Explaining the Recent Decline in Domestic Violence

Contemporary Economic Policy(ISSN 1074-3529)Vol. 21, No. 2, April 2003, 158–172 © Western Economic Association International

EXPLAINING THE RECENT DECLINEIN DOMESTIC VIOLENCE

AMY FARMER and JILL TIEFENTHALER*

According to the Department of Justice (DOJ), the incidence of domestic vio-lence decreased during the 1990s. Understanding the causes of this decline couldoffer important insight into designing effective policies to continue this trend. Thisarticle uses the Area-Identified National Crime Victimization Survey (NCVS), thesame data used to generate the DOJ’s national estimates, merged with county-level variables, to examine the determinants of women reporting abuse. The resultsindicate that there are three important factors that likely contribute to the decline:(1) the increased provision of legal services for victims of intimate partner abuse,(2) improvements in women’s economic status, and (3) demographic trends, mostnotably the aging of the population. (JEL J12, J16, J22)

I. INTRODUCTION

Domestic violence is a serious problemin our society with significant social costs.Despite the tremendous toll on both the vic-tims and society, domestic violence was notrecognized as a public health issue in theUnited States until relatively recently. Withthe women’s movement of the 1970s, domes-tic violence was increasingly recognized as apublic (not a private) issue. The result hasbeen growing public and private initiativesto eradicate domestic violence. After morethan 20 years of effort, the rate of domes-tic violence appears to be declining. Accord-ing to a recent report published by the U.S.Department of Justice (DOJ, 2000), violenceagainst women by intimate partners fell by

∗This work was completed while the authors werefellows at the Carnegie Mellon Census Research DataCenter with funding from the National Consortium ofViolence Research. The authors thank Dave Merrilland Sara Markowitz for data and computing assistanceand two anonymous referees for helpful comments. Theopinions and findings herein are those of the authorsand do not necessarily represent the opinions or officialfindings of the U.S. Bureau of the Census. Any errorsremain those of the authors.Farmer: Associate Professor, Department of Economics,

Walton College of Business, University of Arkansas,Fayetteville, AR 72701. Phone 1-501-575-6093, Fax 1-501-575-3241, E-mail [email protected]

Tiefenthaler: Associate Professor, Department of Eco-nomics, Colgate University, Hamilton, NY 13346.Phone 1-315-228-7523, Fax 1-315-228-7033, E-mailtiefenthaler@mail. colgate.edu

21% between 1993 and 1998 from 1.1 millionviolent incidents to 876,340 incidents.1

What factors explain this recent decline?Economic models of domestic violence pre-dict that violence against women will declineas women’s alternatives outside their rela-tionships improve. (See Tauchen et al. [1991],and Farmer and Tiefenthaler [1997], furtherdiscussed in section II.) One way to improvebattered women’s alternatives is by pro-viding shelters, hotlines, and other servicesthat help make leaving their relationshipsrealistic for these women. Federal, state,and local governments as well as numerousnonprofit groups have contributed to increas-ing the availability of services for battered

1. It is important to note that there is controversyover the DOJ’s statistics on the incidence of domesticviolence. For example, in a recent national study, theNational Violence against Women Survey, Tjaden andThoennes (1998) find the rate of domestic violence to besignificantly higher than that resulting from the DOJ’sNational Crime Victimization Survey. However, there isno other time series on the incidence of domestic vio-lence to question the NCVS finding that the rate ofdomestic violence is falling over time.

ABBREVIATIONS

BJS: Bureau of Justice StatisticsDOJ: Department of JusticeNCS: National Crime SurveyNCVS: National Crime Victimization Survey

158

FARMER & TIEFENTHALER: DECLINE IN DOMESTIC ABUSE 159

women throughout the country over the past25 years.

Though programs that provide servicesto battered women (such as shelters) mayprovide women with short-term alternativesto staying with their abusers, improvingwomen’s economic status (for example, byincreasing educational attainment) will resultin more battered women being able toachieve self-sufficiency in the long run. If bat-tered women can support themselves, theyare both more likely to leave and have morepower within their relationships if they stay.As a result, economic equality for women—both at the individual and community level—is also predicted to lower the incidence ofdomestic violence.

Do better alternatives for women explainthe decline in domestic violence in the 1990s?Have both improved economic status andmore service provision provided women withbetter alternatives to abusive relationshipsand therefore lowered the incidence of inti-mate partner abuse? This article uses theArea-Identified National Crime Victimiza-tion Survey (NCVS) (see section III for morediscussion of the data set), made available tous through a grant from the National Con-sortium for Violence Research in coopera-tion with the U.S. Census Bureau, to examinethe determinants of the incidence of inti-mate partner abuse. The NCVS is the samedata used by the DOJ to construct theirannual estimates of victimization. Becausethe Area-Identified NCVS includes detailedgeographical identifiers, the authors are ableto investigate the effects of county-level vari-ables, including the existence of programsfor battered women, welfare payments, andwomen’s overall economic status on the inci-dence of abuse at the individual level. Byexamining the determinants of intimate part-ner abuse at the individual level, this articlecan provide some insight into which factorsexplain the decline in the incidence of domes-tic violence nationally.

Using a probit analysis, the authors exam-ine the determinants of an individual womanreporting abuse (see section IV). The find-ings generate three important factors that arelikely to have contributed to the decline inviolence against women in the 1990s. First,although shelters, hotlines, and counselingprograms targeted at battered women arefound to have no significant impact on the

likelihood of domestic abuse, the availabilityof legal services in the county of residencehas a significant negative effect on the like-lihood that an individual woman is battered.Given that the provision of legal services forvictims of domestic violence has increaseddramatically in the 1990s, the authors con-clude that legal services provision is onelikely significant factor in explaining thedecline. Both the improvement in women’seconomic status and demographic changes inthe population may have also contributed toexplaining the decreased incidence of inti-mate partner abuse. This analysis indicatesthat increased education, for both batteredwomen themselves and for women in general,significantly lowers the rate of abuse. Giventhat women’s educational attainment contin-ued to increase in the 1990s, this variableappears to contribute to the decline. Some ofthe decline in the rate of domestic violencealso may be the result of demographic trends.The population is aging, and older womenare significantly less likely to be victims of thistype of abuse.

II. BACKGROUND

In a recent press release (May 17, 2000),the DOJ Bureau of Justice Statistics (BJS)reports that violence against women by inti-mate partners fell by 21% from 1993 through1998. This statistic was calculated from theNCVS, an annual survey on the incidenceof all types of crimes, including violence byintimates (current or former spouses, girl-friends, or boyfriends). National estimates onthe rate of domestic violence are only avail-able from 1993 because the NCVS (formerlycalled the National Crime Survey [NCS]) wassignificantly redesigned in 1992. Previous esti-mates of intimate partner abuse were foundto suffer from a serious problem of under-reporting, but the redesigned survey includesseveral questions specifically concerning inti-mate partner abuse. However, trend data areavailable on intimate partner homicide sincethe 1970s, and these data support a long-termdecline in domestic violence. The BJS reportsthat between 1976 and 1998, the number ofmale victims of intimate partner homicide fellan average 4% per year, and the number offemale victims fell an average 1% per year.

In addition to documenting the decline inthe rate of domestic abuse, the BJS report

160 CONTEMPORARY ECONOMIC POLICY

by Rennison and Welchans (2000) outlinesthe characteristics of the victims. Data fromthe NCVS indicate that being young, black,poor, and divorced or separated all increaseto the likelihood of a woman being a vic-tim of intimate partner abuse. Specifically,women ages 20–24 are the most likely tobe victimized, and black women are 35%more likely to be abused than white womenand 2.5 times more likely than women ofother races. Women in the lowest incomehouseholds have seven times the abuse ratesof those in the highest income households.Finally, women with children under age 12experience twice the rate of abuse than thosewithout young children.

What factors explain the apparent declinein intimate partner abuse? Economic theorypredicts that the incidence of abuse declinesas women gain economic independence andtherefore gain power in their relationships.Farmer and Tiefenthaler (1997) show thatas women’s alternatives to their relationshipsimprove, they should experience less vio-lence because as women gain credible threatsto exit their relationships, they can assertmore power within the relationships. Menare forced to lessen the violence or risklosing their partners. Consequently, womenare more likely to leave, or they suffer lessviolence if they choose to stay. Given thistheory, the authors expect that as women’soutside alternatives improve, intimate part-ner abuse should decline. This could occurdue to improvements in individual women’seconomic status. Women who have greaterearnings or earning potential are more likelyto leave abusive relationships because theycan support themselves. If men respond towomen’s increased power by lowering the vio-lence, women may decide to stay, but clearlythe incidence of violence has declined andthe women are better off. In addition, over-all gender equality in the community mightprovide battered women with better alterna-tives and therefore more credible threats ofleaving. For example, a woman’s threat toleave her abuser is much more credible ifshe lives in an area where a large percent-age of women are employed and women’swages are high. Finally, outside options couldalso be improved via services provided tohelp battered women, such as shelters, wel-fare benefits, and civil legal services to assist

women with protection orders, child support,and custody.

Farmer and Tiefenthaler (1997) find sup-port for the theory as women who withthe highest personal incomes (this includesboth wages and nonwage income, such aschild support and public assistance) experi-ence the least amount of abuse. The BJS find-ing that poor, young, minority women withyoung children are most likely to be vic-tims of intimate partner abuse is also consis-tent with the theory given that these womenhave the fewest alternatives to their relation-ships. Other studies support the notion thatwomen’s alternatives affect the level of vio-lence that they experiences. Gelles (1976)and Pagelow (1981) both find evidence thatwomen with access to fewer resources areless likely to leave their abusers. Kalmuss andStraus (1990) indicate that women who arehighly dependent on marriage suffer greaterabuse, whereas several studies (see, for exam-ple, Coleman and Straus, 1986, and Allenand Straus, 1980) find that women in male-dominated marriages experience more vio-lence. Finally, overall gender inequality hasbeen linked to higher rates of abuse acrossstates and countries.2

Although the empirical literature supportsthe importance of women’s economic alter-natives as a determinant of domestic vio-lence, there is little empirical work that exam-ines the effect of service provision on the rateof female abuse. However, one study doesexamine the effect of service provision onthe rate at which women kill their husbands.Dugan et al. (1998) examines the effects ofdomesticity, women’s economic power, andresources for battered women on intimatepartner homicides in 29 U.S. cities overfour biannual periods. The results indicatethat both women’s economic power and ser-vices provided for battered women lower therate at which women kill their husbands.The authors contend that women with bet-ter alternatives are more likely to use themrather than resort to killing their abusers toprotect themselves.

2. Straus (1994) and Yllo and Straus (1990) use agender equity index to show that states exhibiting moreinequality have higher rates of abuse (although Yllo andStraus find a curvilinear relationship). Levinson (1989)finds that countries with high gender inequality also havehigher rates of wife abuse.


III. DATA

The U.S. Census Bureau on behalf of theBJS undertakes the NCVS annually. The pri-mary goal of the NCVS is to collect detailedcrime data for calculation of national esti-mates of crime rates by both the type of crimeand for various subpopulations. In additionto detailed information on each incident ofcrime reported, the data include basic demo-graphic information, such as gender, race,education, age, and income. The NCVS sam-ple is made up of about 50,000 housing unitsselected with a stratified, multistage cluster-ing design. The sample design results in indi-viduals living in large, metropolitan areasbeing overrepresented in the sample. TheNCVS, previously called the NCS, was sig-nificantly redesigned in 1992. Consequently,NCS data prior to 1992 are not compatiblewith the 1992–98 surveys. One of the mainpurposes of the 1992 redesign was to gener-ate more accurate data on the incidence ofdomestic violence by adding several probingquestions about acts of violence committedby known and intimate offenders. The newerdesign increased the rates at which victimsreported domestic violence and consequentlyprovides more accurate estimates of the rateof intimate partner violence.

Although the public-use NCVS (ICPSR6406) does not include geographical iden-tifiers, through a grant from the NationalConsortium for Violence Research in coop-eration with the Census Bureau and the BJS,the authors were granted access to the Area-Identified NCVS data. The Area-IdentifiedNCVS has standard geographical area iden-tifiers for all housing units sampled includ-ing, most important for our purposes, countyidentifiers. The county identifiers allowed usto merge community-level variables with eachindividual’s data.

The final sample used for our analysis is525,615 observations on women ages 18 andover from the 1992–98 NCVS. It is importantto note that the NCVS is a panel, interview-ing the same households over several timeperiods, and, as a result, the data set doesnot represent 525,615 different women. Weidentified the women who were victims ofintimate partner abuse using the DOJ defini-tion: a victim of violent crime (including rape,sexual assault, robbery, aggravated assault,and simple assault) committed by a current

or former spouse or boyfriend. Women whohad experienced such an incident in the pastsix months (the NCVS takes incident reportson all crimes within the last six months)were categorized as “battered women.” Thefinal sample includes 1863 observations onbattered women.3 Table 1 compares the char-acteristics of these women with the charac-teristics of the other women in the sample.

The descriptive statistics are consistentwith those reported by the DOJ (2000).Women who are victims of domestic violenceare, on average, younger, have more children,and are more likely to be employed thanother women. Black women, women with lit-tle formal education, and women who live inhouseholds with relatively low incomes areoverrepresented in the sample of batteredwomen. Strikingly, divorced and separatedwomen are much more likely to report beingabused than are married women. Fifty-onepercent of the sample of battered women aredivorced or separated, compared with only13% of the other women. However, this dif-ference could simply reflect that divorced orseparated women are more likely to be will-ing to report abuse than are married women.

To examine the effects of community vari-ables on the incidence of domestic violence,the authors used the state and county FederalInformation Processing Standards codes inthe Area-Identified NCVS to merge county-level variables to the individual-level data.The authors chose the county as our unitof analysis because all of the households inthe NCVS have county identifiers (not thecase for some of the smaller area identifiers).The county variables are proxies for the eco-nomic alternatives and status of women in thecommunity. Services for battered women maypresent them with improved options outsidethe relationship. Consequently, the authorsused the 1994 National Directory of Domes-tic Violence Programs to create a countyspreadsheet of the existence and number of

3. Approximately 0.3% of our sample of womenover age 18 reported to be victims of domestic violence.This is about half of the rate of 7.5 per 1000 reportedby the BJS Statistics using the same data. The discrep-ancy resulted because this article categorizes a woman asbattered if she has been a victim in the past six months,whereas the DOJ generates annual estimates.


TABLE 1Average Characteristics of Battered Women Compared with Other Women

Women Reporting Abuseby Intimate Partner Other Women

(n= 1863) (n= 523,752)

Continuous variablesAge 31�20 46�03# children under age 12 1�12 0�51# children ages 12–17 0�31 0�25

Dummy variables (%)Employed 61 55Race

White non-Hispanic 74 77Black 16 11Hispanic 8 8Other minority race 2 4

Marital statusMarried 16 57Divorced or separated 51 13Single 33 30

EducationLess than high school 22 18High school degree (only) 67 60College degree or more 10 20Education not reported 1 2

RegionNew England 4 5Middle Atlantic 11 15South Atlantic 14 18East North Central 19 17West North Central 6 7East South Central 6 6West South Central 13 11Mountain 9 5Pacific 18 16

IncomeHH income less than $10,000 29 12HH income between $10,000 and $20,000 22 16HH income between $20,000 and $40,000 23 26HH income over $40,000 16 31HH income not reported 10 14

programs providing services to help batteredwomen in each county.4 We supplementedthe information in the directory by callingthose programs that were listed in the direc-tory but did not provide information onthe types of services provided. The authorsused the directory and supplemental infor-mation to create variables for the types ofservices offered by the programs, includ-ing hotlines, shelters (including the numberof beds), safe homes, counseling, emergencytransport, rape counseling, programs forvictims’ children, programs for batterers, and

4. The directory is a periodic (not annual) publica-tion, therefore, the authors used the 1994 county datafor all seven years of individual-level data.

legal services. Research indicates that bat-tered women often rely on welfare paymentsas a means of escaping abusive relationships.5

Therefore, the authors also merge thestate-level average welfare payment (takenfrom selected Statistical Abstracts) to theindividual-level data.

Although a woman’s own education andemployment status are likely to have thestrongest impact on her economic status out-side the relationship, the economic status of

5. The U.S. General Accounting Office (1998) sur-veys the literature on the incidence of domestic violenceamong welfare recipients, which is shown to be signifi-cantly greater (between 16% and 56%) than that of thegeneral population (1–2%).


women in the community in which she livesmay also improve her options. Women wholive in areas where women are well repre-sented in the labor force and relatively edu-cated are likely to have more credible threatsof leaving abusive relationships. Thereforethe percent of women in the labor force andthe percent of women with college degreesdivided by the percent of men with collegedegrees are also merged with the individual-level data. The population of the county isalso included because women who live inmore urban areas are likely to have moreoptions. All of these county variables aretaken from selected editions of the Countyand City Data Book. Table 2 presents someselected statistics on the county variables.

A minority of counties has services for bat-tered women, but the majority of women livein counties where such services are available.For example, only 35% of U.S. counties havea shelter for battered women, yet 82% ofAmerican women live in counties with at leastone shelter. Some services are much more

TABLE 2Selected Characteristics of County Variables

Variable County Level Individual Level (Weighted)

County (of residence) has at least one � � �Program 42% 85%Shelter 35% 82%Hotline 40% 85%Safe home 8% 20%Legal services 30% 74%Counseling 39% 83%Rape counseling 21% 41%Batterer’s program 12% 43%Children’s program 27% 74%Emergency transportation 33% 71%

Average number in county (of residence)Programs 0.59 3.52Shelters 0.43 2.35Hotlines 0.53 3.02Safe homes 0.09 0.40Legal services 0.38 2.19Counseling programs 0.52 3.02Rape counseling programs 0.23 0.83Batterer’s programs 0.15 0.96Children’s programs 0.35 2.42Emergency transportation services 0.39 1.89

Weighted means (by individuals)AFDC payment $358Population 986,754Education ratio (% women with college 0.90

degrees/% men with college degrees)Female labor force participation rate 57%

likely to be available than others are. Coun-seling, hotlines, and shelters are the mostwidely available services, and safe homes andrape counseling are available to less than halfof all women.

One could argue that the costs of com-mitting abuse are also an important determi-nant of the incidence of domestic violence,and therefore any changes in these costscould explain the decline in domestic vio-lence in the 1990s. If this is the case, thecosts, or legal punishments, should also beincluded in the analysis. The authors do notinclude these costs here for two reasons.First, most changes in the legal environmentoccurred in the 1980s, and there has beenno significant shift during the time period ofour data. In fact, between 1984 and 1989,arrest rates rose 70% for minor assaults,indicating that police were increasing theirinterventions significantly before the 1990s. Asignificant year for law enforcement reformwas 1984, after the Minneapolis Domestic


Violence Experiment indicated that arrestserved as a deterrent for future abuse. Sweep-ing reform followed this research and therecommendations of the 1984 Attorney Gen-eral’s Task Force on Domestic Violence (seeSchmidt and Sherman, 1993). No such majorinitiatives occurred during the period underinvestigation in this paper. Second, since the1984 research, a large number of studieshave questioned the effectiveness of arrestas a deterrent. (See Buzawa and Buzawa,1996, for a survey.) In particular, Schmidtand Sherman (1993) criticize Sherman’s orig-inal work on the Minneapolis experiment,representing a major reversal of his findingsconcerning the efficacy of arrest as a deter-rent. Other recent studies including Gelles(1996) and Hirschel and Hutchison (1996)cast doubt on the role of punishment indeterring domestic abuse.

IV. RESULTS

To examine the causes of the decline indomestic violence in the 1990s, the authorsfirst examine the determinants of a womanreporting to be a victim of abuse. Variablesthat significantly lower (increase) the inci-dence of domestic violence that are alsotrending upward (downward) can be identi-fied as contributing factors to the decline inthe rate of domestic violence nationally. Insection IV.A, the authors identify the vari-ables that significantly impact the likelihoodof a woman reporting domestic abuse. Insection IV.B, the article examines whether ornot the significant variables are trending inthe right direction in terms of contributingto the decline in the incidence of domesticviolence.

A. The Determinants of Domestic Violence

Table 3 summarizes the results from aprobit estimation of the determinants of awoman reporting to be a victim of intimatepartner abuse.6 Whether or not each variableis a significant determinant of the incidence

6. These results were generated from a probit esti-mation because the dependent variable is categorical(1 = abused by an intimate partner, 0 = not abusedby an intimate). The NCVS is a stratified, multi-stage cluster sample and as a result is not entirelyrandom. Consequently, weighting regression with app-ropriate corrections of the standard errors is required to

of domestic violence (at 95% confidence)is reported in the table. For the significantvariables, whether or not the variable has apositive or negative effect on abuse is alsoreported. The full results from the probit esti-mation are presented in the appendix.7

As the results indicate, most of the ser-vice variables are not significant factors inexplaining the incidence of domestic abusein the NCVS data. With the exception oflegal services, none of the services specif-ically designed to help victims of domes-tic violence impact the likelihood of abuse.Women living in counties with shelters, hot-lines, safe homes, emergency transportation,programs for batterers, children’s programs,and counseling are not significantly less likelyto be victims of intimate partner abuse thanwomen who live in counties without theseservices.8 However, women who live in coun-ties with legal assistance programs to helpbattered women are significantly less likelyto report abuse. Because legal services helpwomen with practical matters (such as pro-tective orders, custody, and child support)they appear to actually present women withreal, long-term alternatives to their relation-ships. These results reject the DOJ (2000)claim that an increase in services provided for

obtain unbiased estimates. Weighting reduced the sam-ple size to 530,487 because weights are missing for somewomen. The estimates could also be biased by the panelnature of the sample. Including multiple observationson the same woman is likely to causes correlation ofthe errors. The authors also corrected for this potentialproblem.

7. The results from an alternative specification ofthe model that omits potentially endogenous variables(marital status, woman’s employment status, and house-hold income) are also included and discussed in theappendix. Marital status is the most likely to be endoge-nous because women who are married may be less likelyto admit to being abused. Reestimating the model with-out this one variable results in no significant changes inthe results (in terms of signs and significance).

8. In other specifications of the empirical model, theauthors substituted the existence of any domestic vio-lence program in the county and the number of programsproviding services in a county for the types of servicesthese programs offered. Although there were no signif-icant changes in the signs and significance of the othervariables, neither the existence of a domestic violenceprogram (equal to 1 if any of the nine service variablesare equal to 1) nor the number of these programs wasa significant predictor of the incidence of domestic vio-lence. Replacing the existence of a shelter with the num-ber of shelter beds in the county also did not significantlyalter the results.


TABLE 3Probit Results (n= 530,487)

Significant Sign

Individual-level variablesAge Yes −Black Yes −Hispanic Yes −Other minority race Yes −High school only NoCollege degree Yes −Education not reported NoNumber of children< age 12 Yes +Number of children between ages 12 and 18 Yes +Employed NoMarried Yes −Household income, <$10,000 Yes +Household income, $10,000< x < $20,000 Yes +Household income, $20,000< x < $40,000 Yes +Household income not reported Yes +

Regional variablesMiddle Atlantic NoSouth Atlantic NoEast North Central Yes +West North Central NoEast South Central Yes +West South Central Yes +Mountain Yes +Pacific Yes +

County variables% females in labor force NoCollege education gender ratio (female/male) Yes −Number of shelters NoNumber of safe homes NoNumber of emergency transport services NoNumber of hotlines NoNumber of counseling programs NoNumber of rape counseling programs NoNumber of batterers’ programs NoNumber of children’s programs NoNumber of legal services programs Yes −AFDC average payment Yes +Population No

Time No

battered women is the major explanation forthe recent decline in the reported incidenceof domestic violence.

Women’s economic status is also found tobe a significant predictor of the likelihoodof abuse. The gender education ratio (% ofwomen with college degrees/% of men withcollege degrees) has a negative and signifi-cant effect on the likelihood that a womanreports that she is a victim of intimate part-ner abuse. It is interesting to note that whenthe authors entered both the percentage ofwomen and men holding college degrees asregressors instead of the education ratio,

women’s educational attainment had a sig-nificant and negative effect on domestic vio-lence, whereas the men’s variable significantlyincreased the incidence of abuse. Economicstatus at the individual-level also matters.More educated women and women who livein high-income households are less likely tobe victims of abuse.

The results on some of the otherindividual-level variables offer additional pol-icy implications and insight into the dynamicsof spousal abuse. Several of the results sup-port the descriptive statistics on domestic vio-lence presented by the BJS. Younger women,


women with young children, and women wholive in low-income household are more likelyto be victims of intimate partner abuse. How-ever, although the incidence of domestic vio-lence is higher among black women, onceone controls for marital status, black women,like other minority women, are less likelyto be abused than are non-Hispanic whitewomen. The negative effect of being marriedon abuse is likely both the result of selection(if a woman is abused, the relationship is lesslikely to be intact) and reporting (women aremore likely to report abuse if they are notliving with the abuser).

The most surprising result we find is thatthe generosity of welfare in the state of res-idence actually had a significant and positiveeffect on the likelihood of abuse. There areseveral possible explanations for this coun-terintuitive result. One explanation is thatwomen who leave abusive relationships maymigrate to states that are more generouswith welfare. As summarized in a U.S. Gen-eral Accounting Office (1998) report, stud-ies indicate that a significant proportion ofwomen on welfare, between 55% and 65%,report having been abused by an intimatepartner in the past. In addition, Blank (1988),Enchautegui (1997), and Borjas (1999) allfind that the poor do tend to migrate tostates that offer more generous welfare pay-ments. Another explanation is that stateswith greater incidences of domestic violencehave responded with increasing welfare pay-ments. However, it should be noted thatdropping the state welfare payment fromthe regression does not significantly alterthe coefficients on the remaining variables.Therefore, endogeneity does not appear tobe a significant problem. Finally, state wel-fare generosity is the only state-level vari-able included in the regression. Thereforethe positive relationship between welfare pay-ments and the likelihood of abuse may resultbecause some important state-level determi-nants of the incidence of domestic violenceare omitted from the regression. Whateverthe explanation for this result, it is clear thatmore work needs to be done on the rela-tionship between welfare and domestic abuse.The high rate of victimization among womenon welfare indicates that welfare is used asan escape route for many battered women.Clearly, the approach of simply including the

variance in generosity across states as an indi-cator of abuse in a wider analysis of theeffects of community-level variables is miss-ing part of the story.

The results also show regional variationin the incidence of domestic violence. Rela-tive to women living in New England (as wellas the Middle Atlantic, South Atlantic, andWest North Central), those who live in theEast North Central, East South Central, WestSouth Central, Mountain, and Pacific regionsof the country are more likely to reportabuse. These results likely reflect differencesin attitudes and values across regions as wellas any systematic regional variation in pro-grams that offer economic alternatives to bat-tered women not included as regressors.

The time trend variable is not a signifi-cant predictor of the likelihood of abuse oncethe control variables are added as regres-sors. This result suggests that it is changesin the individual- and community-level vari-ables over time that is generating the declinein the incidence of domestic violence in theUnited States and that the important vari-ables for explaining this decline are includedin the current analysis. The following sectionexamines the national trends in these signifi-cant variables.

B. Trends in the Significant Variables

The best way to examine the causes ofthe reported decline in domestic violenceover time in the United States would beto use time-series data on the incidence ofdomestic violence in the United States. How-ever, as discussed in section III, the NCVSwas redesigned in 1992 to improve report-ing of domestic violence. Consequently, dataprior to this time is not reliable, and there-fore a significant time-series of estimates ofthe rate of domestic violence in the UnitedStates does not exist. The authors attemptto estimate the determinants of the reporteddecline by using a panel data set at theindividual-level to first isolate the importantdeterminants of the incidence of domesticviolence and then to examine whether ornot these significant variables are trendingin the right direction to possibly contribute tothe decline in the rate over time. Althoughthe authors cannot say that the variables thatare both significant determinants of the inci-dence of domestic violence at the individ-ual level and trending in the right direction


nationally are definitively the causes of thetrend downward in the rate of domestic vio-lence, they are likely suspects.

Most services provided to help batteredwomen do not impact the likelihood of abuse,but the provision of legal services significantlylowers the incidence of domestic violence.However, for legal services to contribute tothe decline in domestic violence in 1990s,the provision of legal services for batteredwomen must have increased over this timeperiod. According to the 1986 National Direc-tory of Domestic Violence Programs, in 1986there were 336 legal services programs serv-ing victims of domestic violence. By 1994, thenumber increased to 1190 programs nation-wide, an increase of 254%! Between 1994and 2000, the number of legal programs forbattered women increased to 1441 programs.Clearly, the expansion of legal assistance tobattered women has accounted for part of thedecline in the incidence of domestic violencenationwide. The expansion of legal serviceshas mostly resulted from existing programsfor victims of intimate partner abuse addinglegal services to their lists of services pro-vided as opposed to new programs open-ing their doors. Part of the credit for theexpansion of legal services goes to the fed-eral government. With the passage of theViolence against Women Act in 1994, thefederal government made a commitment tomeet the needs of women who are victimsof violence. One grant program set up torespond to the legal needs of women whoare victims of intimate partner violence isthe Domestic Violence Victims’ Civil LegalAssistance Program. Since 1998, this programhas provided more than $60 million to morethan 200 nonprofit, nongovernmental organi-zations to provide civil legal services to vic-tims of domestic violence.

Another important determinant of thelikelihood of a woman reporting abuse is edu-cation. Both the woman’s own educationalstatus (having a college degree) and the rel-ative education of women in her community(the percent of women in the county with col-lege degrees relative to the percent of men inthe county with college degrees) significantlyimpact the likelihood of abuse. From 1993 to1998, the percentage of women nationwidewith college degrees increased from 17.9%to 20.7%, an increase of almost 16%. Theincrease was even more significant for black

women increasing from 10.9% to 13.6%, rep-resenting more than a 35% increase in theproportion of college-educated black women.The increase in women’s educational attain-ment over the time period is likely to play asubstantial role in diminishing the incidenceof domestic abuse.

Household income is also found to be asignificant predictor of the likelihood thatthe woman is a victim of domestic violence.Household income is likely to be signifi-cant both because women are more likelyto have earnings, and therefore economicpower, in households with higher incomesand because as men’s incomes (and educa-tion status) increase, they are less likely toabuse their partners (see Farmer and Tiefen-thaler, 2000). The suggestion that women’seconomic power is at least partially the expla-nation for the negative relationship betweenhousehold income and domestic violence issupported by the fact that when householdincome is dropped from the list of indepen-dent variables, women’s employment statushas a significant and negative effect on thelikelihood of abuse.

According to census statistics, over theperiod from 1993 to 1998, both medianhousehold income and women’s medianincome increased. Median household incomein 1993 (1998 adjusted dollars) was $35,241.By 1998 that figure has risen to $38,885,an increase of 10.3%. The economic pros-perity of the 1990s has been cited as alikely cause of the decrease in the overallcrime rate, and it appears to have been acause of the decrease in domestic violenceas well. Women’s median income increasedalmost 18% from $13,800 to $16,258 (1998adjusted dollars). Not only did women’s realincome rise, indicating an increase in outsideoptions, but women’s income as a percentageof men’s income rose over this period as well.In 1993 women’s median income was 49.7%of men’s, whereas in 1998 that figure hadincreased to 53.0%. An increase in women’searning power relative to men’s implies thatwomen have both more opportunities forself-sufficiency and fewer gains from mar-riage. Both of these factors will increase thelikelihood that women leave abusive relation-ships. Finally, women’s labor force partici-pation rate rose two percentage points from58.5% in 1993 to 60.5% in 1998.


One of the most important demographictrends of the twenty first century will be theincreasing age of the populations in devel-oped countries. This demographic trend,which started in the late twentieth century,appears to have an important impact onthe incidence of domestic violence. Youngerwomen experience significantly more violencewith women between the ages of 20 and 24most likely to be victims. During the 1990s,the percentage of women in this age categoryfell from 35.8% in 1993 to 32.0% in 1998.This decline of 3.8 percentage points repre-sents a 10.6% decline in the percentage ofthe U.S. population that belongs to the high-est risk age group. As the population contin-ues to age, the incidence of domestic violencewill continue to decline.

Race is also a significant factor in thereported incidence of abuse and the chang-ing racial composition of the current popu-lation, another important demographic trendthat will continue in the twenty first century,appears to be another significant factor inexplaining the decline in domestic violence.Although the BJS report suggests that blackwomen are more likely to be abuse victims,once other factors such as income and maritalstatus are controlled for, this study finds thatblack women are less likely to be abused thantheir non-Hispanic white counterparts. Simi-larly, we find that Hispanic women as well aswomen of all other races are less likely to bevictims than non-Hispanic white women. Thepercentage of the female population that iswhite non-Hispanic fell 1.2 percentage points,from 38.1% to 36.9%, between 1993 and1998. This decline represents a significantdemographic shift over such a short period oftime. Holding other factors constant, as thepopulation continues to become more raciallydiverse, the reported incidence of domesticviolence should continue to decline. How-ever, if the increase in racially diversity is theresult of increased immigration and there-fore is accompanied by lower educational sta-tus among women (and men) and more poorhouseholds, one is not likely to see a declinein intimate partner abuse.

The number of children that a woman hasis also a significant determinant of abuse.Women with children, especially young chil-dren, are more dependent on their relation-ships and have fewer alternatives for self-sufficiency outside their relationships and

consequently are more likely to be abused.Although the fertility rate declined dramati-cally in previous decades, there was no sig-nificant change in this variable in the 1990s,and therefore it is not a significant factor inexplaining the decline in domestic violence.

The regression results indicate that mar-ried women are less likely to report beingabused. During the period from 1993 to 1998,the percentage of married women fell from56.5% to 54.9%, representing a 1.6 percent-age point drop. As a result, the authorsexpect that the increase in the percentagesof divorced and separated women wouldincrease the rate of reported intimate part-ner abuse. Clearly, any positive effect of theincrease in divorce on domestic violence isoutweighed by trends in other more impor-tant variables.

In addition to individual characteristics,this article finds that some community vari-ables matter as well. Geographic location,possibly indicative of cultural norms and atti-tudes, is a significant predictor of the likeli-hood of abuse. Specifically, women living inthe East North Central, East South Central,West South Central, Mountain, and Pacificregions report more abuse than women liv-ing in New England, the Middle and SouthAtlantic, and the West North Central regionsof the country. The states with the great-est population increases over in the 1990sare in the Pacific, Mountain, and SouthAtlantic regions. Although women living inthe Pacific and Mountain regions experiencesignificantly more violence, those living inthe South Atlantic do not. However, giventhe magnitude of the population increasesin the Pacific and Mountain regions relativeto the rest of the United States (8 of 10states with the greatest population growth arein these two regions), this variable does notgenerally support the trend to lower ratesof abuse and may in fact be working in theopposite direction.

Finally, the authors find that as welfarepayments rise, so does abuse. Given that the1996 welfare reform has decreased the over-all generosity of most states’ welfare pro-grams, this result is consistent with a lowernational rate of abuse. However, the positiverelationship between abuse and welfare gen-erosity is a curious result with no theoreticalbasis. It may be explained by women leavingabusive relationships migrate to states that


offer more generous support. In any case,one should be very careful in citing decreasedwelfare generosity as a source of the nationaldecline in domestic violence.

This analysis indicates that morewidespread provision of legal services forbattered women, improved educational andeconomic status for women, and demo-graphic trends explain the decline in domes-tic violence in the 1990s. With the exceptionof migration to the Mountain and Pacificregions of the country, a slight decline inmarriage rates, and the fact that the averagenumber of children has remained constantin the 1990s, every significant variable inour regression is trending in a direction thatpredicts a lower rate of abuse.

The psuedo-R2 from the estimation of thefull model indicates that the right-hand-sidevariables explain only 14% of the variation inthe dependent variable, the probability of anindividual woman being a victim of domes-tic violence. Certainly many other importantvariables, such as the abuser’s and the vic-tim’s family backgrounds, have been omit-ted (this information is not included in theNCVS). However, these variables are likelyto be more important in explaining why somewomen are abused and others are not at apoint in time than in explaining the declinein domestic violence over time. As previouslydiscussed, the incidence of domestic violencedeclined by 21% between 1993 and 1998 from1.1 million violent incidents to 876,340 inci-dents. Therefore, the probability of an indi-vidual woman being abused fell from 0.011%to 0.0084%, a decline in the likelihood ofbeing abused of 0.0026%. Using the partialderivatives from the full model (presented inthe appendix) and the change in the aver-age national statistics for the dependent vari-ables in 1993 and 1998,9 the key variables—the improvement in women’s educational andeconomic status, the increase in the num-ber of legal programs, and the aging ofthe population—explain approximately 22%of the decline (almost 50,000 incidents) indomestic violence over this time period.Clearly, future work that includes time-seriesanalysis would produce more accurate esti-mates. However, these results suggest thatpast trends that are likely to continue into

9. Because of data limitations, the authors use 1994and 1999 data for the number of legal service programs.

the twenty first century will have a significantimpact on the incidence of domestic violencein the United States in the future.

V. CONCLUSION

The goal of this article is to ascertainthe source of the decline in intimate part-ner abuse from 1993 to 1998 cited in arecent report by the BJS. The authors findthree significant factors in explaining thedecline: (1) the increased provision of legalservices for victims of intimate partner abuse,(2) improved educational and economic sta-tus for women, and (3) demographic trends,including the aging of the population and anincrease in racial diversity. Specifically, U.S.women have become older, more educated,richer, and more likely to belong to a minor-ity race. In addition, the provision of legalservices has exploded in the past decade.All of these factors are significant determi-nants of abuse and are trending in the rightdirection to explain the reduced incidence ofdomestic violence in the United States.

These results have important policy impli-cations given that factors that explain thedecline in domestic violence in the 1990sare likely to be key variables in continuingto lower abuse in decades to come. First,the availability of legal services has a sig-nificant negative effect on the incidence ofabuse. The Violence against Women Act hasbeen an important impetus for funding inthe area of civil legal assistance. The con-tinued expansion of the availability of civillegal services will likely continue to lowerthe incidence of intimate partner abuse inthe future. Other services—hotlines, shelters,job training, outreach, and counseling—arenot significantly related to women’s reportsof domestic violence in the NCVS, but giventheir use, these services are clearly valu-able to battered women. However, expansionof these services should focus on providingwhat clients need to become self-sufficient.Although short-term housing and counsel-ing provide women with important temporarysafety and support (and therefore may signifi-cantly reduce the incidence of serious injuriesresulting from domestic violence), they aremore likely to lower the overall incidence ofdomestic violence if they provide women with


long-term, realistic alternatives to their rela-tionships. Increased funding is necessary toprovide these types of services.

Women’s educational attainment, a keyindicator of economic status, is an impor-tant determinant of the likelihood of abuse.The authors find that a woman’s own formalschooling (a college degree) and the edu-cational attainment of women in her com-munity lower the likelihood that she reportsabuse. Therefore, continuing the trend ofimproving women’s education will likely bea key factor in eliminating domestic vio-lence. Many women have gained access tohigher education through community col-leges and part-time college enrollment. Fund-ing to subsidize community colleges andpolicies that make it easier for women toenroll in college (for example, subsidizedday care for students) would likely generate

TABLE A-1Probit Results (n= 530,487)

Model 1 Model 2

Individual-level variablesAge −0�0212 �−0�000115�∗∗ −0�0167 �−0�000744�∗∗Black 0�0623 �0�000364�∗∗ −0�1372 �−0�000525�∗∗Hispanic −0�2388 �−0�000984�∗∗ −0�2852 �−0�000908�∗∗Other minority race −0�3081 �−0�001128�∗∗ −0�3399 �−0�000971�∗∗High school only −0�1212 �−0�000686�∗∗ −0�0052 �−0�000023�College degree −0�3321 �−0�001397�∗∗ −0�1293 �−0�000515�∗∗Education not reported −0�2059 �−0�000844�∗∗ 0�0648 �0�000318�Number of children<age 12 0�0899 �0�000489�∗∗ 0�1507 �0�000671�∗∗Number of children between ages 12 and 18 0�0079 �0�000043� 0�0325 �0�000145�∗∗Employed 0�0323 �0�119�Married −0�6168 �−0�003422�∗∗Household income, <$10�000 0�2797 �0�001729�∗∗Household income, $10�000 < x < $20�000 0�2090 �0�001161�∗∗Household income, $20�000 < x < $40�000 0�0711 �0�000333�∗∗Household income not reported 0�1056 �0�000531�∗∗

Regional variablesMiddle Atlantic −0�0131 �0�801� 0�0201 �0�732�South Atlantic 0�0402 �0�516� 0�0787 �0�256�East North Central 0�1062 �0�049�∗∗ 0�1540 �0�010�∗∗West North Central 0�0721 �0�232� 0�0907 �0�188�East South Central 0�1074 �0�161� 0�1938 �0�026�∗∗West South Central 0�1865 �0�009�∗∗ 0�2170 �0�006�∗∗Mountain 0�2320 �0�00173�∗∗ 0�2492 �<0�001�∗∗Pacific 0�1253 �0�008�∗∗ 0�1796 �0�001�∗∗

County and state variables% females in labor force −0�0022 �−0�159� 0�0010 �0�602�College education gender ratio (female/male) −0�2274 �−0�003�∗∗ −0�2135 �0�010�∗∗Shelter −0�0039 �−0�833� −0�0193 �0�353�Hotline 0�0196 �0�271� 0�0200 �0�292�Counseling 0�0034 �0�886� 0�0105 �0�703�Batterers’ program 0�0049 �0�713� 0�0163 �0�254�

continued

many positive outcomes for society, includ-ing a lower incidence of domestic violence.Women’s employment and earnings are alsoimportant in providing them with alterna-tives to their abusive relationships. There-fore, policies to eliminate the wage gap wouldalso likely have the desired impact on theincidence of violence against women.

APPENDIX

The results from two separate regressions are includedin Table A-1. Model 2, which generated the resultsreported in Table 3, includes all variables that are likelyto affect the likelihood of abuse. Model 1 omits thosevariables—employment status, household income, andmarital status—that may be endogenous. The authorsreport the results both with and without the potentiallyendogenous variables instead of attempting IV tech-niques because the NCVS does not include variables to


TABLE A-1 continued

Model 1 Model 2

Children’s program 0�0034 �0�817� 0�0059 �0�711�Rape counseling −0�0007 �−0�955� 0�0083 �0�558�Emergency transportation 0�0158 �0�245� −0�0027 �0�855�Safe home −0�0254 �−0�164� −0�0324 �0�102�∗Legal services −0�0262 �−0�031�∗∗ −0�0279 �0�032�∗∗AFDC average payment 0�0003 �0�015�∗∗ 0�0003 �0�033�∗∗Population −0�0195e-06 �−0�293� −0�0213e-07 �0�918�

Time trend −0�0015 �−1�426� 0�0003 �0�240�Constant −1�677∗∗ −2�1156 �<0�001�Psuedo R2 0�09 0�14

∗ Significant at 10%.∗∗ Significant at 5%.

use as proper instruments. The probit coefficients arefollowed by the partial derivatives (evaluated at the sam-ple means) in parenthesis.10

REFERENCES

Allen, C., and M. A. Straus. “Resources, Power andHusband-Wife Violence,” in The Social Causes ofHusband-Wife Violence, edited by M. A. Strausand G. T. Hotaling. Minneapolis: University ofMinnesota Press, 1980.

Blank, R. M. “The Effect of Welfare and Wage Levels onthe Location Decisions of Female-Headed House-holds.” Journal of Urban Economics, 24, 1988,186–211.

Borjas, G. J. “Immigration and Welfare Magnets.” Jour-nal of Labor Economics, 17(4), 1999, 607–37.

Buzawa, E., and C. Buzawa. Do Arrests and Restrain-ing Orders Work? Newbury Park, CA: Sage Publica-tions, 1996.

Coleman, D. H., and M. A. Straus. “Marital Power, Con-flict and Violence in a Nationally RepresentativeSample of American Couples.” Violence and Vic-tims, 1(2), 1986, 141–57.

Dugan, L., D. Nagin, and R. Rosenfeld. “Explainingthe Decline in Intimate Partner Homicide: TheEffects of Changing Domesticity, Women’s Sta-tus, and Domestic Violence Resources.” WorkingPaper, National Consortium on Violence Research,1998.

Enchautegui, M. E. “Welfare Payments and OtherDeterminants of Female Migration.” Journal ofLabor Economics, 15, 1997, 529–54.

Farmer, A., and J. Tiefenthaler. “An Economic Analysisof Domestic Violence.” Review of Social Economy,55(3), 1997, 337–58.

. “The Employment Effects of Domestic Vio-lence.” Colgate University Working Paper 100-04,2000.

10. The partial derivatives presented for dummy vari-ables are for the discrete change from 0 to 1. Inevaluating the percentage contribution of the significantvariables in explaining the decline in domestic violence,the regular partials were used.

Gelles, J. “Abused Wives: Why Do They Stay?” Journalof Marriage and the Family, 38, 1976, 659–68.

. “Constraints against Family Violence: How WellDo They Work?,” in Do Arrests and RestrainingOrders Work?, edited by E. Buzawa and C. Buzawa.Newbury Park, CA: Sage Publications, 1996,30–42.

Hirschel, D., and I. Hutchison. “Realities and Implica-tions of the Charlotte Spousal Abuse Experiment,”in Do Arrests and Restraining Orders Work?, editedby E. Buzawa and C. Buzawa. Newbury Park, CA:Sage Publications, 1996, 54–82.

Kalmuss, D., and M. Straus. “Wife’s Marital Dependencyand Wife Abuse,” in Physical Violence in AmericanFamilies, edited by M. A. Straus and R. J. Gelles.New Brunswick, NJ: Transaction Publishers, 1990.

Levinson, D. Family Violence in Cross-Cultural Perspec-tive. Newbury Park, CA: Sage Publications, 1989.

Pagelow, M. D. Woman-Battering: Victims and TheirExperiences. Beverly Hills, CA: Sage Publications,1981.

Rennison, C. M., and S. Welchans. “Intimate Part-ner Violence.” Bureau of Justice Statistics SpecialReport, May 2000, NCV 178247.

Schmidt, J., and L. Sherman. “Does Arrest DeterDomestic Violence?” American Behavioral Scientist,36(5), 1993, 601–9.

Straus, M. A. “The National Family Violence Surveys,”in Physical Violence in American Families, edited byM. A. Straus and R. J. Gelles. New Brunswick, NJ:Transaction Publishers, 1990.

. “State-to-State Differences in Social Inequalityand Social Bonds in Relation to Assaults on Wivesin the United States.” Journal of Comparative Fam-ily Studies, 25(1), 1994, 7–24.

Tauchen, H. V., A. D. Witte, and S. K. Long. “Domes-tic Violence: A Nonrandom Affair.” InternationalEconomic Review, 32(2), 1991, 491–511.

Tjaden, P., and N. Thoennes. “Prevalence, Incidence, andConsequences of Violence against Women: Find-ings from the National Violence against WomenSurvey.” Research in Brief, U.S. Department of Jus-tice, National Institute of Justice, Washington, DC.NCJ 172837, 1998.

U.S. Department of Justice, Bureau of Justice Statis-tics. “Intimate Partner Violence against Women


Declined from 1993 through 1998.” Bureau of Jus-tice Statistics Press Release, May 17, 2000. Availableonline at www.ojp.usdoj.gov/bjs/pub/press/ipv.pr.

U.S. General Accounting Office. Domestic Violence:Prevalence and Implications for Employment amongWelfare Recipients. GAO/HEHS-99-12. November1998.

Yllo, K. A., and M. A. Straus. “Patriarchy and Vio-lence against Wives: The Impact of Structural andNormative Factors,” in Physical Violence in Amer-ican Families: Risk Factors and Adaptations to Vio-lence in 8,145 Families, edited by M. A. Strausand R. J. Gelles. New Brunswick, NJ: TransactionPublishers, 1990.

Explaining the Recent Decline in Domestic Violence

Documents

Transcript of Explaining the Recent Decline in Domestic Violence