women’s employment status, coercive control, and intimate partner violence in mexico

17
ANDRE ´ S VILLARREAL University of Texas at Austin Women’s Employment Status, Coercive Control, and Intimate Partner Violence in Mexico Findings from previous studies examining the relation between women’s employment and the risk of intimate partner violence have been mixed. Some studies find greater violence toward women who are employed, whereas others find the opposite relation or no relation at all. I propose a new framework in which a woman’s employment status and her risk of violent victimization are both influenced by the level of control exercised by her partner. Con- trolling men will actively prevent women from working and are also more likely to physically harm their partners. Using a statistical model in which the effect of omitted characteristics on women’s employment and their risk of violence are allowed to be correlated reverses the esti- mated association between employment and violence. The final results show that employ- ment reduces women’s risk of violence. Data for the study are drawn from a sample of over 30,000 Mexican women in intimate relation- ships. The findings have broad implications given the increase in female labor force partici- pation in many parts of the world. Research on domestic violence has increased dra- matically over the past two decades. Studies con- ducted in the United States have revealed new information about the frequency and severity of violence perpetrated against women by their part- ners, as well as the types of factors that place women at greater risk (see Holtzworth-Monroe, Bates, Smutzler, & Sandin, 1997; Johnson & Ferraro, 2000; Miller & Knudsen, 1999, for re- views). One unresolved issue in the literature on domestic violence is the effect that women’s eco- nomic status, and specifically their participation in the labor market, has on their risk of violent victimization. Different theoretical perspectives lead to opposite conclusions regarding the possi- ble effect of women’s employment. Marital de- pendency theorists argue that employed women are less economically dependent on their partners and are therefore less likely to tolerate abuse (Kalmuss & Straus, 1990; Straus, Gelles, & Steinmetz, 1980; Strube & Barbour, 1983). A second theoretical tradition known as resource theory sees violence as a result of power derived from an imbalance in access to resources (Allen & Straus, 1980; Goode, 1971). Male partners who cannot derive power from their employment status or greater economic resources will use vio- lence to assert their dominance in the relation- ship. A feminist perspective similarly predicts a higher risk of violence against employed women. Feminist scholars place greater impor- tance on the symbolic value attached to men’s greater earnings potential. They argue that men’s larger economic contribution to the household is tied to the construction of their masculinity. Men in intimate relations will therefore perceive their female partners’ employment as a threat to which they might respond violently (Dobash & Dobash, 1979; Yllo, 1993; Yllo & Straus, 1990). Empirical findings on the relation between women’s employment status and the risk of Population Research Center, University of Texas, 1 Univer- sity Station, G1800, Austin, TX 78712 ([email protected]. edu). Key Words: domestic violence, employment, intimate part- ner/marital abuse, Mexico, power. 418 Journal of Marriage and Family 69 (May 2007): 418–434

Upload: andres-villarreal

Post on 21-Jul-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

ANDRES VILLARREAL University of Texas at Austin

Women’s Employment Status, Coercive Control,

and Intimate Partner Violence in Mexico

Findings from previous studies examining therelation between women’s employment and therisk of intimate partner violence have beenmixed. Some studies find greater violencetoward women who are employed, whereasothers find the opposite relation or no relationat all. I propose a new framework in whicha woman’s employment status and her risk ofviolent victimization are both influenced by thelevel of control exercised by her partner. Con-trolling men will actively prevent women fromworking and are also more likely to physicallyharm their partners. Using a statistical modelin which the effect of omitted characteristics onwomen’s employment and their risk of violenceare allowed to be correlated reverses the esti-mated association between employment andviolence. The final results show that employ-ment reduces women’s risk of violence. Datafor the study are drawn from a sample of over30,000 Mexican women in intimate relation-ships. The findings have broad implicationsgiven the increase in female labor force partici-pation in many parts of the world.

Research on domestic violence has increased dra-matically over the past two decades. Studies con-ducted in the United States have revealed newinformation about the frequency and severity of

violence perpetrated against women by their part-ners, as well as the types of factors that placewomen at greater risk (see Holtzworth-Monroe,Bates, Smutzler, & Sandin, 1997; Johnson &Ferraro, 2000; Miller & Knudsen, 1999, for re-views). One unresolved issue in the literature ondomestic violence is the effect that women’s eco-nomic status, and specifically their participationin the labor market, has on their risk of violentvictimization. Different theoretical perspectiveslead to opposite conclusions regarding the possi-ble effect of women’s employment. Marital de-pendency theorists argue that employed womenare less economically dependent on their partnersand are therefore less likely to tolerate abuse(Kalmuss & Straus, 1990; Straus, Gelles, &Steinmetz, 1980; Strube & Barbour, 1983). Asecond theoretical tradition known as resourcetheory sees violence as a result of power derivedfrom an imbalance in access to resources (Allen &Straus, 1980; Goode, 1971). Male partners whocannot derive power from their employmentstatus or greater economic resources will use vio-lence to assert their dominance in the relation-ship. A feminist perspective similarly predictsa higher risk of violence against employedwomen. Feminist scholars place greater impor-tance on the symbolic value attached to men’sgreater earnings potential. They argue that men’slarger economic contribution to the household istied to the construction of their masculinity. Menin intimate relations will therefore perceive theirfemale partners’ employment as a threat to whichthey might respond violently (Dobash & Dobash,1979; Yllo, 1993; Yllo & Straus, 1990).

Empirical findings on the relation betweenwomen’s employment status and the risk of

Population Research Center, University of Texas, 1 Univer-sity Station, G1800, Austin, TX 78712 ([email protected]).

Key Words: domestic violence, employment, intimate part-ner/marital abuse, Mexico, power.

418 Journal of Marriage and Family 69 (May 2007): 418–434

violence are similarly divided. Some studies findgreater violence toward women who are em-ployed (DeMaris, Benson, Fox, Hill, & Van Wyk,2003; Hornung, McCullough, & Sugimoto,1981), whereas others find the opposite relation(Kalmuss & Straus, 1990; Straus et al., 1980),or no relation at all (Fox, Benson, DeMaris, &Van Wyk, 2002; Kaukinen, 2004). In this article,I suggest that the different estimates of the effectof women’s employment may be a result of theway in which statistical models used to test theassociation between women’s employment andtheir risk of violence have been specified. AsGibson-Davis, Magnuson, Gennetian, andDuncan (2005) have recently suggested, unmea-sured characteristics of women are likely to affectboth their likelihood of employment and their riskof violence leading to biased estimates whenwomen’s employment is used as a predictor ofviolence in standard regression analysis. I pro-pose a conceptual framework in which femaleemployment and violence are both determinedby the level of control exercised by the malepartner. Controlling partners will actively pre-vent women from working and are also morelikely to physically harm their partners. I test thismodel using data from a survey of Mexicanwomen in intimate relations.

Intimate partner violence constitutes an impor-tant social problem in Mexico. Dıaz-Olavarrieta,Ellertson, Paz, Ponce de Leon, and Alarcon-Segovia (2002) note that almost half of all femalehomicides in Mexico City are a result of partnerabuse. Estimates of the incidence of violenceagainst women from surveys range from 9%(Dıaz-Olavarrieta et al., 2002) to 45% (Rivera-Rivera, Salmeron-Castro, & Castro, 2004) de-pending on the population sampled, the exactquestions asked, and the baseline period (e.g.,previous 12 months, length of the relationship).In addition to the comparatively high rate of inti-mate partner violence, Mexico constitutes animportant test case for theories that posit a relationbetween women’s employment status and theirrisk of violence because of the rapid increase inwomen’s labor force participation, and the pres-ence of a strong gender ideology defining men asthe primary breadwinners (Garcıa & de Oliveira,1994). Between 1990 and 2004, the female laborforce participation rate in Mexican urban areasincreased from 34% to 41% (Instituto Nacionalde Estadıstica, Geografıa e Informatica [INEGI],1990, 2004a). These changes are in turn tied tobroader trends in women’s educational attainment

and family formation among other factors. Dur-ing the decade of the 1990s, the percentage ofadult women with lower secondary education ormore increased from 35% to 45% (INEGI, 1993,2001). Yet, despite women’s advances in educa-tion and employment, they nevertheless facea gender ideology in which the use of violenceby men to control the behavior of female partnersis still widely accepted (Garcıa & de Oliveira;Gonzalez de la Rocha, 1994). If resource theoristsare correct and violence toward women is a com-pensatory mechanism for men’s loss of domi-nance in other domains such as the economicdomain, then we should expect the effect ofwomen’s higher status to be greater in a countrysuch as Mexico where women’s advances aremore recent and where traditional norms regard-ing women’s roles are likely to lag behind.

Female Labor Force Participation andDomestic Violence

A large body of research has examined the rela-tion between household socioeconomic condi-tions and intimate partner violence. Moststudies find a higher incidence of physical abuseamong economically disadvantaged groups. Oneexplanation for this finding is that economicdeprivation creates stress and frustration amongcouples, which may ultimately lead to conflictand violence (Farrington, 1986; Straus, 1990).Because a lack of employment of either partnercan cause economic hardship, stress and frustra-tion theories predict a lower risk of domestic vio-lence when either or both partners are employed.Other theoretical traditions explicitly distinguishthe effect that a woman’s employment may haveon physical abuse within heterosexual couples.Resource theorists argue that violence is causedby the disparity in economic resources betweenpartners rather than their sum total. Being the pri-mary breadwinner within the household is oneway men derive power within a relationship.Men who lack employment will be more likelyto use violence as a last attempt to maintain dom-inance. Consistent with this approach, some stud-ies find that a greater difference between men andwomen’s status results in a higher incidence ofpartner abuse (Allen & Straus, 1980; Hornunget al, 1981; Kaukinen, 2004). Nevertheless, asAnderson (1997) points out, if intimate partnerviolence is simply attributed to a discrepancy inaccess to resources, then a disparity in employ-ment status and income favoring a man would

Women’s Employment and Violence in Mexico 419

be as likely to result in his female partner’s use ofviolence against him. Yet, this is not the case. Inher study of domestic violence in U.S. families,Anderson examines the use of violence by bothmale and female partners. She finds that lowerrelative economic status is associated with men’s,but not women’s, perpetration of violence. An-derson argues that this gender difference in theuse of violence to compensate for a lower eco-nomic position is consistent with a feminist per-spective according to which a higher earningcapacity has a symbolic value beyond its purelymaterial rewards. For men, a higher economicstatus is tied to their construction of their maleidentities. In their study of the effect of labor forceparticipation on the risk of violence among Cana-dian women, Macmillan and Gartner (1999) alsotreat employment as a symbolic resource. Theyfind that the effect of women’s labor force partic-ipation is conditioned on their partners’ employ-ment status: When her partner is employed,a woman’s labor force participation lowers herrisk of systematic abuse. But when her partneris unemployed, a woman’s labor force participa-tion significantly increases her risk of systematicabuse. This conditional effect strongly suggeststhat men see employment in terms other than sim-ply a way to attain economic resources. As Mac-millan and Gartner (1999) argue, unemployedmen will use coercive tactics including violenceagainst their employed partners in order to ‘‘rein-state their dominance’’ (p. 956).

A different theoretical tradition suggests thatwomen’s employment reduces rather than in-creases the incidence of intimate partner vio-lence. Marital dependency theorists argue thatwomen who lack sufficient economic resourcesof their own, and in particular those who areunemployed, are at greater risk of violence. Insome versions of this argument, the causal con-nection proposed between women’s unemploy-ment and violence seems to rely on a selectioneffect (women who are employed are more likelyto exit abusive relationships), whereas in otherversions of marital dependency theory, menwho know that their female partners are lessbeholden to them and can leave choose not tophysically harm them. In either case, the relationbetween female employment and the risk of vio-lence is hypothesized to be negative. Strube andBarbour (1983) specifically examine women’sdecisions to stay or leave abusive relationships.They find that women’s economic dependence,measured both objectively as not being employed

outside the home, and subjectively in terms ofwomen’s own assessments of their economicposition, is significantly related with the decisionto remain with their abusive partners. Similarly,Kalmuss and Straus (1990) find that womenwho are more dependent on their husbands aremore likely to experience physical abuse.

Coercive Control and IntimatePartner Violence

Recent work on domestic violence has recog-nized the need to examine the relation betweenphysical abuse and the level of control exercisedby men over their female partners (Johnson &Ferraro, 2000). This relation is implicit inresource theory insofar as the use of violence isconceptualized as a compensatory strategy forthe loss of power within the relationship. Afterall, one way to define power is simply the abilityto control a partner’s behavior (Blood & Wolfe,1960; Kaukinen, 2004; Macmillan & Gartner,1999; Trichenor, 1999). Various studies haveelaborated on the relation between control andviolence. In an early formulation, Finkelhor(1983) proposes that power asymmetries underlievarious forms of family violence. He suggeststhat physical abuse is a way that the more power-ful gain control of a situation. Similarly, Stets(1988) uses detailed information from interviewsof women and their abusive partners to developan interactionist model of domestic violence inwhich the desire to control women’s behavior isthe central motivating factor for physical abuse(see also Stets & Pirog-Good, 1987). Controllingbehavior in Stets’s (1988) framework includeslimiting women’s contact with friends (especiallymale friends) and monitoring their spending hab-its. When male partners perceived a challenge totheir authority in these domains, the outcome wasoften violent.

Several survey-based studies have attemptedto measure the extent of control by a male partnerin a relationship and its effect on the risk of vio-lence. In their study of the effect of female laborforce participation on abuse among intimate part-ners, Macmillan and Gartner (1999) explicitlymodel physical abuse as a consequence of coer-cive control by the male partner. They definecoercive control as an attempt by a man to limita woman’s contact with other men or with familyand friends, an insistence on knowing where sheis and who she is with, and the prevention of heraccess to or knowledge of family income. They

420 Journal of Marriage and Family

find coercive control to be an important predictorof abuse. Coleman and Straus (1990) use datafrom the National Family Violence Survey tocompute an index of marital power that is basedon earlier work by Blood and Wolfe (1960). Theyuse self-reported assessments about who hasthe final say in decisions regarding six issues:‘‘buying a car, having children, what house orapartment to take, what job either partner shouldtake, whether a partner should go to work orquit work, and how much money to spend eachweek on food’’ (p. 289). They find that male-dominant relationships in which the man morefrequently has the final say have the highestamount of conflict (although the strength of theassociation is reduced if both partners agree onthe ‘‘appropriateness’’ of the unequal arrange-ment). Moreover, they find that ‘‘the powerstructure of the relationship is strongly andindependently related to violence’’ (p. 300,emphasis in original). The conceptual frame-work proposed in this article is consistent withthat of Coleman and Straus, and Macmillan andGartner in that coercive control is seen as a pre-dictor of violence. By contrast, Johnson (1995)proposes a framework in which coercive controlis part of a larger pattern of violence and controlused by men against their partners. This distinc-tion is also discussed by Kaukinen (2004,p. 457). The extent to which control and vio-lence may be considered part of the same con-struct is examined empirically below.

Cross-cultural work on domestic violence hassimilarly found the power exercised by men overtheir partners as important predictors of abuse.Using information from 90 societies included inthe Human Relations Area Files, Levinson (1989)finds domestic authority, defined as men’s abil-ity to make decisions regarding the spending offamily resources, child care and kinship rela-tions as ‘‘the single most powerful predictorof wife-beating frequency’’ (p. 73). The fact thatthe relation between interpersonal control at thehousehold level and violence holds for a sampleof largely preindustrial societies suggests thatthe causal mechanism may be generalized beyondthe contemporary Western context where avast majority of the research on the topic hasbeen carried out. Similarly, a recent study usingnationally representative samples of womenfrom nine countries available as part of theDemographic and Health Surveys finds thathusbands’ controlling behavior was associatedwith a higher risk of violence against married

women in all countries included in the study(Kishor & Johnson, 2004, pp. 53 – 71). Control-ling behavior in this instance was defined as‘‘whether the respondent’s husband is jealous orangry if she talks to other men; he frequentlyaccuses her of being unfaithful; he does not per-mit her to meet her girlfriends; he limits hercontacts with her family; he insists on knowingwhere she is all the time; and he does not trusther with money’’ (Kishor & Johnson, 2004,p. 68).

Research on domestic violence in Mexico isstill relatively scarce; yet, the studies that existalso suggest that violence is closely associatedwith the level of control that men have withinthe household. In one of the few works on thesubject, Oropesa (1997) uses survey data toexamine marital power in Mexico. He definespower using respondents’ answers to a generalquestion of who has the last word on householddecisions, as well as the use of violence by themale partner. Consistent with a modernizationperspective according to which economic devel-opment is associated with cultural and structuralchanges that enhance women’s status, Oropesafinds that wives’ education both lowers the riskof violence and promotes more egalitarian deci-sion making. He further demonstrates that deci-sion making and violence against femalepartners are strongly correlated. In her ethno-graphic study of working-class families in theMexican city of Guadalajara, Gonzalez de laRocha (1994) similarly argues that violence isused by Mexican men to maintain their domi-nant position and to ‘‘reaffirm or confirm anasymmetric family discipline’’ (p. 143). Finally,in a study of domestic violence among MexicanAmericans, Sugihara and Warner (2002) alsofind that dominance in a relationship is associ-ated with violent behavior. Sugihara and Warneridentify four different dimensions of dominance(power, possessiveness, decision making, anddevaluation). Interestingly, although control anddecision making are associated with male vio-lence, power and possessiveness appear to beconducive to physical assault by both men andwomen.

An important limitation of this large body ofresearch that demonstrates a close relationbetween the coercive control exercised by menand their use of violence against female partnersis that it fails to acknowledge how the level ofcontrol may also influence other factors that inturn affect the risk of violence. In particular,

Women’s Employment and Violence in Mexico 421

men who have a high level of control over theirfemale partners may be expected to both restrictthe latter’s participation in the labor market andphysically harm them. In fact, some of the sur-vey questions used to measure control or powerin relationships in the studies just described spe-cifically refer to whether men have influenceover their partner’s employment or other activi-ties outside the home. For example, the indexsuggested by Blood and Wolfe (1960) and usedby Coleman and Straus (1990), as well asothers, asks whether a man has the final say onwhether his partner should go to work or quitwork, and what job she should take. Women’semployment status cannot therefore be assumedto be independent of the level of control in mul-tivariate regression analyses. I argue that femaleemployment should instead be treated as anendogenous variable.

This endogenous model is illustrated in thesimplified path diagram shown in Figure 1 inwhich other predictors of employment and vio-lence have been omitted for simplicity. As indi-cated by the signs in each path in the diagram,a higher level of coercive control by men overtheir partners is expected to lower the probabilityof female employment and increase the risk ofviolence. By contrast, the direction of associationbetween women’s employment and the risk ofintimate partner violence is difficult to predictbecause there are competing arguments in bothdirections as discussed above. An important fea-ture of the proposed model is captured by thedotted line representing the correlation betweenthe error terms in the statistical equations usedto predict women’s employment and their riskof intimate partner violence. As explained in theanalytical strategy section below, allowing theerror terms to be correlated is necessary forobtaining unbiased estimates of the effect of

employment on violence. Conceptually, how-ever, the fact that the error terms are allowed tobe correlated means that the extent to whichunmeasured characteristics of women affect boththeir likelihood of being employed and their riskof violence is explicitly taken into account in themodel and measured empirically in the analysisthat follows.

Other Predictors of Violenceand Employment

In addition to the main variables of interest shownin Figure 1, the regression models below alsocontrol for women’s basic demographic character-istics, including their age, educational attainment,indigenous background, and urban residence.Research on domestic violence in Mexico hasconsistently found women with lower educa-tional levels to be at greater risk of intimatepartner violence (Dıaz-Olavarrieta et al., 2002;Dıaz-Olavarrieta & Sotelo, 1996; Rivera-Riveraet al., 2004). Similarly, because indigenous groupsare among the most disadvantaged sectors ofMexican society, I expect them to be at higherrisk of domestic violence in direct analogy withresults for minority groups in the United States(Gelles, 1993; Lockhart, 1987; Tjaden & Thoennes,2000). Previous work has also found con-siderably higher rates of homicidal violence inMexican rural areas compared to urban areas(Villarreal, 2002, 2004), although it is not clearwhether this higher incidence of violence is spe-cifically directed against women.

Several characteristics of Mexican householdsmay also affect the risk of intimate partner vio-lence. First, I expect a greater number of peopleliving within the same residential space anda greater number of children to increase the over-all level of stress and lead to the use of violence.Research has found Mexican women living inhouseholds with more children to be at greaterrisk of violence (Dıaz-Olavarrieta et al., 2002;Dıaz-Olavarrieta & Sotelo, 1996). Second, it isa relatively common practice for Mexican house-holds to include members other than thosebelonging to the nuclear family. I hypothesizethat the presence of extended household mem-bers will reduce violence between a couple (con-trolling for the overall level of crowding) becausethey might intervene on the woman’s behalf andexert greater control over her partner’s violentbehavior. This hypothesis is consistent withfindings from Finkler’s (1997) study of domestic

FIGURE 1. RELATION BETWEEN COERCIVE CONTROL BY

A PARTNER, FEMALE EMPLOYMENT, AND INTIMATE

PARTNER VIOLENCE (OTHER PREDICTORS OMITTED)

ControlViolence

Employment–

+

+/–

1

2

1,2

422 Journal of Marriage and Family

violence in a sample of patients in a Mexicanclinic. She finds that married women living innuclear households have a higher reported inci-dence of marital conflict and violence than thoseliving with extended family members.

The marital status of the couple is also animportant predictor of intimate partner violence.Research in the United States has found thatwomen in cohabiting relationships are at higherrisk of violence by their partners (Stets, 1991;Stets & Straus, 1990; Yllo & Straus, 1981).Although women who are cohabiting may befreer to leave abusive relationships and maytherefore be expected to experience lower levelsof violence, researchers have argued that theirgreater isolation and the implied lower level ofcommitment by the male partner may actuallyplace them at greater risk. More recently, Kenneyand McLanahan (2006) argue that the higher rateof domestic violence among cohabiting couplesis a result of a selection effect whereby the leastviolent cohabiting couples are more likely tomarry and the most violent married couples aremore likely to divorce. A couple’s marital statusmay also be confounded with the length of therelationship because married couples are morelikely to have been together for a longer periodof time. Women who have been together withtheir partners for longer may be at lower risk ofviolence among other reasons because violent in-cidents are a precipitating factor for terminatinga relationship.

A woman’s partner’s employment status mayalso affect the risk of intimate partner violence.Unemployment not only creates stress and frus-tration but also allows the male partner to remainat home for longer periods of time, thereforeincreasing the probability of violent interaction.Women whose partners are unemployed havebeen found to be at greater risk of violence instudies of the general population in the UnitedStates (Hornung et al., 1981; Straus et al.,1980). One study among Hispanic couples in par-ticular finds a substantially higher risk of violenceagainst women whose husbands are unemployed(Straus & Smith, 1990).

Research on the intergenerational transmis-sion of violence suggests that having grown upin a violent household as children increasesthe risk of victimization later in life (see Stithet al., 2000, for a review). Mexican studies ofdomestic violence consistently find women’s ex-perience of physical abuse as children to be sig-nificant predictors of intimate partner violence

(Dıaz-Olavarrieta et al., 2002; Rivera-Riveraet al., 2004). In the statistical analysis below,I therefore control for whether a woman grewup in a household where there was violence.

Finally, women’s employment status is alsomodeled as a function of many of the same pre-dictors as domestic violence in the statistical anal-ysis below, including most importantly coercivecontrol by their partners. In addition, three otherpredictors of women’s employment are intro-duced: a woman’s work orientation, the amountof financial help she receives from relatives,and the number of preschool children currentlyliving with her. Women who are more committedto their right to work outside the home will bemore likely to seek employment. By contrast, re-ceiving financial aid from relatives will reducethe need for securing additional income throughwork and is therefore expected to reduce awoman’s probability of participating in the laborforce. Finally, the presence of preschool children(6 years of age or younger) will generally hampera woman’s ability to manage family and work ob-ligations, and will therefore reduce her odds ofworking outside the home. The appropriatenessof these three instrumental variables used in theendogenous model is discussed in greater detailbelow.

METHOD

Sample

The data used in this study are drawn from theNational Survey on Household RelationshipDynamics (Encuesta Nacional sobre la Dinami-ca de las Relaciones de los Hogares) carriedout in 2003 by the Mexican National Institutefor Statistics, Geography and Informatics (INE-GI), the governmental organization responsiblefor the decennial census (INEGI, 2004b). Thesampling strategy involved first creating a rosterof all residents in a national probability sampleof Mexican households. In each household, awoman 15 years of age or older currently livingwith a male partner was then selected at randomout of all household residents meeting these cri-teria. An in-person interview was conductedwith each woman selected. The dataset included33,709 women for whom a partner could beidentified. Of those cases between 7.1% and11.5% were deleted because of missing valuesdepending on the model (sample sizes are re-ported in each table). Additional tests indicated

Women’s Employment and Violence in Mexico 423

that missing values did not deviate statisticallyfrom the mean. Sampling weights were used inall statistical analyses to adjust for differencesin sampling across communities with differentlevels of urbanization.

Measurement

The key dependent variable in the analysis belowis the occurrence of physical violence againsta woman by her partner during the previous year.Physical violence is defined as an affirmativeanswer by the female respondent to any of the fol-lowing eight questions: In the past 12 months hasyour partner: pushed you or pulled your hair?Tied you up? Kicked you? Thrown an object atyou? Hit you with his hands or an object? Triedto strangle you? Attacked you with a knife orblade? Fired a weapon at you? At the nationallevel, 9.3% of the female respondents reporthaving experienced at least one of these formsof abuse during the previous year. Althoughhigh, this overall incidence of violence is lowerthan some estimates from studies conducted inMexico (Krug, Dahlberg, Mercy, Zwi, & Lozano2002; Oropesa, 1997; Rivera-Rivera et al., 2004)but consistent with others (Dıaz-Olavarrieta et al.,2002). As discussed above, differences in thepopulation sampled, the wording of the specificquestions asked (i.e., which types of violent inci-dents are included), as well as the baseline period(e.g., previous 12 months, length of relationship)make comparisons of the overall rate betweenstudies difficult.

The level of control exercised by a male partneris measured by a woman’s answers to questionsregarding her freedom to perform daily activitiesincluding working, going shopping, and visitingfriends or relatives. Respondents to the surveywere asked questions of the form: ‘‘If you workor wanted to work for pay would you have toask permission from your partner, inform him,or do nothing?’’ Because needing to ask permis-sion to perform these routine activities impliesthat a partner has considerable power over awoman’s actions, answers to these questionsserve as good indicators of coercive control. Re-sponses to the questions were coded as 1 if awoman required permission and 0 if not. Table 1shows the percentage of Mexican women inintimate relations who report needing to askpermission from their partners for each categoryof activities. A staggering 40% of women reportneeding to ask permission to work, and somewhat

smaller percentages need permission to performother activities. More importantly, consistentwith the model of intimate partner violence pro-posed in the previous section, these measures ofmen’s control over their partners’ activities areassociated with both a higher incidence of vio-lence and lower employment rates. As shown inthe aggregate statistics in Table 2, women whorequire permission to work are 64% more likelyto experience violence and 52% less likely towork.

In the analysis below, I use two alternativemeasures of coercive control. First, I createa scaled measure of control using the answers toall four questions. The alpha reliability coeffi-cient is high (.84), suggesting that these fourquestions are indeed measuring the same under-lying phenomenon. Nevertheless, because someof these questions do not specifically ask aboutmen’s control over their partner’s employmentstatus, it could be argued that the assumed causalrelation between coercive control and employ-ment is in fact reversed: Women’s employmentmay affect the level of control exercised by herpartner over such things as going shopping or vis-iting friends or relatives. Women who work maybe freer to perform these daily activities. For thisreason, I test additional models in which the levelof control is measured using only answers to thefirst question regarding whether a woman needspermission from her partner to work. Using onlyanswers to this question is particularly important

Table 1. Percent Women Who Report Needing to Ask

Permission for Daily Activities

Need to Ask Permission to %

Work 39.9

Go shopping 17.5

Visit relatives 23.9

Visit friends 23.8

Note: N ¼ 31,765.

Table 2. Percent of Women Who Are Employed and Report

Violence by Whether They Need Permission to Work

Permission % Employed % Violence

Does not need permission

to work

48.1 7.5

Needs permission

to work

23.3 12.3

Note: N ¼ 31,765.

424 Journal of Marriage and Family

in the second part of the analysis where women’semployment is treated as an endogenous variable.

In the models of intimate partner violencetested below, the level of control exercised bya woman’s partner is used as a predictor of herrisk of violence. Nevertheless, following thework of Johnson (1995) and Johnson and Ferraro(2000), it may be argued that coercive control isnot a separate measure from violence but insteadforms part of a broader pattern of violence andcontrol. To establish the discriminant validity ofthe measure of coercive control used here, I per-formed some additional tests. First, I employedalpha scoring using the responses to the fourquestions about coercive control along with themeasure of physical violence and found violenceto be weakly correlated with the remaining items(item-rest correlation ¼ .10). Second, physicalviolence also loaded poorly with the four itemsfor coercive control when the principal factormethod was used (uniqueness ¼ .98). This evi-dence suggests that my variables for coercivecontrol and violence are measuring differentphenomena.

The regression models for physical violencecontrol for basic sociodemographic charac-teristics including a woman’s age. A woman’seducational attainment is measured using fourcategories: primary education or less, middleschool, high school or technical education, andcollege or more. Membership in one of the coun-try’s indigenous groups is measured using indi-viduals’ ability to speak an indigenous language.Although clearly an imperfect measure, languageproficiency is the standard way in which indige-nous groups are identified in the Mexican census.According to the 2000 Mexican population cen-sus, 7.2% of women 15 years of age or older be-longed to an indigenous group in 2000 (INEGI,2001), compared to 7.8% of women 15 years orolder in the (weighted) sample. An indicator ofurban residence is also introduced in the regres-sion models to control for differences in the levelof violence between urban and rural areas.

Three variables related to the household sizeand composition are included in the regressionmodels. First, the total number of residentsdivided by the number of rooms in the dwellingis used to account for a possible effect of crowd-ing. Second, I also control for the total numberof children a respondent has currently livingwith her. Third, I introduce a variable indicatingthe presence of extended family members inthe household. At the national level, 32% of the

women surveyed live in households with nonnu-clear members (a nuclear family is defined as awoman, her partner, and any children they mayhave).

A woman’s marital status and the number ofyears she has been living with her partner are alsointroduced as predictors of violence. Followingmy discussion above, I include a woman’s part-ner’s employment status as a predictor of vio-lence. Unemployment is expected to increasethe risk of violent incidents. A partner’s incomefrom wages, however, is not included as a predic-tor in the statistical analysis below because it ishighly collinear with his employment status (onlyemployed partners report income). Nevertheless,in separate models not presented here, a partner’slogged income from wages was not found tobe a significant predictor of violence when usedinstead of employment. A woman’s wage incomeis excluded from the model for the same reason,although the financial assistance she receivesfrom relatives is included in the final modelbelow.

The regression models control for whether awoman grew up in a household where there wasviolence (regardless of who was the aggressorand who was the target of aggression). Althougha woman’s partner’s experience as a child wouldadmittedly be a better predictor than her ownexperience, information about a partner’s child-hood is provided by the female respondent whooften does not know the circumstances in whichhe grew up, including whether there was violencein his family of origin. A variable constructedusing information about men’s childhood experi-ence provided by their partners generated toomany missing values (but confirmed the resultsdiscussed below in separate analyses not pre-sented here).

Finally, three instrumental variables areincluded as predictors of a women’s employmentstatus: a woman’s work orientation, the amountof financial help she receives from relatives,and the number of preschool children currentlyliving with her. Work orientation is defined usingrespondents’ answer to the question whethera woman should be free to work if her husband’ssalary is not enough. Answers to this questionmay be said to capture a respondent’s own opin-ion regarding women’s economic role. Second,the amount of financial assistance from relativesis introduced in the regression models. Third,the presence of preschool children (6 years ofage or younger) is expected to reduce a woman’s

Women’s Employment and Violence in Mexico 425

ability to work. The effect of preschool children,however, is measured after controlling for thetotal number of children a woman has becausethe latter is included as a predictor of both em-ployment and violence. Kalmuss and Straus(1990) include the presence of children 5 yearsof age or younger in their measure of marital de-pendency. Nevertheless, they specifically arguethat having young children reinforces a woman’sdependence on her partner because they limit herparticipation in the labor force (p. 372). In otherwords, the effect of preschool children on a wom-an’s risk of violence is mediated by their effect onher employment status, consistent with the modelpresented here.

Analytical Strategy

In the conceptual framework developed in thepreceding sections, I argued that the level ofcontrol exercised by a woman’s partner is onecharacteristic that is likely to affect both her prob-ability of employment and her risk of intimatepartner violence. Yet, there are other characteris-tics affecting both outcomes, many of which arenot included in the statistical models in partbecause they are difficult or impossible to mea-sure. The effect of such unmodeled or unob-served characteristics are captured in the errorterms corresponding to the equations for wom-en’s employment and violence in the statisticalmodels below. It may be shown, however, thatif the same unmeasured characteristics affect botha woman’s employment status and her risk of vio-lence, and her employment status is included asa predictor of violence, then the estimate of theassociation between employment and violenceobtained from standard regression techniqueswill be biased (for a general discussion, seeGreene, 2003, chapter 5). This is known as endo-geneity bias in the econometric literature. It is forthis reason that the error terms in the equations foremployment and violence are explicitly allowedto be correlated in the complete model of intimatepartner violence below. As a comparison of theresults of the regression model allowing the er-rors to be correlated (referred to as an endogenousmodel below) and those from the model that doesnot will show this aspect of the model specifica-tion makes a significant difference and, in fact, re-verses my estimates of the effect of women’semployment on violence.

Statistical models with endogenous variablesand correlated errors are often estimated with

techniques developed for structural equations.Traditional procedures for estimating structuralequation models are ideal when the dependentvariables are continuous, but they are not wellsuited for dealing with models involving twodichotomous endogenous variables such as thosein this study, namely, women’s employment anddomestic violence. In a recent contribution to thisjournal, Gibson-Davis et al. (2005) estimate amodel where women’s employment is treated asan endogenous variable using the two-stage leastsquares method by coding employment as a con-tinuous variable given by the total number ofquarters in which a woman is employed duringa 2-year interval. Unfortunately, the measure ofemployment available in the Mexican survey,as in many other surveys, is dichotomous. For thisreason, I employ a maximum likelihood methodto estimate a statistical model where both em-ployment and violence are modeled as probitfunctions of other predictors and the errorscorresponding to both equations are allowed tobe correlated. Specifically, the error terms areassumed to have a bivariate normal distribution,and their correlation is estimated. Because theunderlying equations for both variables havea probit form, these models are often referred toas bivariate or multivariate probit models, butthey are distinguished from ordinary probitmodels precisely in that the error terms for twoor more dichotomous variables being predictedare jointly distributed using a higher order nor-mal distribution (see Greene, 2003, pp. 710 –719, for a general discussion of multivariateprobit models). A crucial statistical property ofbivariate probit models is that one of the dichot-omous variables may be included as a predictorof the other even without the use of instrumen-tal variables, so long as the errors are correlated.This property is demonstrated by Greene (2003,pp. 715 – 716) (see also Maddala, 1983, pp. 122 –125; Wilde, 2000; and Greene, 1998). Theendogenous probit model presented below isestimated using the statistical software packageaML (version 2.08) developed by Lee Lillardand Constantijn Panis (Lillard & Panis, 2003).

Although they are not strictly necessary,instrumental variables will help to strongly iden-tify the endogenous probit model and lead toimproved estimates of the endogenous effect.As explained above, three variables are used asinstruments for a woman’s employment status:her work orientation, the financial help she re-ceives from relatives, and the number of

426 Journal of Marriage and Family

preschool children currently living with her.These three variables can reasonably be expectedto be good predictors of a woman’s employmentstatus. More importantly, their effect on domesticviolence is likely to be mediated by a respondent’semployment status. By contrast, the amount ofresidents per room, the presence of violence inthe partner’s family of origin, and the length ofa woman’s current relationship cannot reason-ably be expected to predict her employment sta-tus and yet may be thought to predict intimatepartner violence for the reasons mentionedabove. They are therefore only included as pre-dictors of violence and excluded from theemployment equation. It is important to note,however, that even though a woman’s work ori-entation, the financial help she receives from rel-atives, and the number of preschool children areused as instruments in the regression modelsbelow for substantive reasons, their inclusion as

predictors of violence in additional models notpresented here did not alter the principal resultregarding the direction of association betweenwomen’s employment and their risk of violence.In other words, regardless of whether these threevariables are modeled as instruments, the mainfinding reported below holds true.

Results of a Baseline Model

Table 3 shows the results of simple probit re-gression models of physical violence againstMexican women in intimate relationships.Women’s employment is treated as an exogenouspredictor of violence in these models and isassumed to be uncorrelated with the error term.The positive and significant coefficient ofwomen’s employment status seems to suggest thatemployed women are at greater risk of violencefrom their partners when all other relevant

Table 3. Summary of Probit Regression Models Predicting Intimate Partner Violence

Variable

Model 1 Model 2

B SE B B SE B

Sociodemographic characteristics

Age �0.030** 0.005 �0.028** 0.005

Middle school education �0.043 0.080 �0.017 0.080

High school education or technical �0.073 0.093 �0.035 0.094

College education or more �0.112 0.165 �0.078 0.162

Indigenous �0.213** 0.081 �0.236** 0.077

Urban areas 0.181* 0.071 0.179* 0.070

Partner employed 0.102 0.110 0.141 0.105

Household characteristics

Residents per room 0.052* 0.024 0.039 0.024

Total number of children 0.013 0.018 0.011 0.018

Nonnuclear household �0.040 0.068 �0.028 0.067

Relationship characteristics

Married �0.275** 0.071 �0.279** 0.069

Length of relationship 0.020** 0.006 0.020** 0.005

Intergenerational transmission of violence

Violence in family of origin 0.448** 0.060 0.439** 0.059

Control

Coercive control (dichotomous) 0.247** 0.063

Coercive control (scale) 0.225** 0.034

Employment

Woman employed 0.205** 0.059 0.202** 0.060

Constant �0.958** 0.221 �0.942** 0.205

Log likelihood �8724.5 �8947.7

N 30,041 31,324

Note: Weighted sample.

*p , .05. **p , .01.

Women’s Employment and Violence in Mexico 427

characteristics, including her partner’s employ-ment status, are taken into account. Further statis-tical tests not presented here indicate nointeraction effect between a woman’s employ-ment status and that of her partner in contrast tothe findings of Macmillan and Gartner (1999).Second, as expected, coercive control by a malepartner significantly increases a woman’s riskof violence. Both relations hold regardless ofwhether coercive control is defined as a dichot-omous variable indicating if a woman specificallyneeds permission to work (Model 1) or as a scaledvariable using answers to all four questions(Model 2). The choice of which measure ofcoercive control is used does not substantivelyaffect the results for the remaining variables inthe models.

The results also indicate that older women areat lower risk of physical violence, which is gener-ally consistent with prior research across manycountries (Krug et al., 2002). A woman’s overalleducation level does not appear to affect her prob-ability of experiencing violence. Contrary toexpectations derived from findings for disadvan-taged minority groups in the United States,women who are members of an indigenous groupare considerably less likely to suffer physical at-tacks from their partners. Although there is insuf-ficient information to reach a firm conclusion, onepossible interpretation for this finding is that fam-ily practices particular to the various indigenousgroups in Mexico discourage the use of violenceagainst female partners. Alternatively, the lowerrates of violence may be the result of under-reporting among this particular group of women.Also contrary to expectations, the presence ofextended family members in the household doesnot discourage the use of violence. Further testingin which the specific relation of extended house-hold members to the respondent (i.e., mother,father) is taken into account also failed to yieldstatistically significant results. Women in urbanareas are found to have a higher incidence ofdomestic violence. This result conflicts with pre-vious findings on homicidal violence in Mexico,which show a higher overall homicide rate inrural areas (Villarreal, 2002, 2004).

Consistent with results in the United Statesas well as other countries, married women inMexico have a lower incidence of domestic vio-lence and those who are in more longstandingrelationships have a higher incidence. Crowdingappears to marginally increase the probability ofviolence from intimate partners, thus supporting

the notion that a greater number of people livingin close proximity creates stress and increasesconflict between the couple. Finally, havinggrown up in a violent family as a child sharply in-creases a woman’s risk of violence by her partnerlater in life. In fact, this measure of violence in thefamily of origin is one of the strongest predictorsof intimate partner violence.

Endogenous Model

In the previous section I tested models of intimatepartner violence in which a woman’s employ-ment status and coercive control were includedas predictors. The results indicated that employedwomen are at greater risk of domestic violence.This finding is consistent with resource theory,which predicts that male partners will compen-sate for their lack of dominance in the economicrealm by using violence. Nevertheless, as dis-cussed in the theoretical section above, a wom-an’s employment status is not an exogenousfactor but is instead also partly explained by thelevel of coercive control exercised by a male part-ner. Moreover, many of the same unmodeledcharacteristics of Mexican women that increasetheir risk of violence are also likely to be associ-ated with their odds of being employed. In thissection, I test a model of intimate partner violencein which women’s employment is considered anendogenous variable, and the errors in the equa-tions corresponding to women’s employmentand their risk of violence are allowed to be corre-lated. I use the dichotomous variable for coercivecontrol as a predictor instead of the continuousmeasure. This variable specifically measureswhether a woman needs permission from herpartner to work. As explained above, using thisspecific variable ensures that the assumed causaldirection between control and employment is notin fact reversed, that is, that coercive control bya partner affects a woman’s probability ofemployment and not the other way around. Forinstance, it could be argued that employedwomen are less likely to need permission to goshopping or visit friends or relatives (the otheritems used in the scaled measure of coercive con-trol); yet, it would be difficult to sustain thata woman’s employment affects her likelihoodof needing permission to work.

Table 4 shows the results of the endogenousmodel. The most important finding is the negativecoefficient associated with a woman’s employ-ment status when used as a predictor of her risk

428 Journal of Marriage and Family

of violence. Employed women are less likely toexperience intimate partner violence. This resultis a complete reversal of that in Table 3 above,where employment is assumed to be an exoge-nous variable. The new finding supports martialdependency theory instead of resource theoryinsofar as employed women, who are lessdependent on their partners for economic re-sources, have a lower risk of violence. Thereversal of the coefficient also indicates theimportance of treating female employment asan endogenous variable predicted by some ofthe same modeled and unmodeled characteris-tics of women as the risk of intimate partnerviolence. As noted in the bottom of Table 4, thecorrelation between the error terms for employ-ment and violence (q1,2) is statistically signifi-

cant indicating that the null hypothesis that theerror terms are independent can be rejected.Because the correlation between the error termsis positive, we know that the unmodeled charac-teristics of women either increase both theirprobability of being employed and their risk ofviolence or decrease them both.

As expected, greater control by male partnersover women’s employment decisions reducedtheir probability of employment even when allother relevant factors are controlled. In otherwords, needing permission from their partnersto work significantly reduces women’s likelihoodof employment. The coefficients for the instru-mental variables are all highly significant, thusconfirming their choice as instruments: Havingmore preschool children substantially reduces

Table 4. Summary of Multivariate Probit Regression Model Predicting Intimate Partner Violence With Women’s

Employment as Endogenous Variable

Variable

Employment Violence

B SE B B SE B

Sociodemographic characteristics

Age �0.005** 0.001 �0.029** 0.002

Middle school education 0.068** 0.022 �0.022 0.029

High school education

or technical

0.344** 0.022 0.014 0.039

College education or more 0.631** 0.031 0.040 0.061

Indigenous 0.224** 0.031 �0.155** 0.046

Urban areas 0.202** 0.021 0.226** 0.029

Partner employed 0.299** 0.031 0.181** 0.050

Household characteristics

Residents per room 0.045** 0.010

Total number of children 0.036** 0.006 0.012 0.008

Nonnuclear household �0.034* 0.017 �0.041 0.022

Relationship characteristics

Married 0.030 0.021 �0.260** 0.026

Length of relationship 0.020** 0.002

Intergenerational transmission of violence

Violence in family of origin 0.429** 0.024

Instruments for employment

Preschool children �0.221** 0.012

Work orientation 0.187** 0.023

Financial help from relatives �0.137** 0.024

Control

Coercive control �0.507** 0.017 0.111* 0.047

Employment

Woman employed �0.426* 0.169

Constant �0.620** 0.060 �0.787** 0.101

Note: Weighted sample. N ¼ 29,837. Log likelihood ¼ �26769.8730. q1,2 ¼ 0.386**.

*p , .05. **p , .01.

Women’s Employment and Violence in Mexico 429

the probability of being employed, controlling forthe total number of children, whereas the latterhas no direct effect on the risk of violence. Awoman’s work orientation increases the likeli-hood of employment, and financial help from rel-atives reduces it. The coefficients for some of theremaining predictors of employment confirmsome expectations, whereas others reveal someunexpected patterns of association. Having moreeducation and living in urban areas understand-ably increase the probability of employment.More interestingly, being a member of an indige-nous group increases the likelihood of employ-ment among women. This pattern may reflectthe necessity of seeking employment for mem-bers of this highly disadvantaged group. Mar-riage appears to have no effect on women’semployment decisions, and the presence ofextended family members decreases women’sprobability of being employed. Finally, havinga working partner actually increases a woman’sprobability of employment. This relation mayhave to do with the partner sorting process bywhich employed women are more likely to findpartners who are also employed.

DISCUSSION

In this article, I have proposed a conceptualframework in which a woman’s employment sta-tus and her risk of intimate partner violence areboth influenced by the level of control exercisedby her partner. Controlling men actively preventa woman from working and are also more likelyto physically harm their partners. Although previ-ous studies have shown that controlling men aremore likely to harm their female partners, theyhave rarely measured the influence that thesesame men have on women’s employment deci-sions. Part of the reason that a partner’s control in-creases a woman’s risk of violence is preciselythat it reduces her ability to work, thereby makingher more dependent. Mexican men appear toexert considerable influence over their partner’semployment decisions, as evidenced by the factthat 40% of women report needing permissionto work for pay.

An important feature of the statistical modelspresented in this article is that they take intoaccount the extent to which unmeasured andunobserved characteristics of women simulta-neously affect their likelihood of being employedand their risk of intimate partner violence. Whenthe simultaneous effects that these unmeasured

characteristics have on women’s employmentand on their risk of violence were taken intoaccount, the estimated effect of employment onviolence was the opposite of that found in simplermodels with uncorrelated errors. The positive andstatistically significant correlation among theerror terms in the equations for employmentand violence suggests that unmeasured character-istics of Mexican women are increasing both theirprobability of being employed and their risk ofviolence, or decreasing both, and that the failureto account for the joint distribution of the errorterms leads to biased estimates in the simplermodel.

According to the final endogenous model, em-ployed women are at lower risk of intimate part-ner violence. This finding is consistent withmarital dependency theory according to whichemployed women are less dependent on theirpartners for support and are therefore less likelyto tolerate abuse. If true, this bodes well for thefuture because women’s labor force participa-tion is on the rise not only in Mexico but inmany other countries as well. Developing coun-tries such as Mexico provide a particularly goodtest for resource theory, which generally pre-dicts a higher incidence of violence against em-ployed women. If violence toward women isa compensatory mechanism for men’s loss ofdominance in other domains such as the eco-nomic domain, then we should expect the posi-tive effect of women’s employment status ontheir risk of violence to be greater in a countrysuch as Mexico where women’s advances in theeconomic realm are more recent and where tra-ditional norms regarding women’s roles arelikely to lag behind. Yet, female employment isfound to decrease the probability of violence.

The findings from this study also have impor-tant implications for research on women’s laborforce participation. A large body of work hasexamined the determinants of female employ-ment (Goldin, 1990; Mammen & Paxson, 2000;Rosenfeld, 1996; Yu, 2001). Studies often focuson the impediments to women’s participation inlabor markets such as the difficulty in balancingwork with childrearing. These sorts of impedi-ments are indeed important, as the statistical anal-ysis above showed. But research on the correlatesof female labor force participation has giveninsufficient attention to the influence that partnersmay have over women’s employment decisions.In Mexico at least, men exert substantial controlover their partner’s choice regarding whether to

430 Journal of Marriage and Family

work. Clearly, needing permission to seekemployment is an extreme form of control overa woman’s behavior. Yet, male partners mayinfluence women’s decisions to work and whichform of employment they seek in more subtleways that are yet to be fully explored. Furtherwork is still required to fully understand whenand how women’s employment decisions areinfluenced by their partners.

Limitations

The analysis presented in this article has severallimitations. First, the statistical model in whichemployment is treated as an endogenous variableis tested using data from a single survey con-ducted in 2003. Data from a clinical trial in whicha group of women are randomly assigned to workor worker training, whereas others are not are notcurrently available in Mexico. Gibson-Daviset al. (2005) use such information from a sampleof welfare recipients in the United States andutilize the random assignment of women intoworking and nonworking groups to identifya two-stage least squares model. Their model isnot intended to measure the effect of a partner’scontrol over a woman’s employment and her riskof violence. Nevertheless, Gibson-Davis et al.find evidence that ignoring the extent to whichunmeasured characteristics of women affectboth their level of employment and their risk ofviolence leads to biased estimates of the effectof employment (in the same direction as thebias found in this study). Using cross-sectionaldata imposes some limitations compared to theresearch design used by Gibson-Davis et al. Ran-dom assignment clearly provides a more rigorousinstrument than those available in this study. Thecross-sectional models presented here are alsosusceptible to questions about causal ordering.Yet the strategy proposed here relies on the useof an endogenous probit model, which may beidentified without the use of instruments, andmay therefore be of use to many researcherswho typically do not have data generated throughsuch complex experimental designs as Gibson-Davis et al. At the very least, the method pro-posed here may serve as a diagnostic in othercross-sectional studies of domestic violencewhere variables such as women’s employment(as well as others) may be suspected of beingendogenous.

Second, a partner’s control over a woman wasspecifically measured as her need to obtain

permission to work precisely to avoid the possi-bility of the reverse causal ordering, namely,that a woman’s employment leads to a lower (orpossibly higher) level of control by her partner.Nevertheless, I cannot be certain that the causaldirection is not reversed, nor that violence doesnot instead (or simultaneously) affect a woman’semployment or the level of control exercised byher partner. A reasoned argument for the pro-posed model has been made using existing theoryand empirical research. Yet, other possibilitiesremain. More research is clearly required to fullydisentangle the various causal connections.

Finally, although the proposed model of inti-mate partner violence represents an improvementover one that treats women’s employment as anindependent predictor of violence, it may stillbe argued that the model makes the unreasonableassumption that the level of control exercised bya man over his female partner is unaffected by thesame unmeasured characteristics that affect therisk of violence. Previous studies have in factfound coercive control and power in a relationshipto be a function of some of the same predictorsas violence (i.e., Macmillan & Gartner, 1999;Oropesa, 1997). In separate analyzes not pre-sented here, I tested several models in which bothfemale employment and control are treated asendogenous predictors of violence. The resultsgenerally supported the main finding reportedabove, namely, that women’s employment re-duces and does not increase her risk of violenceby her partner, although the effect of coercivecontrol on women’s employment was oftenreversed in these models.

NOTE

Research was supported by a grant from the Harry Frank Gug-genheim Foundation. I thank Daniel Powers and ConstantijnPanis for helpful advice on statistical methods. I also thankWilliam Parish for useful comments on a previous draft of thispaper.

REFERENCES

Allen, M., & Straus M. (1980). Resources, power

and husband-wife violence. In M. Straus & G.

Hotaling (Eds.), The social causes of husband-wifeviolence (pp. 188 – 208). Minneapolis: University

of Minnesota Press.

Anderson, K. (1997). Gender, status, and domestic

violence: An integration of feminist and family

violence approaches. Journal of Marriage and theFamily, 59, 655 – 669.

Women’s Employment and Violence in Mexico 431

Blood, R., & Wolfe, D. (1960). Husbands and wives.

Glencoe, IL: Free Press.

Coleman, Diane H., and Murray A. Straus. (1990).

Marital power, conflict, and violence in a nationally

representative sample of American couples. In

M. Straus & R. Gelles (Eds.), Physical violencein American families: Risk factors and adaptationsto violence in 8,145 families (pp. 287 – 304).

New Brunswick, NJ: Transaction.

DeMaris, A., Benson, M., Fox, G., Hill, T., &

Van Wyk, J. (2003). Distal and proximal factors

in domestic violence: A test of an integrated

model. Journal of Marriage and Family, 65,

652 – 667.

Dıaz-Olavarrieta, C., Ellertson, C., Paz, F., Ponce de

Leon, S., & Alarcon-Segovia, D. (2002). Preva-

lence of battering among 1780 outpatients at an

internal medicine institution in Mexico. Social Sci-ence & Medicine, 55, 1589 – 1602.

Dıaz-Olavarrieta, C., & Sotelo, J. (1996). Domestic

violence in Mexico. Journal of the American Med-ical Association, 275, 1937 – 1941.

Dobash, R., & Dobash, R. (1979). Violence againstwives: A case against the patriarchy. New York:

Free Press.

Farrington, K. (1986). The application of stress the-

ory to the study of family violence: Principles,

problems and prospects. Journal of Family Vio-lence, 1, 131 – 147.

Finkelhor, D. (1983). Common features of family

abuse. In D. Finkelhor, R. Gelles, G. Hotaling, &

M. Straus (Eds.), The dark side of families: Cur-rent family violence research (pp. 17 – 28). Bev-

erly Hills, CA: Sage.

Finkler, K. (1997). Gender, domestic violence and

sickness in Mexico. Social Science & Medicine,

45, 1147 – 1160.

Fox, G., Benson, M., DeMaris, A., & Van Wyk, J.

(2002). Economic distress and intimate violence:

Testing family stress and resource theories. Jour-nal of Marriage and Family, 64, 893 – 807.

Garcıa, B., & de Oliveira, O. (1994). Trabajo femeninoy vida familiar en Mexico [Female work and familylife in Mexico]. Mexico City, Mexico: El Colegiode Mexico.

Gelles, R. (1993). Through a sociological lens:

Social structure and family violence. In R. Gelles &

D. Loseke (Eds.), Current controversies onfamily violence (pp. 31 – 46). Newbury Park, CA:

Sage.

Gibson-Davis, C., Magnuson, K., Gennetian, L., &

Duncan, G. (2005). Employment and the risk of

domestic abuse among low-income women. Jour-nal of Marriage and Family, 67, 1149 – 1168.

Goldin, C. (1990). Understanding the gender gap:An economic history of American women. New

York: Oxford University Press.

Gonzalez de la Rocha, M. (1994). The resources ofpoverty: Women and survival in a Mexican city.

Oxford, U.K.: Blackwell.

Goode, W. (1971). Force and violence in the family.

Journal of Marriage and the Family, 33, 624 – 636.

Greene, W. (1998). Gender economics courses in lib-

eral arts colleges: Further results. Journal of Eco-nomic Education, 29, 291 – 200.

Greene, W. (2003). Econometric analysis (5th ed.)

Upper Saddle River, NJ: Prentice Hall.

Holtzworth-Monroe, A., Bates, L., Smutzler, N., &

Sandin E. (1997). A brief review of the research on

husband violence. Aggression and Violent Behav-ior, 2, 65 – 99.

Hornung, C., McCullough, C., & Sugimoto, T.

(1981). Status relationships in marriage: Risk fac-

tors in spouse abuse. Journal of Marriage and theFamily, 43, 675 – 692.

Instituto Nacional de Estadıstica, Geografıa e Infor-

matica. (1990). Avance de informacion economica:empleo [Economic information news: Employment].

Aguascalientes, Mexico: Author.

Instituto Nacional de Estadıstica, Geografıa e Infor-

matica. (1993). Estados Unidos Mexicanos: XIcenso general de poblacion y vivienda, 1990. Re-sultados definitivos [United Mexican States: XI

General population and household census, 1990.

Final Results]. Aguascalientes, Mexico: Author.

Instituto Nacional de Estadıstica, Geografıa e Infor-

matica. (2001). Tabulados basicos: Estados Uni-dos Mexicanos: XII censo general de poblacion yvivienda, 2000 [Basic Tables: United Mexican

States: XII General population and household cen-

sus, 2000]. Aguascalientes, Mexico: Author.

Instituto Nacional de Estadıstica, Geografıa e Infor-

matica. (2004a). Estadısticas economicas: Indica-dores de empleo y desempleo [Economic statistics:

Indicators of employment and unemployment].

Aguascalientes, Mexico: Author.

Instituto Nacional de Estadıstica, Geografıa e Infor-

matica. (2004b). Encuesta nacional sobre la dina-mica de las relaciones de los hogares [National

survey on household relationship dynamics].

Aguascalientes, Mexico: Author.

Johnson, M. (1995). Patriarchal terrorism and com-

mon couple violence: Two forms of violence

against women. Journal of Marriage and the Fam-ily, 57, 283 – 294.

Johnson, M., & Ferraro, K. (2000). Research on domes-

tic violence in the 1990s: Making distinctions. Jour-nal of Marriage and the Family, 62, 948 – 963.

432 Journal of Marriage and Family

Kalmuss, D., & Straus, M. (1990). Wife’s marital

dependency and wife abuse. In M. Straus & R.

Gelles (Eds.), Physical violence in American fami-lies: Risk factors and adaptations to violence in8,145 families (pp. 369 – 382). New Brunswick,

NJ: Transaction.

Kaukinen, C. (2004). Status compatibility, physical

violence, and emotional abuse in intimate relation-

ships. Journal of Marriage and Family, 66, 452 –

471.

Kenney, C., & McLanahan, S. (2006). Why are co-

habiting relationships more violent than marriages?

Demography, 43, 127 – 140.

Kishor, S., & Johnson, K. (2004). Profiling domesticviolence: A multi-country study. Calverton, MD:

ORC Macro.

Krug, E., Dahlberg, L., Mercy, J., Zwi, A., &

Lozano, R. (2002). World report on violenceand health. Geneva, Switzerland: World Health

Organization.

Levinson, D. (1989). Family violence in cross-cultural perspective. Newbury Park, CA: Sage.

Lillard, L., & Panis, C. (2003). aML multilevel multi-process statistical software, version 2.0. Los

Angeles: EconWare.

Lockhart, L. (1987). A reexamination of the effects

of race and social class on the incidence of marital

violence: A search for reliable differences. Journalof Marriage and the Family, 49, 603 – 610.

Macmillan, R., & Gartner, R. (1999). When she

brings home the bacon: Labor-force participation

and the risk of spousal violence against women.

Journal of Marriage and the Family, 61, 947 –

958.

Maddala, G. S. (1983). Limited-dependent and quali-tative variables in econometrics. New York: Cam-

bridge University Press.

Mammen, K., & Paxson, C. (2000). Women’s work

and economic development. Journal of EconomicPerspectives, 14, 141 – 164.

Miller, J., & Knudsen, D. (1999). Family abuse and

violence. In M. Sussman, S. Steinmetz, & G.

Peterson (Eds.), Handbook of marriage and thefamily (pp. 705 – 741). New York: Plenum Press.

Oropesa, R. (1997). Development and marital power

in Mexico. Social Forces, 75, 1291 – 1318.

Rivera-Rivera, L., Salmeron-Castro, J., & Castro, R.

(2004). Prevalence and determinants of male part-

ner violence against Mexican women: A popula-

tion-based study. Salud Publica de Mexico, 46,

113 – 122.

Rosenfeld, R. (1996). Women’s work histories.

Population and Development Review, 22, 199 –

222.

Stets, J. (1988). Domestic violence and control. New

York: Springer-Verlag.

Stets, J. (1991). Cohabiting and marital aggression:

The role of social isolation. Journal of Marriageand the Family, 53, 669 – 680.

Stets, J., & Pirog-Good, M. (1987). Violence in dat-

ing relationships. Social Psychology Quarterly, 50,

237 – 246.

Stets, J., & Straus, M. (1990). The marriage license

as a hitting license: A comparison of assaults in

dating, cohabiting and married couples. In

M. Straus & R. Gelles (Eds.), Physical violence inAmerican families: Risk factors and adaptationsto violence in 8,145 families (pp. 227 – 244).

New Brunswick, NJ: Transaction.

Stith, S., Rosen, K., Middleton, K., Busch, A., Lun-

deberg, K., & Carlton, R. (2000). The intergenera-

tional transmission of spouse abuse: A meta-

analysis. Journal of Marriage and the Family, 62,

640 – 654.

Straus, M. (1990). Social stress and marital violence

in a national sample of American families. In M.

Straus & R. Gelles (Eds.), Physical violence inAmerican families: Risk factors and adaptationsto violence in 8,145 families (pp. 181 – 201).

New Brunswick, NJ: Transaction.

Straus, M., Gelles, R., & Steinmetz, S. (1980).

Behind closed doors: Violence in the Americanfamily. Garden City, NY: Anchor.

Straus, M., & Smith, C. (1990). Violence in Hispanic

families in the United States: Incidence rates and

structural interpretations. In M. Straus & R. Gelles

(Eds.), Physical violence in American families:Risk factors and adaptations to violence in 8,145families (pp. 341 – 367). New Brunswick, NJ:

Transaction.

Strube, M., & Barbour, L. (1983). The decision to

leave an abusive relationship: Economic depen-

dence and psychological commitment. Journal ofMarriage and the Family, 45, 785 – 793.

Sugihara, Y., & Warner, J. (2002). Dominance and

domestic abuse among Mexican Americans: Gen-

der differences in the etiology of violence in inti-

mate relationships. Journal of Family Violence,

17, 315 – 340.

Tjaden, P., & Thoennes, N. (2000). Full report of theprevalence, incidence, and consequences of vio-lence against women: Findings from the nationalviolence against women survey. Report NCJ

183781. Washington, DC: National Institute of

Justice.

Trichenor, V. (1999). Status and income as gendered

resources: The case of marital power. Journal ofMarriage and the Family, 61, 638 – 650.

Women’s Employment and Violence in Mexico 433

Villarreal, A. (2002). Political competition and vio-

lence in Mexico: Hierarchical social control in

local patronage structures. American SociologicalReview, 67, 477 – 498.

Villarreal, A. (2004). The social ecology of rural vio-

lence: Land scarcity, the organization of agricul-

tural production, and the presence of the state.

American Journal of Sociology, 110, 313 – 348.

Wilde, J. (2000). Identification of multiple equation

probit models with endogenous dummy regressors.

Economics Letters, 69, 309 – 312.

Yllo, K. (1993). Through a feminist lens: Gender,

power and violence. In R. Gelles & D. Loseke

(Eds.), Current controversies on family violence(pp. 47 – 62). Newbury Park, CA: Sage.

Yllo, K., & Straus, M. (1981). Interpersonal violence

among married and cohabiting couples. Family Re-lations, 30, 339 – 347.

Yllo, K., & Straus, M. (1990). Patriarchy and vio-

lence against wives: The impact of structural

and normative factors. In M. Straus & R. Gelles

(Eds.), Physical violence in American families:Risk factors and adaptations to violence in 8,145families (pp. 383 – 399). New Brunswick, NJ:

Transaction.

Yu, W. (2001). Family demands, gender attitudes,

and married women’s labor force participation:

Comparing Japan and Taiwan. In M. Brinton

(Ed.), Women’s working lives in East Asia (pp.

70 – 95). Stanford, CA: Stanford University Press.

434 Journal of Marriage and Family