The Employment Costs of Domestic Violence

Amy Farmer, University of Tennessee

& Jill Tiefenthaler, Colgate University

While there is much anecdotal evidence that being battered has negative effects on women’s job performance, or even the probability of being employed, there is little research either theoretical or empirical on this issue. The first step in understanding this relationship is to build a theoretical model that recognizes the simultaneity of violence and women’s income/employment. While previous models recognize the effect of women’s income on violence, they ignore the reverse relationship between violence and a woman’s employment income. The violence is likely to increase her likelihood of absence and tardiness, reduce her productivity while on the job and decrease her chances of advancement. In addition, much anecdotal evidence exists that some abusers intentionally attempt to sabotage her work achievements by bothering her at work or even forcing her to quit. The model presented in this paper assumes these negative employment effects exist while the empirical work tests this assumption of the simultaneous effects of employment (hours of work or income) on violence and violence on employment. The theoretical model is also extended to incorporate the possibility of additional strategic behavior by the abuser.

Our original study, Farmer and Tiefenthaler [1997], uses a strategic model to analyze the impact of the woman’s independent alternatives, including income, on the level of violence. The model assumed that the man’s utility is increasing in violence via self esteem enhancement or other psychological factors that might induce a man to batter. The woman’s utility is decreasing in violence and both have utility increasing in consumption of all other goods. Specifically, the woman’s utility function is UW(V, C, () where C = (IW + t)/P represents consumption, IW represents her personal income, t is a transfer of income from him, P is an aggregate price measure and ( accounts for marital specific utility such as the presence of children. Similarly, the man’s utility function is UM (V, C) where C = (IM - t)/P. The woman chooses to stay or leave based upon whether her external utility, denoted  EMBED Equation.2 
, exceeds the utility she receives within the relationship. Her external utility depends partly upon the availability of services as well as her personal income. Her utility within the relationship is affected by violence and the level of financial support she receives from her husband. In addition, the presence of children, for example, may impact her financial needs. Also included in the general model is her ability to affect her income through her labor/leisure decision.

The man’s strategy includes choosing the violence level as well as financial subsidies to the woman. He chooses an optimal combination of violence and financial compensation that maximizes his utility subject to the constraint that she remains in the relationship. This optimization implies that he will raise violence and lower transfers until the constraint becomes binding (i.e. her utility falls to her threat point  EMBED Equation.2 
). Of course, multiple combinations of violence and consumption will provide her with exactly  EMBED Equation.2 
. The figure below illustrates all such combinations on the woman’s indifference curve labeled  EMBED Equation.2 
. Note that her utility is improving toward the origin as his consumption falls (her transfers rise) and violence falls. Among all combinations on this indifference curve (which serves as his constraint), he chooses the optimal point by equating his marginal rate of substitution between violence and consumption with hers. Graphically, this point is the tangency between his indifference curve and hers.

The general predictions from the model are discussed in Farmer and Tiefenthaler [1997], but of particular interest here is the prediction that a rise in her income should decrease violence. The empirical results support the strategic nature of the model. In other words, as her threat point rises his choice of the level of violence falls; this relationship indicates some level of rational, strategic behavior on his part.

In this paper, we build upon this initial research in a couple of ways. First, it is important to recognize that her income not only affects the violence, but that violence may simultaneously affect her income. To incorporate this possibility we model her income as a decreasing function of violence. Specifically, Iw=f(V) where f is a decreasing function. This impacts his optimization decision since the marginal disutility she receives from violence rises. Assuming her external options are the same in either case, this implies that in order to keep her within the relationship, he must either commit a lower level of violence or subsidize her with even larger transfers in order to compensate her for her income loss.

Graphically, as violence lowers her income her indifference curves between V and CM become steeper; specifically, becomes steeper. In other words, when the violence diminishes her income, his level of transfers to her need to be higher for each level of violence in order to keep her utility identical to that of her external options. This additional transfer, represented graphically by a lower level of CM, is needed to compensate her for her lost income.

In either case he forces her to her lowest tolerable level of utility which remains unchanged. In fact, it is he who loses utility since each commission of violence requires additional transfers to compensate for her lost income. Note that the final solution lies on a lower indifference curve for the man, and the level of violence unambiguously falls.

This analysis assumes that her external level of utility is the same whether or not violence affects her income within the relationship. However, a more general model endogenizes the impact of the violence on her threat point. If violence affects her income within a relationship, then the impact may remain long after she leaves. Even if her employment improves somewhat after leaving, it is unrealistic to think there are no lasting affects on her income. In this case, if violence affects her external income, then it also unambiguously lowers her utility. In other words, would represent a lower level of utility for the woman. However, his incentives to commit violence are ambiguous. Preliminary investigation indicates that there will be effects similar to an income and substitution effect. By lowering earning potential and thus her threat point (which is his constraint), additional violence effectively raises his income. This allows him to commit greater violence and provide lower transfers. However, as described above, the relative cost of violence has risen. If the harm done to her earning potential within the relationship exceeds the harm that would persist if she chose to leave, then he must compensate her for this additional loss; if not, she will leave. This higher cost of violence will induce lower violence. However, any lost earning potential that is permanent he need not compensate her for because her ability to leave the relationship has diminished.

The assumption that violence affects a woman’s income as well as her income impacting the level of violence is tested empirically. First, The National Family Violence Survey, 1985 is used to examine how being in a battering relationship affects both a woman’s probability of being employed and her occupational attainment. This data set includes information on the woman’s past and current employment status as well as her occupation. The likelihood that a woman’s past employment status affects her involvement in a violent relationship is accounted for empirically. These data are also used to examine the effects of additional incidents of violence on her probability of being employed and her probability of achieving advancement given that she is in a violent relationship. The National Family Violence Survey also includes information of the number of days of work the woman missed as a result of the violence for those women who report abuse. This information is also used as a dependent variable in regression analysis to investigate how the severity of abuse affects the productivity of women who are employed. This survey, however, does not include information on a woman’s personal income or wage. Consequently, the effects of violence on a woman’s income given that she is in a violent relationship are estimated using Omaha and Charlotte replication studies of the Minneapolis Experiment. Again, the simultaneity of income and violence are accounted for empirically.

Preliminary data analyses support the hypothesis that violence impacts the woman’s income. This result indicates that the abuser could further optimize by strategically affecting her employment in some optimal manner. The theoretical analysis described thus far assumes that violence affects her income in some predetermined manner, and he chooses the violence strategically with that knowledge in mind. However, evidence suggests that his strategic behavior may not stop there. In fact, it may be true that the abuser can intentionally sabotage her employment efforts by harassing her at work, placing bruises in visible places that may increase her absences, etc. This is an additional level of strategic behavior that is currently being investigated in an extension of the model.

We propose first to analyze these strategies of sabotage in a game-theoretic model with asymmetric information in which the abuser chooses a level of intervention into her employment situation. Ultimately the theoretical model will produce predictions concerning when such strategic intervention proves most fruitful for the abuser. We expect that the strategic model will offer specific predictions concerning the demographics of the woman whose income is most likely to be harmed by strategic as opposed to inadvertent intervention into her employment. These predictions may include the type of employment or her level of education. We plan to test these theories with data concerning his behavior as a function of these demographic variables. These empirical tests will assist us in understanding which model provides a more accurate description of the truth based on the demographics of the man and woman. In other words, when might it be expected that a woman’s employment suffers as an unintentional consequence of the abuse, and when might the man be choosing violence with this specific intention in mind? This analysis will not only contribute to our understanding of the determinants of the level of violence, but it makes our estimations of the impact of violence on income more accurate. Ultimately these estimates will provide us with a more accurate measure of the income each woman loses for every incidence of violence conditional on her demographic situation. These figures can then be aggregated across all households using the national demographic figures on victims of domestic violence to determine a national total loss in employment earnings for these victims.