male Suicide
Notes: Figures 1 and 2 show the estimated coefficients (evaluated as elasticities at the unweighted cell mean) from regressing the suicide rate
on dummy variables for whether unilateral divorce laws have been in effect for 1-2 years, 3-4 years, 5-6 years etc; as shown, dummies are
also included for similar leads. State and year fixed effects are also included.
Figures – p.7
Figure 3
-12
-6
0
6
-12
-6
0
6
-12
-6
0
6
-4
-2
0
2
-4
-2
0
2
-4
-2
0
2
-10-8 -6 -4 -2 0 2 4 6 8 101214161820 -10-8 -6 -4 -2 0 2 4 6 8 101214161820 -10-8 -6 -4 -2 0 2 4 6 8 101214161820 -10-8 -6 -4 -2 0 2 4 6 8 101214161820
Younger than 10 years Ages 10-14 Ages 15-19 Ages 20-24
Ages 25-34 Ages 35-44 Ages 45-54 Ages 55-64
Ages 65-74 Ages 75-84 Aged 85 or older All Women, RHS
Coefficient Estimates +/- 1 standard error
Change in Aggregate Suicide Rate
Annual suicides per million females
Years Since (until) Unilateral Divorce Laws Adopted
Each panel reports results from a separate regression, including all controls listed in Table 2 and state and year fixed effects.
Bottom right panel: Top line reproduces aggregate result from Figure 1. Bottom line sums the results from preceding panels. Scale on RHS.
Contributions of each age group to aggregate decline in suicide rates
Effects of Unilateral Divorce Laws on Female Suicide
Figures – p.8
Figure 4
Effect of Unilateral Divorce on Females Murdered by Intimates
-30%
-20%
-10%
0%
10%
20%
(>=9)
(
7-
8)
(
5-
6)
(
3-
4)
(1-2)
0
1-2
3-4
5
-
6
7
-8
9-1
0
11-12
13
-14
15
-16
17
-18
>=1
9
Years since (to) Unilateral divorce introduced
Regression Coefficients:
% Change in Intimate Homicide Rate
by Spouse
by Non-Stranger
by Family Member
Notes: Figure 4 shows the estimated coefficients (evaluated as elasticities at the unweighted cell means) from three regressions, each
focussing on a different definition of the female intimate homicide rate. Each line plots the coefficients on dummies indicating whether
unilateral divorce laws have been in effect for 1-2 years, 3-4 years, 5-6 years etc; as shown, dummies are also included for similar leads.
State and year fixed effects are also included.
