10.1177/1049731504272908RESEARCH ON SOCIAL WORK PRACTICE Hetling, Born / FAMILY VIOLENCE OPTION AND WELFARE

Examining the Impact of the Family ViolenceOption on Women’s Efforts to Leave Welfare

Andrea HetlingCatherine E. Born

University of Maryland, Baltimore

Objective: The establishment of the Family Violence Option (FVO) in 1997 was met with some controversy, as criticsbelieved waivers from time limit and work requirements would hinder women’s ability to leave welfare and findemployment. Method: Using administrative and interview data from Maryland, multivariate equations analyze ifdomestic violence disclosure, administrative documentation, or waiver use had a statistically significant affect onone year employment and welfare use outcomes of individuals. Results: Waiver holders did not differ from nonvictims,but victims who are not documented received fewer months of welfare and earned less income. Conclusions: Findingsdo not indicate that FVO waivers encourage women to stay on welfare longer. However, the poor outcomes of undocu-mented victims indicate that some individuals may be slipping through the cracks of a well-intentioned policy.

Keywords: domestic violence; Family Violence Option; welfare receipt; employment

In 1997, while reforming federal welfare programs, theU.S. Congress established the Family Violence Option(FVO) to prevent reforms from adversely affecting thosewelfare recipients who are domestic violence victims.The FVO requires states that choose to adopt the option toscreen and identify abuse victims, make service referrals,and offer waivers from program requirements that mayunfairly penalize or endanger current or former victims ofabuse. This research investigated if known domestic vio-lence victims, and in particular those who received anFVO waiver, remained on the welfare rolls longer andworked less than other welfare recipients, as many criticsspeculated.

The purpose of the research was to assess what differ-ence the FVO, in practice, has made on women’s ability toleave welfare and find employment. The major researchquestion investigated whether an experience of violenceor, more specifically, one’s classification as a waiver user,waiver nonuser, or narrative discloser affected individual

self-sufficiency as reflected by participation in cash assis-tance programs and securing employment. In otherwords, controlling for a number of individual, agency,and community characteristics, did the welfare oremployment outcomes of domestic violence victims,including those who held an FVO waiver, differ fromthose of other welfare recipients?

BACKGROUND

Theoretical approaches and anecdotal accounts sug-gest that domestic violence is a substantial barrier formany welfare recipients in achieving self-sufficiency.Sociologists and psychologists, in particular, haveoffered extensive studies showing that battered womenface a multitude of barriers in escaping the violence intheir lives and gaining economic independence (Kantor& Jasinski, 1998; Pagelow, 1981; Walker, 1979). Lackingfinancial resources or the control of those resources,abused women are often faced with the dilemma ofchoosing violence or poverty (Davis, 1999; Raphael,1996). Although it is commonly accepted that escapingdomestic violence often forces women into poverty,lesser known are the reasons why these experiences per-petuate poverty even after an escape, or prevent womenfrom leaving welfare and entering work.

In addition to the time and energy needed to resolvelegal matters related to orders of protection, divorce, andchild custody, research in psychology and family studieshas determined other direct and indirect consequences ofintimate partner abuse that complicate the process of

143

Authors’ Note: A previous version of this article was presented at the 8thAnnual Society for Social Work and Research Conference on January 16, 2004.An expanded version of this study was also presented to the faculty of the Grad-uate School of the University of Maryland, College Park, in July 2002 in partialfulfillment of the requirements for the degree of Doctor of Philosophy. The dis-sertation was completed by Andrea Hetling under the direction of William A.Galston, School of Public Affairs, University of Maryland, College Park. Theresearch was supported by funding from the Maryland Department of HumanResources. Correspondence may be addressed to Andrea Hetling, PhD, FamilyWelfare Research and Training Group, School of Social Work, University ofMaryland, 525 West Redwood Street Baltimore, Maryland 21201; e-mail:ahetling@ssw.umaryland.edu.

Research on Social Work Practice, Vol. 15 No. 3, May 2005 143-153DOI: 10.1177/1049731504272908© 2005 Sage Publications

achieving independence. Abuse may harm physical ormental health, and severe beatings can cause injuries thatnecessitate immediate medical attention (Straus, 1986).Additionally, mental health problems, most notablyposttraumatic stress disorder (PTSD), may cause difficul-ties that interfere with work and training efforts (Brush,2000). Difficulties may include an inability to leave thehouse, to consider the future, to concentrate or to learnnew tasks, and to have social interactions (Murphy,1997). Even when women are physically and mentallyable to participate in either training or work, narrative evi-dence suggests that many abusive partners and formerpartners are threatened by these actions and attempt,often successfully, to sabotage these efforts (Brandwein,2000; Raphael, 2000).

As a result of the effect of abuse on both health andwork, many formerly battered women lack the necessaryeducation or work experience to find adequate employ-ment. Women who have recovered from injuries and havecompletely broken free from their abusers, those whohave settled legal matters, and those who are safe fromfurther violence may continue to have a difficult timeachieving self-sufficiency because of paltry work histo-ries and lack of education and training (Raphael &Tolman, 1997; Shepard & Pence, 1988).

Also of consequence for welfare programs is howcommon experiences of domestic violence are amongrecipients. Empirical research shows a high rate ofdomestic violence incidence among cash assistancerecipients (e.g., Tolman & Raphael, 2000). Surveys dem-onstrate that between 20% and 65% of women on welfareare current or past victims of domestic violence (Allard,Albelda, Colten, & Cosenza, 1997; Browne & Bassuk,1997; Curcio, 1997; Raphael, 1996). The rate is espe-cially stark in comparison with the general populationwhose reported prevalence rates are 1.5% for abusewithin the past 12 months and 24.8% for lifetime domes-tic violence occurrences (Tjaden & Thoennes, 1998).

If it is the case that the populations of domestic vio-lence victims and welfare recipients significantly overlapand that the barriers created by domestic violence are sub-stantial, then for welfare reforms to succeed, these pro-grams should be designed to meet the unique needs of thisgroup of women. It is with these theories in mind that theFVO was included in the 1996 federal welfare reform leg-islation, the Personal Responsibility and Work Opportu-nity Reconciliation Act (PRWORA). The FVO, a volun-tary state level option in the legislation, obligates states toscreen for domestic violence victims, requires them tooffer services and referrals to local battered women’sorganizations, and permits states to exempt battered

women from Temporary Assistance for Needy Families(TANF) program requirements (Personal Responsibilityand Work Opportunity Reconciliation Act, 1996).

Critics of the FVO claimed that FVO waivers takeaway incentives to become self-sufficient, leading waiverholders to stay on welfare longer and forgo opportunitiesto gain employment experience. Moreover, the languagemay be too broad and could allow for very minimal rea-sons to be accepted as exemptions from working orencourage false claims of domestic violence (Satel,1998). Last, critics claim that the FVO is unnecessarybecause in cases of severe violence, the law, without theoption, provides enough leeway to include those cases ashardship exemptions (Satel, 1998).

Reality, however, has not borne out either the largeprevalence numbers put forth by proponents or the num-ber of false claims feared by critics. Studies based on wel-fare program administrative data indicate that very fewwomen, approximately 1% of the active caseload, dis-close domestic violence to their caseworkers and evenfewer have opted for FVO waivers (Lennett, 1997;Raphael & Haennicke, 1999). There is a discrepancybetween the prevalence of domestic violence among wel-fare recipients based on disclosures to researchers anddomestic violence advocates, with conservatively one inevery four women estimated to be experiencing abuse,and the rate at which the FVO is used, with about one inevery hundred women receiving services, and this dis-crepancy has led to concerns. Questions regarding theappropriateness of the legislation itself as well as the vari-ety of ways it is being implemented locally are spurringresearch evaluating the policy and its screening and ser-vice delivery practices (Angelari, 1998; Burt, Zweig, &Schlichter, 2000; Raphael & Haennick, 1999). In addi-tion to this research, an understanding of the characteris-tics and outcomes of the women who have used the FVOis critical in deciphering who the policy is reaching andwhat effect it is having.

Because of mostly data restrictions and confidentialityconcerns, very few researchers have been able to examinethe demographic characteristics or outcomes of womenwho disclose and receive services or waivers under theFVO. Research based on women identified through sur-veys have examined the welfare outcomes of domesticviolence victims and indicate that women who have expe-rienced recent abuse have poorer employment experi-ences (Tolman & Rosen, 2001). However, comparablework on women who have disclosed to welfare casework-ers and thus are recipients of FVO services is not yetavailable.

144 RESEARCH ON SOCIAL WORK PRACTICE

The presented research evaluated the effect of the FVOon self-sufficiency outcomes using measures of employ-ment and welfare use of FVO waiver recipients, otherknown domestic violence victims, and nonvictims. Basedon findings from violence research showing that abusedand formerly abused women may have difficulty in rap-idly attaining financial self-sufficiency, we hypothesizedthat all domestic violence victims would have poorer wel-fare and employment outcomes than other welfare recipi-ents. Considering the leeway offered FVO waiver recipi-ents under the law, we hypothesized that these womenmay fare even worse than victims who did not hold awaiver especially considering that they may be in a moresevere or recent relationship to have qualified for awaiver.

METHOD

Using Maryland state administrative data for March1998 to June 2001 and interview data collected from wel-fare personnel, individual-level outcomes related to Tem-porary Assistance for Needy Families (TANF) participa-tion and employment were examined for four groups: (a)domestic violence victims who were administrativelymarked and held FVO waivers, (b) victims who wereadministratively marked but did not hold waivers, (c) vic-tims who were not administratively marked but whosecase narratives indicated abuse, and (d) TANF recipientswith no indication of abuse.

Sample

Three domestic violence victim samples and onenonvictim sample were identified as the analytic groupsof interest using the official Maryland state automatedadministrative data system, the Client Information Sys-tem (CIS), for cash assistance program participation. Allfour groups were comprised of households headed bywomen receiving TANF in Maryland for one or moremonths between March 1998 and June 2000. Each house-hold included an adult female payee or casehead in whosename the benefits were paid as well as children and otheradults in the household, most of whom were included inthe TANF grant. March 1998 was chosen as the beginningdate for the sample because conversion to a new adminis-trative data system was completed during that month andthe previous system did not record any information ondomestic violence.

First, the universe of Maryland welfare recipients whowere administratively marked as domestic violence

victims between March 1998 and June 2000 were identi-fied (n = 554). Administrative markers include: a yes orno field for domestic violence, a shelter field with a valueindicating that the individual was living in a domestic vio-lence shelter, and three program exemption fields withvalues indicating that the exemption reason was domesticviolence. Women with at least one administrative markerwere then divided into two groups: waiver holders (n =261, exempt from at least one of the three programrequirements—time limits, work requirements, or childsupport requirements—with or without the other admin-istrative fields marked) and waiver nonholders (n = 293,those with only a domestic violence mark or domesticviolence shelter value).

Second, a 5% random, proportionately stratified sam-ple (n = 3,781) of all other welfare recipients from thesame time period was drawn, ensuring a representativesample of seven regions of Maryland, the six regionsdefined by the Maryland Department of Planning, andBaltimore City, which was treated as a separate regionbecause of its size. Individual case narratives were readand coded to identify abuse within this sample. Case nar-ratives are typed notes entered by the welfare caseworker.In the automated system, caseworkers can access a freeform space in which they can annotate an ongoing narra-tive on each welfare case and their interactions withcaseheads. Cases with narrative indication of abuse wereflagged and comprise the third victim sample (n = 184)and are referred to as narrative disclosers. For comparisonpurposes, all other cases in the random sample were keptas a fourth analytic group. Although members of the com-parison group were not marked in either the administra-tive system or in their case narratives as having discloseddomestic violence and are referred to as nonvictims in thearticle, it is important to recognize the possibility thatthey did experience domestic violence but decided not todisclose the experience or that their caseworker did nottake note of it. Table 1 presents the description and size ofthe four samples, which, together, constitute a total studysample of 4,335 cases.

Statistical Models and Outcome Variables

Multiple regression statistical models explored 1-yearpostdisclosure outcomes and were designed to explainemployment and welfare use during the 1-year follow-upperiod. Specifically, the models were constructed toexamine if being identified as a domestic violence victimcan explain variation in the achievement of economicself-sufficiency and were analyzed using Stata 7.0. Themodels were based on the following equation:

Hetling, Born / FAMILY VIOLENCE OPTION AND WELFARE 145

Outcome = � + �Waiver + �Nonwaiver + �Narrative Disclosure +�XDem + �XAgency + �XJuris + � .

Waiver, nonwaiver, and narrative disclosure variablesrepresent the inclusion of the three domestic violence vic-tim groups in the model. The nonvictim group served asthe control or reference group and was thus not includedin the model. The effect of an FVO waiver, administrativedocumentation, and domestic violence disclosure is ex-plained in comparison to the outcomes of sample mem-bers in the nonvictim group. The results as discussed inthe text and presented in the tables reflect this design. Thematrices of demographic, agency, and jurisdictional vari-ables were included in the model as important indepen-dent variables that likely influence the outcome variablesand are discussed in more detail later on in this section.The statistical significance and effect size of these vari-ables are not discussed in the results section or in the cor-responding table, however, but noted simply as controlsbecause they are not central to the research question orhypotheses.

Variations of the model were used to explain follow-upcash assistance program participation and employmentexperiences. Three measures of cash assistance use wereconstructed using the CIS database. First, the months ofreceipt variable captures the total number of months ofbenefit receipt in the 12 months following the studymonth and ranges from 0 to 12. Mean months of receiptfor the entire sample is 7.34 months. After a number oftests, ordinary least squares (OLS) regression was used inthis analysis. Initially, a skewness-kurtosis test of thevariable indicated nonnormality. A number of transfor-mations (e.g., log, exponent, reciprocal) were attemptedto yield a more normal distribution, but none were suc-cessful. Thus, an ordinal logit model was used to test therobustness of the model. Given the similar results of themodel, the use of OLS regression models to explain thenumber of follow-up months of TANF was deemed suffi-cient. The raw coefficients are straightforward in this caseand are interpreted as a one unit change in the

independent variable leads to an x unit change in theoutcome or dependent variable.

Second, exit is a dichotomous variable measuring theproportion of families that experienced a consecutive60-day or longer exit from TANF during the follow-upperiod. Approximately half, 55.02%, of the entire sampleexperienced such an exit during the follow-up year.Third, among exiters, the recidivism variable indicateswhether the individual returned to TANF during the1-year follow-up period. Among all exiters, 10.19%experienced a return within the year. A probit model wasused to examine these two dichotomous variables; themarginal effect of the variable is presented in addition tothe raw coefficient for ease of interpretation. The margin-al effect is interpreted as a percentage point increase ordecrease in the outcome variable. Specifically, each addi-tional unit increase of the predictive variables leads to anx percentage point increase or decrease in the outcomevariable.

Turning to employment experiences, data were gath-ered from the Maryland Automated Benefits System(MABS), the official state database of employment cov-ered by Unemployment Insurance (UI). Again, threemeasures were analyzed. First, quarters worked is a countof the number of quarters an individual worked in aUI-covered job in Maryland during the four calendarquarter period following the quarter containing the studyor focal date. In other words, it measures the number ofquarters worked during a 1-year period, which beginswith the calendar quarter immediately following (and notincluding) the study date. The range of values is 0 to 4,and the mean number of quarters worked for the entiresample is 1.64. Second, the number of employers rangesfrom 0 for individuals who did not work during the 1-yearfollow-up period to a maximum of 12 employers in aUI-covered job in Maryland. The average number ofemployers for the entire sample was 1.16 employers dur-ing the year following the study date quarter. Both vari-ables, quarters worked and number of employers, wereanalyzed using the ordinal logit technique, commonly

146 RESEARCH ON SOCIAL WORK PRACTICE

TABLE 1: Description of the Four Samples

Percentage ofGroup Definition Type Size (n) Active Caseload

Waiver holders Victims who are administratively marked and hold waivers Universe 261 0.31

Waiver nonholders Victims who are administratively marked and do not hold waivers Universe 293 0.35

Narrative disclosers Victims who are not administratively marked as victims but Sample 184 Represents 5.12whose narratives indicate violence

Nonvictims Welfare recipients with no indication of domestic violence Sample 3,597 Representsentire caseload

used in modeling categorical dependent variables. Theresults of this type of model are similar to the probitmodel in that the raw coefficient is not sufficient in pre-senting a logical interpretation of the effect of the vari-ables. In this case, the log odds results are presented inbrackets to facilitate interpretation. In these models, theeffect is interpreted as the odds of having more units ofthe outcome variable are x% larger or smaller for eachadditional unit of the predictor variable.

Third, the range of the annual earnings variable is from$0 for individuals not employed during the follow-upyear to $73,679.94. The mean or average annual earningsduring the follow-up year is $4,269.47. Earnings is a con-tinuous variable, but it is not logged as is conventionallydone with earnings. Although it is common practice,because of a skewed distribution, to log annual earningsfor the general population, this correction was not helpfulhere. The distribution of earnings for the four samplesmore closely approximates a normal distribution than thedistribution for the general population. As the samplesconsist of welfare recipients, it makes sense that it isunlikely to find individuals earning sufficiently largeamounts as to skew the distribution as drastically as in thecase for the general population. The measure is analyzedusing OLS.

Predictor Variables

Predictor variables fall into three categories: individ-ual characteristics, agency characteristics, and commu-nity indicators.

Individual characteristics. Three state-automatedadministrative data systems were used to gather individ-ual demographic, cash assistance program, and employ-ment data. These were: CIS; the Automated InformationManagement System/Automated Master File; and theMABS. Measures of individual demographic characteris-tics included in the model were age, race, marital status,language, citizenship, disability, pregnancy, age at firstbirth, number of children, age of youngest child,child-only case, and two adult case. Previous studies havealso shown that TANF and employment history areimportant indicators of the likelihood of exiting or return-ing to welfare; thus, individual measures of the months ofTANF receipt out of the previous 60 and previous 12months, the number of quarters employed during lifetimeand in the past year, number of employers during pastyear, and earnings during past year were included in themodels.

Agency characteristics. A series of telephone inter-views was conducted with local department of social ser-vices personnel in each of the 24 jurisdictions, the 23counties and the independent, and incorporated city ofBaltimore to obtain specific information on how the FVOhas been implemented and what effect they perceive it hashad. The interviews were semistructured and had ques-tions in seven main topic areas: (a) background and staff-ing, (b) trainings, (c) screening and disclosures, (d) waiv-ers, (e) referrals and relationships with other agencies, (f)miscellaneous issues (including data entry and confiden-tiality), and (g) opinion questions.

Where possible, questions were coded as quantitativevariables, usually dichotomous measures, and individualsample members were then coded as having visited anagency with the specific characteristic. The variableshypothesized to influence self-sufficiency outcomes,either directly or indirectly, included measures of an indi-vidual’s experience with an agency’s FVO personnel,training, screening, and waiver policy. The first variablemeasures whether the sample member visited an officewith an in-house FVO expert. The second variablereflects the type of organization that conducted thedomestic violence training at the individual’s agency(local organization vs. state personnel or no training).Third, screening methods were also deemed potentiallyimportant. Thus, the next four variables are the type ofstaff who conducted the screening (team led vs. an indi-vidual screener), the screening tools (state-designedquestions vs. locally designed ones), when the client wasinformed of FVO waivers (before vs. after disclosingabuse), and whether written material was distributed.Last, two measures of waiver practice, specifically whogranted the waivers (individuals vs. consultations andteams) and whether waivers were granted automatically,were included.

Community indicators. To control for important socio-economic community characteristics of sample mem-bers’ place of residence, data on Maryland’s 24 jurisdic-tions were compiled from a variety of sources. Data onsociodemographic and economic variables were gatheredfrom the Maryland Departments of Planning; HumanResources; Labor, Licensing, and Regulation; and Healthand Mental Hygiene; the U.S. Bureau of EconomicAnalysis; and the Uniform Crime Report of the U.S.Department of Justice. These data contained informationon housing, crime, vital statistics, and human capital.

Because of the large number of variables and the extentof multicollinearity among them, the creation of moresuccinct and manageable metaindicators was deemed

Hetling, Born / FAMILY VIOLENCE OPTION AND WELFARE 147

necessary (Hetling-Wernyj & Born, 2002). Based on theresults of correlation analyses, two metaindicators(sociodemographic risk and economic risk, composed ofthe sociodemographic and economic variables respec-tively), were constructed. In terms of the sociodemo-graphic metaindicator, infant mortality rate, proportion ofthe population more than age 25 with a bachelor’s degree,high school dropout rate, percentage of female-headedhouseholds, and three crime rates (murder, robbery, andbreaking and entering) were taken into account. For theeconomic metaindicator, both macroeconomic opportu-nity (i.e., the variables of unemployment rate, povertyrate, and job growth rate between 1994 and 1999) andpersonal economic well-being (i.e., average weeklywages, median household income, average per capitaincome, cost of living index, and fair market rent) wereconsidered. The measures were used to assign each juris-diction to one of three overall risk categories (high,medium, or low). If a jurisdiction fell in the worst third ofvalues on the majority of the variables (e.g., top third forunemployment rate, bottom third for median householdincome), it was classified as high risk. Likewise, jurisdic-tions in the best third were assigned to the low risk cate-gory, whereas those with the majority of values aroundthe state mean or median or with mixed values wereassigned to the medium risk category.

Using these two measures, a metarisk indicator wasthen created. Jurisdictions with at least one low ranking,and no high rankings in either the economic or socio-demographic metaindicators were designated low risk.The second category, neutral risk, includes counties withboth medium economic and medium sociodemographicrisk indicators. Finally, jurisdictions with at least onemetaindicator in the high risk ranking and neither in thelow risk ranking were grouped into the high risk category.Individual sample members were then coded as havingresided in a low, medium, or high risk county.

County-level data on community resources related todomestic violence were gathered from a variety ofdomestic violence hotline data and information clearinghouses. These domestic violence community servicevariables were analyzed and used to create a domesticviolence service indicator. A four-tiered categorizationwas created based on whether a domestic violence serviceprovider existed in that jurisdiction, the number of shelterbeds per 1,000 residents in the county or region if no pro-vider existed in that county, and the number of domesticviolence arrests per 1,000 residents. Jurisdictions with anin-county agency, more than 0.1 shelter beds per 1,000residents, and more than 4.0 arrests per 1,000 residentswere classified as having a very high level of domestic

violence services. Jurisdictions meeting two of thesethree criteria were considered to have a high level of ser-vices. Medium level jurisdictions exceeded the cut-offvalue on only one criterion, and low level ones met noneof them. Individual sample members were then coded ashaving resided in a county with a very high, high,medium, or low level of domestic violence services.

RESULTS

Table 2 contains the results from six models address-ing various aspects of self-sufficiency of the samplemembers one year after the study date. Columns 1 to 3present results on three measures of welfare dependency:months of TANF receipt, exits, and recidivism. Threemeasures of employment status—the number of quartersworked, earnings for the poststudy date year, and thenumber of employers an individual had—are presented incolumns 4 to 6.

Cash Assistance Program Participation

Measures of participation in Maryland’s TANF pro-gram comprise the first three models of outcomes in thefollow-up period. Specifically, months of TANF receiptand whether an individual exited welfare were exploredfor all individuals. Recidivism was also investigated forthose individuals who experienced a 60-day exit duringthe follow-up period. These three measures together pro-vide a thorough picture of welfare dependence during theyear following disclosure.

Months of cash assistance receipt. Column 1 in Table 2contains results from an OLS model exploring the factorsassociated with the number of months of TANF receiptduring the follow-up period. Controlling for other indi-vidual and macrolevel factors, waiver holders received2.11 fewer months of TANF than nonvictims, but this dif-ference is not statistically significant. In contrast, waivernonholders and narrative disclosers received more thaneight fewer months of TANF than nonvictims, and thesedifferences are statistically significant. Although waiverholders are the only group of domestic violence victimsnot to receive statistically less cash assistance in thefollow-up period, they also do not receive more monthsthan nonvictims. Thus, one can conclude that, even aftercontrolling for a number of background characteristics,the receipt of a waiver did not increase the probability thata welfare recipient remained on welfare longer than anonvictim.

148 RESEARCH ON SOCIAL WORK PRACTICE

149

TAB

LE

2:

Ind

ivid

ual

Sel

f-su

ffic

ien

cy O

utc

om

es

Cas

h P

artic

ipat

ion

Mod

els

Em

ploy

men

t Mod

els

Mon

ths

of T

AN

FQ

uart

ers

Num

ber

ofA

nnua

lR

ecei

pt (

1)E

xit (

2)R

ecid

ivis

t (3)

Wor

ked

(4)

Em

ploy

ers

(5)

Ear

ning

s (6

)

Mar

gina

lM

argi

nal

Var

iabl

eC

oeffi

cien

tS

EC

oeffi

cien

tS

EE

ffect

Coe

ffici

ent

SE

Effe

ctC

oeffi

cien

tS

E%

Cha

nge

Coe

ffici

ent

SE

% C

hang

eC

oeffi

cien

tS

E

Wai

ver

hold

ers

–0.2

11**

*0.

286

–0.0

120.

091

–0.0

050.

284

0.14

70.

051

–0.2

340.

135

–20.

9–0

.074

0.15

0–7

.1–9

1.06

382.

30W

aive

r no

nhol

ders

–0.8

23**

*0.

281

0.02

70.

091

0.01

10.

380*

*0.

142

0.07

1–0

.035

0.13

2–3

.40.

061

0.14

66.

359

9.16

415.

45N

arra

tive

disc

lose

rs–0

.895

***

0.33

0–0

.009

0.10

4–0

.003

–0.0

400.

181

–0.0

06–0

.051

0.14

1–5

.00.

230

0.15

825

.9–7

94.1

9*40

6.64

Dem

ogra

phic

con

trol

sYe

sYe

sYe

sYe

sYe

sYe

sTA

NF

and

em

ploy

men

tYe

sYe

sYe

sYe

sYe

sYe

shi

stor

y co

ntro

lsA

genc

y F

VO

con

trol

sYe

sYe

sYe

sYe

sYe

sYe

sJu

risdi

ctio

nal c

ontr

ols

Yes

Yes

Yes

Yes

Yes

Yes

Bal

timor

e C

ity c

ontr

olYe

sYe

sYe

sYe

sYe

sYe

sLo

g lik

elih

ood

–276

1.6

–722

.4–5

743.

2–5

690.

1P

seud

oR

20.

074

0.08

00.

114

0.10

5R

20.

123

0.46

8S

ampl

e si

ze4,

335

4,33

52,

385

4,33

54,

335

4,33

5

NO

TE

:TA

NF

=Te

mpo

rary

Ass

ista

nce

forN

eedy

Fam

ilies

;FV

O=

Fam

ilyV

iole

nce

Opt

ion.

The

mod

els

inco

lum

ns1

and

6ar

eor

dina

ryle

asts

quar

esre

gres

sion

mod

els;

thos

ein

colu

mns

2an

d3

are

prob

itm

odel

s.T

hem

odel

sin

colu

mns

4an

d5

are

ordi

nall

ogit

mod

els.

All

stan

dard

erro

rsha

vebe

enco

rrec

ted

forh

eter

oske

dast

icity

.Dep

ende

ntva

riabl

esar

eou

tcom

em

easu

res

fort

he1-

year

fol-

low

-up

perio

d.C

ontr

ols

incl

ude

varia

bles

mea

surin

gde

mog

raph

icch

arac

teris

tics

(age

,rac

e,m

arita

lsta

tus,

lang

uage

,citi

zens

hip,

disa

bilit

y,pr

egna

ncy,

age

atfir

stbi

rth,

num

bero

fchi

ldre

n,ag

eof

youn

gest

child

,chi

ld-o

nly

case

,and

two

adul

tcas

e),T

AN

Fan

dem

ploy

men

this

tory

(mon

ths

ofTA

NF

rece

ipto

utof

prev

ious

60an

dpr

evio

us12

mon

ths,

num

bero

fqua

rter

sem

ploy

eddu

ring

lifet

ime

and

inth

epa

stye

ar,n

umbe

rofe

mpl

oyer

sdu

ring

past

year

,and

earn

ings

durin

gpa

stye

ar),

agen

cyF

VO

impl

emen

tatio

n(lo

cald

omes

ticvi

olen

cepr

ovid

ertr

aini

ng,t

eam

-led

scre

enin

gs,M

aryl

and

Dep

artm

ento

fHum

anR

esou

rces

–des

igne

dqu

estio

ns,w

aive

rsdi

scus

sed

befo

redi

sclo

sure

,gen

eral

DV

info

dist

ribut

ed,w

aive

rsgr

ante

dw

ithou

tcon

sulta

tion,

and

wai

vers

gran

ted

auto

mat

ical

ly),

juris

dict

iona

lcha

ract

eris

tics

(ave

rage

mon

thly

case

load

per

1000

resi

dent

s,po

pula

tion

size

inth

eth

ousa

nds,

risk

met

aind

icat

or,a

nddo

mes

ticvi

olen

cese

rvic

ein

dica

tor)

,and

aco

ntro

lfor

resi

-de

nce

in B

altim

ore.

* p>

.05.

**p

> .0

1.**

*p>

.001

.

Exits. A probit model was constructed to estimate theeffect of domestic violence and waivers on TANF exitsand to hold other individual and macro level variablesconstant. These results are presented in column 2 of Table2. The effect of receiving a waiver, not receiving a waiverbut being administratively marked, and disclosingdomestic violence but not receiving a waiver or beingadministratively marked was very small and statisticallynot significant. Waiver holders were 0.5 percentagepoints less likely to experience a 60-day exit during thefollow-up period. Waiver nonholders were 9.1 percent-age points more likely to experience an exit. And, narra-tive disclosers were 0.3 percentage points less likely toexit than nonvictims. Again, none of these differenceswere statistically significant.

Recidivism. The last model of Table 2, presented incolumn 3, is a probit model that explored the effect ofdomestic violence on recidivism among exiters. AmongTANF exiters, neither waiver recipients nor narrativedisclosers had a statistically significant increase ordecrease in their likelihood to recidivate. Waivernonholders, on the other hand, had a 7.1 percentage pointincrease in their likelihood to recidivate as compared tononvictims.

Employment Experiences

The set of outcomes is presented in columns 4 to 6 ofTable 2. These multivariate models examine the effect ofdomestic violence classification on employment whilecontrolling for the individual, agency, and jurisdictionalvariables included in the models described previously.

Quarters employed. The most obvious measure ofemployment, and most fundamental employment factorto influence self-sufficiency, is employment status, or inother words, whether an individual was employed. Col-umn 4 of Table 2 contains the results of an ordinal logitmodel with quarters worked during the 1-year, orfour-quarter, follow-up period as the dependent variable.Examining these results, all domestic violence victimsworked fewer follow-up quarters than nonvictims did, butnone of these differences were statistically significant.Specifically, the odds of having worked more quarterswere 20.9% smaller for waiver holders than nonvictims,but only 3.4% smaller for waiver nonholders and 5.0%smaller for narrative disclosers than nonvictims.

Number of employers. The number of employers dur-ing the follow-up period is the dependent variable of the

ordinal logit model presented in column 5. The modelincluded both individuals who worked during the follow-up period and those who did not. Thus, the dependentvariable included 0 for those who did not have a job andgoes up to 12 employers on the high end of the range. Inthis model, waiver holders were more likely to have aslightly smaller number of employers as compared tononvictims, but this was not statistically significant andwas probably accounted for by this group’s smaller likeli-hood of working at all. Both waiver nonholders and narra-tive disclosers had a slightly greater likelihood of havinga larger number of employers. None of these differences,however, were statistically significant.

Average quarterly earnings. Results from an OLSregression modeling annual earnings are presented in col-umn 6 of Table 2. In this model, narrative disclosers, onaverage, made $794.19 less than nonvictims did, and thisdifference, as opposed to the results of the other twoemployment models, was statistically significant. In lightof the nonsignificant findings in the quarters workedmodel and the number of employers model, this result isespecially interesting. Although narrative disclosers werenot significantly more likely to work less or to have feweremployers, they did make significantly less moneyaccording to these data. Waiver holders made slightly less($91.06) than nonvictims, and waiver nonholders mademore ($599.16) than nonvictims. However, neither ofthese two differences was statistically significant.

DISCUSSION AND APPLICATIONSFOR SOCIAL WORK PRACTICE

The multivariate models show that, after controllingfor important individual and macrolevel variables, differ-ences between domestic violence victims and nonvictimswere rarely statistically significant. The lack of differencebetween the groups may be attributed to two factors.First, the possibility that victims were included in thenonvictim group is strong and discussed further as a studylimitation. Second, domestic violence is not the only bar-rier faced by welfare recipients; substance abuse, learningdisabilities, and mental health issues, to name just a few,may also affect a woman’s ability to become self-sufficient.

Given these possible explanations for the lack of statis-tical significance on many measures, the differences thatare statistically significant become more interesting anddeserve consideration. In terms of employment out-comes, only the annual earnings model showed a

150 RESEARCH ON SOCIAL WORK PRACTICE

difference between nonvictims and any of the three vic-tim groups. In this model, narrative disclosers earned sig-nificantly less in the 1-year follow-up period. Turning tocash assistance participation outcomes, although all vic-tims received fewer months of assistance during the fol-low-up period, this difference was statistically significantonly for waiver nonholders and narrative disclosers. Con-sidering intergroup differences, it seems as if the out-comes of waiver holders do not differ at all fromnonvictims. Similarly, waiver nonholders differ very littlewith slightly fewer months of TANF receipt in the fol-low-up period but a higher likelihood of returning to wel-fare after exiting. However, the outcomes of narrativedisclosers seem worse than those of nonvictims in thatthey are both statistically more likely to earn less and toreceive fewer months of TANF in the follow-up year.

Admittedly, these analyses would have benefited fromthe inclusion of perspectives from the victims themselvesand from frontline caseworkers. However, based on thebreadth of administrative data available and the time,effort, and skill necessary to analyze these data well, theprocess of also conducting extensive personal interviewsor focus groups with either victims or frontline welfareworkers was deemed unfeasible. The decision to not col-lect data from the victims and caseworkers limits the abil-ity of these analyses to consider fully the complexity ofindividual trajectories, decision making, and discretion.It also contributed to the possibility of our inclusion ofdomestic violence victims in the nonvictim group.Because it is likely that a number of women chose not todisclose abuse to their caseworker, these women wouldhave been mistakenly, but unavoidably, categorized asnonvictims. Thus, our analyses are limited to understand-ing the outcomes of known, and not all, domestic vio-lence victims. A second data limitation relates to theemployment data used in the study. Because employmentdata come from the MABS and do not include under-the-table employment, self-employed individuals, andemployment in other states or the federal government, thefigures presented likely underestimate slightly the truelabor force participation of the sample.

A final study limitation concerns difficulties withselection bias. Selection bias occurs when it is not the pol-icy but rather the characteristics of the individuals whochoose or are assigned to the treatment that are affectingoutcomes. Unfortunately, experiments with matchedpairs and implementation of instrumental variables esti-mation were determined to be unfeasible alternatives totest the robustness of the findings. Although the originalresearch plan included such models, the available infor-mation from individual and macrolevel data made the use

of these methods impractical. Specifically, the difficultyin locating suitable instruments for domestic violenceand waiver use prohibited the development of this type ofmodel. Likewise, significantly different demographiccharacteristics made matched pairs extremely difficult.Although the absence of these alternative models is a lim-itation of the study, it does not compromise the impor-tance of the analyses as a whole. The purpose of theresearch was to explain the particular experiences of thefour samples and assess the effect of the FVO thus far, notto predict the outcomes of other and future recipients. Inthis respect, the data and analyses used are appropriateand sufficient and also leave open the door to otherresearchers interested in these questions.

Even considering the study limitations, the findingspresented here certainly do not indicate that abuse of FVOwaivers has been occurring or that the FVO has encour-aged women to stay on welfare longer and not address theissue of self-sufficiency. However, whereas waiver usedoes not seem to have negative consequences in terms ofself-sufficiency outcomes, the outcomes of victims whoare administratively marked but do not use waivers arevery similar. Although low disclosure and documentationrates may be troubling, a low waiver use rate may be moreindicative of the number of victims who need or acceptwaivers than of an unwillingness to grant them. In somecircumstances, waivers may not have been the appropri-ate service for some victims. As these women are markedin the administrative system, it is likely that they were toldof the FVO and received domestic violence services otherthan an FVO waiver.

Of more concern than the low waiver usage rate are thepoor outcomes of victims who are not administrativelymarked in terms of lower earnings and fewer months ofTANF receipt. This combination of fewer earnings andless assistance points to disturbing prospects for thosedomestic violence victims whose experiences are notadministratively documented. Perhaps when caseworkersdo not administratively mark victims of abuse, the work-ers are also less likely to take the abuse seriously and offerhelp in these circumstances. Conversely, research hasshown that the process of leaving abusive situations isoften iterative (Choice & Lamke, 1997). Perhaps thesewomen are not yet ready to fully accept help in dealingwith the violence and choose to return to their abusersinstead. It is unlikely that only narrative disclosers andnot other domestic violence victims were more likely tobe in this situation.

Because the outcomes of administratively marked vic-tims seem positive, it is strongly recommended that theFVO should remain as part of the welfare legislation. The

Hetling, Born / FAMILY VIOLENCE OPTION AND WELFARE 151

FVO has potential as an important tool for identifying andserving domestic violence victims on welfare. The betteroutcomes of administratively marked victims (bothwaiver holders and nonholders) in comparison to narra-tive disclosers seem to point to a positive effect of identifi-cation and documentation on service provision andassistance.

The outcomes of victims who are not administrativelymarked, however, indicate that some individuals may beslipping through the cracks of a well-intentioned policy.This finding suggests that caseworkers must be more dili-gent in recognizing and recording domestic violence andthat their ability to do so can increase the likelihood ofpositive outcomes for clients. Although social workersusually do not function as frontline welfare caseworkersand do not have the responsibility of documentingdomestic violence among welfare recipients, their rolecan aid in identifying victims of domestic violence andensuring that their information is marked in the narrativesas well as the automated system. First, as integral parts ofwelfare agencies, social workers often assist in servingvictims and training frontline workers. In serving theirclients, social workers should work collaboratively withthe welfare caseworkers and with their clients’ permis-sion advocate for FVO services on their behalf and con-firm that the welfare caseworker is properly documentingthe abuse. On a more macrolevel, social workers oftenserve as the resident experts on domestic violence. Thisrole lends itself to training frontline staff in the complexi-ties of domestic violence and to assisting with the devel-opment of screening protocols. Trainings should includean explanation of the benefit of services and the impor-tance of documenting the situation. If trainings have notbeen conducted recently or at all, agency social workersmay suggest their implementation.

Finally, the need for further evaluations of the FVOand research on the outcomes of domestic violence vic-tims is great. It is recommended that future researchinclude a more in-depth use of survey data from both wel-fare recipients and frontline workers. Qualitative researchwould serve to disentangle the questions of caseworkersensitivity and awareness of the issue versus individualwillingness to disclose and be able to address more fullythe distinct outcomes of narrative and administrativelymarked disclosers. Furthermore, continued monitoringand evaluation of screening, documentation, and serviceuse is essential because the FVO and the implementationof the policy are, relatively speaking, still very new. Poli-cies and practices have not yet been set in stone at thejurisdictional level. In this situation, especially, an under-standing of what is happening on the frontlines is critical

in informing evaluations of the results as well as thedevelopment of improved training, screening, and per-haps service delivery methods and partnerships. Last, theanalyses presented here do not exhaust the ways in whichone could use existing administrative or other quantita-tive data. Considering the evolving nature of servicedelivery and the limited understanding of the long-termoutcomes of clients, these data remain an importantsource.

REFERENCES

Allard, M. A., Albelda, R., Colten, M. E., & Cosenza C. (1997). Inharm’s way? Domestic violence, AFDC receipt, and welfarereform in Massachusetts. Boston: University of MassachusettsPress.

Angelari, M. (1998). The family violence option in Maryland: A pre-liminary report. Baltimore: The Women’s Law Center ofMaryland.

Brandwein, R. A. (2000). Toward real welfare reform: The voices ofbattered women. AFFILIA: Journal of Women and Social Work, 15,224-243.

Browne, A., & Bassuk, S. S. (1997). Intimate violence in the lives ofhomeless and poor housed women: Prevalence and patterns in anethnically diverse sample. American Journal of Orthopsychiatry,67, 261-278.

Brush, L. (2000). Battering, traumatic stress, and welfare-to-worktransition. Violence Against Women, 6, 1039-1065.

Burt, M. R., Zweig, J. M., & Schlichter, K. (2000). Strategies foraddressing the needs of domestic violence victims within the TANFprogram: The experience of seven counties. Washington, DC: TheUrban Institute.

Choice, P., & Lamke, L. K. (1997). A conceptual approach to under-standing abused women’s stay/leave decisions. Journal of FamilyIssues, 18, 290-314.

Curicio, W. (1997). The Passaic County study of AFDC recipients in awelfare-to-work program. Passaic County, NJ: Passaic CountyBoard of Social Services.

Davis, M. F. (1999). The economics of abuse: How violence perpetu-ates women’s poverty. In R. Brandwein (Ed.), Battered women,children, and welfare reform: The ties that bind (pp. 17-30). Thou-sand Oaks, CA: Sage.

Hetling-Wernyj, A., & Born, C. E. (2002). Domestic violence and wel-fare receipt in Maryland: How is the family violence option beingimplemented? Baltimore: University of Maryland Press.

Kantor, G. K., & Jasinski, J. L. (1998). Dynamics and risk factors inpartner violence. In J. L. Jasinski & L. M. Williams (Eds.), Partnerviolence: A comprehensive review of 20 years of research (pp.1-43). Thousand Oaks, CA: Sage.

Lennett, J. (1997). Like ships that pass in the night: AFDC policy andbattered women. Law and Policy, 19, 183-203.

Murphy, P. A. (1997). Recovering from the effects of domestic vio-lence: Implications for welfare reform policy. Law and Policy, 19,169-182.

Pagelow, M. D. (1981). Women-battering: Victims and their experi-ences. Beverly Hills, CA: Sage.

152 RESEARCH ON SOCIAL WORK PRACTICE

Personal Responsibility and Work Opportunity Reconciliation Act,(1996).

Raphael, J. (1996). Prisoners of abuse: Domestic violence and welfarereceipt. Chicago: The Taylor Institute.

Raphael, J. (2000). Saving Bernice: Battered women, welfare, andpoverty. Boston, MA: Northeastern University Press.

Raphael, J., & Tolman, R. (1997). Trapped by poverty/trapped byabuse: New evidence documenting the relationship betweendomestic violence and welfare. Ann Arbor, MI: The Project forResearch on Welfare, Work, and Domestic Violence.

Raphael, J., & Haennicke, S. (1999). Keeping battered women safethrough the welfare-to-work journey: How are we doing? Chicago:The Taylor Institute.

Satel, S. L. (1998). The abuse excuse. Women’s Quarterly. RetrievedDecember 9, 2003 from http://www.iwf.org/pubs/twg/wi98f.shtml(available now at http://www.fact.on.dc/newspaper/wq98wi04.htm).

Shepard, M., & Pence, E. (1988). The effect of battering on theemployment status of women. AFFILIA: Journal of Women andSocial Work, 3, 55-61.

Straus, M. A. (1986). Medical care costs of intra-family assault andhomicide. Bulletin of the New York Academy of Medicine, 62,556-561.

Tjaden, P., & Thoennes, N. (1998). Prevalence, incidence, and conse-quences of violence against women: Findings from the NationalViolence Against Women Survey. Washington, DC: National Insti-tute of Justice.

Tolman, R., & Raphael, J. (2000). A review of research on welfare anddomestic violence. Journal of Social Issues, 56, 655-682.

Tolman, R., & Rosen, D. (2001). Domestic violence in the lives ofwomen receiving welfare. Violence Against Women, 7, 141-158.

Walker, L. E. (1979). The battered woman. New York: Harper & Row.

Hetling, Born / FAMILY VIOLENCE OPTION AND WELFARE 153