weapon use (4, 6). Crime-related characteristics examined as risk factors for injury have included the victim-offender relationship status (4, 10), presence of others during the incident (6), and history of offender violence against the victim (4).

This study examined how the use of self-protective behaviors affected the risk of injury by using the traditional and revised measures of self-protective behavior. Additionally, women's perceptions regarding the utility of the self-protective behaviors were examined. Finally, we tested whether the association between self-protective behaviors and injury outcome varied depending on the nature of the victim-offender relationship. Other potential risk factors were included as main effect covariates.

Materials and Methods

NCVS procedure

The NCVS was initiated in 1972 and is the only national database of self-reported victimization. Data are collected from a nationally representative sample of individuals over age 12 years who line in a US household. Households are randomly selected, and all age-eligible persons residing in the household are interviewed. A rotating panel design is used, such that once in the sample, household members are interviewed every 6 months for 3 years. After 3 years, the household is rotated out of the panel, and a new household is added (see reference 2 for a more detailed information). In 1992, the NCVS instrument was redesigned, resulting in a more reliable and valid assessment of crimes perpetrated by intimate partners, family members, and acquaintances (10, 11).

Study population

The sample for this study consisted of 3,206 incidents occurring between 1992 and 1995 in which females aged 12 years and older reported a completed or attempted physical assault, i.e., aggravated or simple, perpetrated by a lone male assailant at any time in the 6 months prior to the interview. A physical assault is defined in the NSVS as "an unlawful physical attack or threat of attack". The physical assault category in the NSVS excludes sexual assaults and robberies (2). Consistent with the NCVS categories, sexual assaults and robberies were excluded from this study (n=907). We also excluded assaults perpetrated by females (n=1,444). Series incidents, defined by the NCVS as six or more similar, but distinct, events that the respondent is unable to describe separately in detail, were included (n=247). Victims of series incidents were asked to describe the most recent incident.

Measures

Injury outcome. Our outcome variable was whether or not an injury was sustained due to a physical assault. Injuries were assessed by asking respondents, "What were the injuries you suffered, if any?" The most common injuries were bruises and cuts (85.4 percent); respondents also reported suffering broken bones or teeth (5.4 percent), internal injuries (1.7 percent), unconsciousness (1.9 percent), gunshot and/or stab wounds (1.3 percent), or other, unspecified injuries (18.2 percent).
Self-protective behaviors The use of self-protective behaviors was assessed by asking respondents if there was "anything you did or tried to do about the incident while it was going on?" Those who answered affirmatively were asked what they did. Responses included attacking the offender with a weapon, chasing or trying to catch or hold the offender, yelling at the offender, and reasoning with the offender. Additionally, respondents who reported using a self-protective behavior were asked "Did any of your actions help the situation in any way?" and "Did any of your actions make the situation worse in any way?" Respondents who answered affirmatively to the former question "How were they helpful?" and respondents who answered affirmatively to the latter question were asked "How did they make the situation worse?" Respondents could indicate more than one way in which the self-protective behavior helped the situations or made it worse.

Respondents who engaged in self-protective behaviors and who were injured were asked additional questions regarding whether their use of self-protective behavior occurred before, during, or after the injury. A woman could have utilized one or more self-protective behaviors at different times relative to the injury occurrence.

We used two variables to assess self-protective behaviors. The first variable, "self-protective-traditional," reflected whether or not the victim had engaged in any form of self-protective behavior at any time during the event. The second variable, "self-protective-revised," reflected whether or not the victim used any self-protective measure before or during, but not after, the injury event. The self-protective-revised measure allowed us to classify respondents who engaged in self-protective behaviors only after being injured as not having used them. By comparing findings based on the two self-protective behavior variables, we were able to assess how the two coding approaches ma yield different results.

Victim covariates. We collapsed victim's age into three levels: 12-17, 18-34, and 35 or more years. Victim's marital status had five categories: married, widowed, divorced, separated, and never married. Victim's education was a dichotomous variable based on whether or not the victim had finished high school. Victim's race/ethnicity had four categories: non-Hispanic White, non-Hispanic Black, Hispanic, and other. Victim's urbanicity was a categorical variable with three levels reflecting Bureau of Justice Statistics codes for metropolitan statistical areas (2): urban, suburban, and rural. Victim's household income had three levels: less than $10,000, $10,000-$29,000, and $30,000 or more.

Offender characteristics. We collapsed offender's age into three levels: less than 18, 18-29, and 30 or more years. Offender's race had three categories: White, Black, and other. Weapon use by the offender during the crime had three categories: no weapon, nonfirearm weapon (e.g. sharp or blunt object), and firearm.

Crime-related characteristics. Victim-offender relationship status was assessed with five categories: intimate partner (i.e., spouse, boyfriend, ex-spouse, or ex-boyfriend), other family member, unrelated acquaintance (e.g., neighbor, friend), casual acquaintance, and stranger. The intimate partner group served as the reference category to which the other four relationship categories were compared. Presence of others during the crime incident was a dichotomous variable (yes/no). History of offender violence against victim was a dichotomous variable reflecting whether or not the incident was the first offense for this particular victim-offender pair.

Data analytic strategy

We used logistic regression as our statistical technique because our outcome measure was dichotomous. We first conducted bivariate logistic regression analysis to test the associations of all study variables with risk of injury. Next, we conducted two multivariate logistic regression analyses, one for each of the self-protective behavior variables. Each model included all the covariates, as well as four interaction terms to determine whether the role of self-protective behaviors in predicting injury depended on the victim-offender relationship category. Entering the covariates and self-protective measure into the logistic regression equation simultaneously allowed us to examine the unique effects of each coding of self-protective behavior when controlling for the effects of the covariates.

Because the NCVS uses a complex sample design, variance estimates must be adjusted to account for correlations among the respondents. These correlations exist within the primary sampling units, within the sampled households, and among the data collected at different enumeration periods, due to the rotating panel design. The sample design effect used in this study was 2.5. This conservative estimate was recommended by the Census Bureau (David Hubble and Tracy James Mattingly, Census Bureau, personal communication, 1998) for analyzing assault data. The design effect was determined by calculating direct variances for estimates of physical assaults by using 1994 NCVS data and then estimating the correlation between 1993 and 1994 assault estimates. We incorporated the design effect by multiplying the standard errors by the square root of the design effect, thereby widening the 95 percent confidence intervals and providing a more conservative test of statistical significance.

Results

Univariate findings

Among the 3,206 incidents of physical assault against the women in our sample, 29 percent resulted in injuries. The distribution of incidents by victim, offender, and crime-related characteristics is shown in table 1. The majority of incidents involved self-protective behaviors during the event (74 percent). However, in 6 percent of the assault incidents (30 percent of all injury incidents in which the victim used a self-protective behavior), the self-protective behavior occurred only after the victim was injured. Thus, for the self-protective-revised measure, the percentage of physical assaults involving self-protective behavior is 68 percent.

Table 1. Descriptions of sample (unweighted data) for 3,206 incidents of attemtped or completed physical assaults experienced by female victims of male offenders in the National Crime Victimization Survey, 1992-1995

Bivariate findings

As shown in table 2, the direction of the association between self-protective measures and injury varied depending on the coding of the variable. For the self-protective-revised measure, self-protective behaviors were associated with a decreased likelihood for injury. In contrast, the self-protective-traditional measure had a positive association with injury risk that approached, but did not attain, statistical significance (p=0.05).

Of the three types of covariates, some victim characteristics, one offender characteristic, and all crime-related characteristics were predictive of injury. In terms of victim characteristics, incidents involving women older than 35 years were less likely to result in injury than were those involving victims aged 12-17, and incidents involving married women were less likely to result in injuries than were those involving never-married victims. Although widowed women were least likely to be injured after an assault, the small number of widowed women in the sample (n=72) resulted in wide confidence intervals, and the odds ratio comparing widowed with never-married victims on injury was not significant. Women with household incomes of less than $30,000 were significantly more likely to be injured than were those with incomes greater than or equal to $30,000; women with household incomes less than $10,000 were the most likely to be injured. Women with at least a high school education were less likely than their respective counterparts to sustain an injury during the assault. A victim's urbanicity and race were unrelated to an injury occurrence.

Only one offender characteristic, weapon use, was significantly related to an injury outcome at the bivariate level. Women assaulted by an intimate partner or ex-partner were more at risk of sustaining an injury than were those assaulted by any other type of offender (e.g., other family member, unrelated acquaintance, casual acquaintance, or stranger). Women revictimized by the same offender and those assaulted when no one else was present were also more likely to be injured than were their respective counterparts.

Multivariate findings

The final multivariate models were conducted after removing the interaction terms because these were not significant, indicating that the effect of self-protective behavior (regardless of coding) on risk of injury did not depend on the victim-offender relationship. In the final model using the self-protective-revised variable, the use of self-protective behaviors continued to be inversely associated with injury risk after other potential risk factors were controlled for this, women who used self-protective actions were at lower risk of injury than were their counterparts who did not use self-protective actions (table 3). Three other variables remained significant in the multivariate model. Victims of intimate partner violence were at greater risk of injury than were victims of nonintimate partner violence. Specifically, women assaulted by family members were less than half as likely as those assaulted by intimates to sustain an injury, women assaulted by a nonrelative were about one third as likely as those assaulted by intimates to sustain an injury, and women assaulted by strangers or by acquaintances were approximately one fifth as likely as those assaulted by intimates to sustain an injury. Victim's education also remained a significant risk factor for injury outcomes. Women with less than a high school education were at increased risk of sustaining an injury after a physical assault. Offender weapon use was also significantly related to injury risk. Women assaulted by male assailants with firearms were half as likely to be injured during the incident as were those assaulted by assailants without weapons.

It should be noted that the model that included the self-protective behavior variable provided a better fit to the data (Chi-Squared/df ratio = 461/24) than did a model that did not include the self-protective behavior variable (Chi-Squared/df ratio = 382/23). This significant Chi-Squared difference of 79 (1 df), indicated that incorporating a variable assessing self-protective behavior into the model improved the prediction of injury risk.

We also conducted two parallel multivariate logistic regression models to determine whether self-protective behavior reduced the risk of both minor injuries (i.e., bruises, cuts, scratches, swellings, chipped teeth) and severe injuries (i.e., knife or stab wounds, gunshot wounds, broken bones or teeth, internal injuries, unconsciousness). Consistent with NCVS coding, for "unspecified injuries", women with hospital stays of 0-2 days were classified as incurring minor injuries, and those with hospital stays of greater than 2 days were classified as incurring severe injuries. A woman's use of self-protective behavior significantly reduced her risk of incurring a minor injury (adjusted odds ratio (AOR) = 0.42, 95 percent confidence interval (CI): 0.30, 0.57). Although the use of self-protective behaviors also reduced the odds that a woman incurred a significant injury, this was not statistically significant (AOR = 0.64, 95 percent CI: 0.29, 1.38).

In the final model for the self-protective-traditional measure, self-protective behaviors continued to manifest a positive, but statistically nonsignificant, effect on risk of injury. A similar pattern emerged for the other covariates. Relationship status, victim's education, and offender weapon use were the only significant risk factors for injury. The nature of their effects on injury risk were similar to their effects in the model with the self-protective-revised measure.

Subjective perceptions of helpfulness of self-protective behaviors

Women who engaged in self-protective behaviors were also asked whether they believed any of their actions helped the situation in any way and whether any of their actions made the situation worse in any way (women could indicate that the self-protective behavior helped the situation and that the self-protective behavior made the situation worse, so these variables were not mutually exclusive). Seventy-five percent of women who engaged in a self-protective behavior reported that their use of a self-protective behavior(s) helped the situation. Of these women, 57 percent reported that the behavior(s) was helpful in avoiding injury or greater injury, 21 percent reported that the self-protective behavior(s) helped scare or chase the offender off, 25 percent reported that the behavior(s) helped them get away from the offender, 7 percent reported that the behavior helped them protect other people, and 16 percent reported that the behavior(s) helped the situation in some other way. Conversely, only 21 percent of the women who engaged in some form of self-protective behavior reported that their behavior made the situation worse. Of these women, 15 percent reported that the behavior(s) led to injury or greater injury, 1 percent reported that the behavior(s) resulted in other people getting hurt, 86 percent reported that the behavior(s) made the offender angrier , and 6 percent reported that the action made the situation worse in some other way.

To determine which factors predicted whether a woman perceived her use of a self-protective behavior as helpful and which predicted whether she perceived her self-protective behavior as making the situation worse, we conducted two parallel logistic regression analyses. Victim-offender relationship was the only variable that predicted both the perception that the use of self-protective behavior(s) was helpful and the perception that the use of self-protective behavior(s) was harmful. Specifically, women assaulted by strangers, unrelated acquaintances, and casual acquaintances were more likely than those assaulted by intimates to perceive their use of self-protective behavior(s) as helping the situation (AOR = 2.14, 95 percent CI: 1.12, 4.09; AOR = 1.79, 95 percent CIL 1.03, 3.13; and AOR = 2.18, 95 percent CI: 1.23, 3.80, respectively). Conversely, women assaulted by strangers , unrelated acquaintances, and casual acquaintances were less likely than those assaulted by intimates to perceive their use of self-protective behavior(s) as making the situation worse (AOR = 0.36, 95 percent CI: 0.18, 0.72; AOR = 0.56, 95 percent CI: 0.31, 0.9; and AOR = 0.43, 95 percent CI: 0.24, 0.77, respectively). There was no significant difference between women assaulted by intimates and those assaulted by family members on their perception that these self-protective measures were helpful or harmful. Further, women who engaged in self-protective behavior(s) before or during the injury were more likely than those who used self-protective behavior only after the injury to perceive their behavior as helpful (75 vs. 68 percent). Conversely, women who engaged in self-protective behavior(s) before or during the injury were significantly less likely than those who used self-protective behavior only after being injured to perceive their use of self-protective behavior as making the situation worse (20 vs. 29 percent).

Table 2. Crude odds ratios and 95% confidence intervals for modeling physical injury by predictors in bivariate logistic regression equations, National Crime Victimization Survey, 1992-1995

¹ The weighting factor adjusts for unequal probabilities of selection and observation and for known age, sex, and race ratios based on the 1990 Adjusted Decennial Census Population totals (2).
² OR, odds ratio; CI, confidence interval.
³ Ninety-five percent confidence interval does not include 1.

Discussion

Results from this study indicate that self-protective behaviors (revised measure), victim-offender relationship status, education of the victim, and offender's weapon use are important to consider when predicting risk of injury. In physical assault situations, we found that women who engaged in self-protective behaviors were less likely to be injured from a physical assault than were those who did not engage in any self-protective behaviors or did so only after incurring an injury. This inverse association between self-protective behaviors and injury diverges from previous findings that have shown that women who engage in self-protective behaviors at any time in the incident are at increased risk of injury in physical assault situations (4), robbery situations (5), and sexual assault situations (5, 6). We are unaware of any other study that has examined self-protective behaviors and risk of injury after reclassifying women who used self-protective behaviors only after being injured into a "no self-protective behavior" group. However, several researchers have recommended using caution in interpreting findings derived from the preredesigned NCVS because data on temporal sequencing were not available (5-7). Our finding that self-protective behaviors are inversely associated with injury is bolstered by the responses of women in our sample when they were asked their opinion regarding the perceived helpfulness of their self-protective behavior; 75 percent reported that the measures helped the situation, whereas on 21 percent reported that the self-protective measures made the situation worse.

Our finding that women victimized by intimate partners are at increased risk of injury after a physical assault is congruent with prior studies. Extant research has consistently revealed that women victimized by intimates are significantly more likely to be injured than are women victimized by nonintimates (10, 12, 13).. Previous research with the NCVS has revealed that 52 percent of women victimized by an intimate partner or ex-partner sustained a physical injury, in contrast to 20 percent of women victimized by strangers (10, 12),

Table 3. Adjusted odds ratios and 95% confidence intervals for modeling physical injury by predictors in multivariate logistic regression equations, National Crime Victimization Survey, 1992-1995

¹ Controlling for vitim's age, marital status, education, race, urbanicity, and household income; for offender's age, race, and weapon use; and for victim/offender relationship status, presence of others, and of offender violence.
² OR, odds ratio; CI, confidence interval.
³ 95% CI does not include 1.

It is not clear why low education would serve as a risk factor for injury. It may be that women of lower education levels are exposed to more violence and, hence, have higher thresholds for interpreting events as physically assaultive. Consequently, the same event may be interpreted as an assault by a woman of higher education level but not by a woman of with a lower one. Thus, women in our sample with more than a high school education may be more likely to have reported minor assaults than would their less-educated counterparts, resulting in a spurious association between low education and injury risk. Conversely, our finding that low education is a risk factor for injury may reflect a true association. Additional research is needed to explain the mechanism through which education is related to injury risk.

There were several study limitations that should be acknowledged. Although the survey queries respondents about both forceful and nonforceful resistance, we did not examine these behaviors separately. Data from the redesigned NCVS do not allow researchers to examine different forms of self-protective behaviors separately while taking into account temporal sequencing. The reason for this is the way in which the questions were asked in the NCVS. Specifically, respondents were asked what behaviors the had engaged in, and the could mark all that applied. Respondents who were injured were then asked whether these actions were taken before, after, or at the same time as the injury. Again, they could mark all that applied. It is not possible to know to which self-protective action(s) a respondent was referring when answering the temporal sequencing questions. This limitation is unfortunate because prior studies have found that the type of self-protective behavior used (i.e., forceful, nonforceful) may differentially affect injury risk (4, 6, 14). For example, Marchbanks et al. (6) found that female victims of attempted or completed rape who engaged only in nonforceful resistance did not increase their risk of injury, but rape victims who used only forceful resistance or both forceful and nonforceful resistance were at increased risk of injury. On the other hand, research on female victims of physical assaults perpetrated by intimate partners has indicated that both types of resistance lead to increased risk of injury (4).

Not only did the use of a secondary data source limit our ability to examine how different forms of self-protective behavior affected injury risk, but it also limited our ability to have more information about the context of the assault. Although the NCVS surveys respondents about several victim and offender characteristics, there is little contextual information about the crime incident. For example, we do not know whether a woman had the opportunity to engage in self-protective behavior or whether she was attacked so quickly that she had no opportunity to protect herself However, this is the only nationally representative database that contains measures of violent victimizations, injuries, and self-protective behaviors.

Another study limitation is that our assessment injury severity was limited. Because the NCVS is a victim-based survey, no data are available on injuries resulting in death. Further, in terms of nonfatal injuries, the small number of women in our sample who experience severe injuries (n=92) may have limited the reliability of our finding regarding the role of self-protective behavior in reducing severe injuries. Some researchers have defined injury severity according to whether or not the victim sought medical care for her injuries. Marchbanks et al. (6) found that rape victims who used both forceful and nonforceful resistance were more likely to require medical treatment that were victims who did not use resistance, although neither form of self-protective behavior alone was associated with medical care. A limitation to using this measure for injury severity, however, is that other factors could influence a victim's decision to seek medical treatment. For example, victims of intimate partner violence may be particularly reluctant to seek care for fear of retaliation by the perpetrator or due to feelings of shame (4). In a study of women who were more physically assaulted, women attached by intimates where more likely to be injured (82 percent) than were those assaulted by strangers (50 percent), but victims of intimate violence were less likely to receive medical care (43 vs. 51 percent). Neither forceful nor nonforceful self-protective behaviors were associated with medical treatment among either group of victims (4).

Another study limitation is that our data do not inform us about the direction of causality. Although we found that self-protective behaviors were associated with a lower risk of injury, we do not know whether the self-protective action itself reduced the likelihood of injury. A victim's perceived risk if injury may influence her decision to engage in self-protective actions. Perhaps women in less-threatening circumstances are more willing to engage in self-protective behaviors, whereas women who fear that injury or death is imminent are more likely to cooperate with the offender. If so, this could explain the observed inverse association between injury and resistance. On a related note, we do not know if a third unmeasured variable might explain the observed association between self-protective behavior and injury. Although we controlled for several potential risk factors for injury, another unmeasured variable may account for the association we found. Finally, we do not know whether our finding that the use of self-protective behaviors decreases the risk of injury among women after physical assault would be replicated among women who have experienced other types of violent crime, such as sexual assaults and robberies.

The incorporation of the temporal sequencing of events has contributed significantly to our knowledge of the association between self-protective behaviors and risk of injury among female victims of physical assaults. The results from this study can provide some guidance regarding the coding of those injured victims who engaged in self-protective behavior only after being injured. We propose that self-protective behaviors that occur after injury could not influence that injury. In fact, it is plausible that these behaviors were influenced by the injury event rather than serving as a risk factor. Moreover, it appears that the positive association observed between self-protective behavior and risk of injury in previous studies is reversed when those who engaged in self-protective behaviors only after being injured are classified as not having use them. We suggest that when attempting to determine the association between use of self-protective behavior and injury risk, it is most accurate to consider only those behaviors that took place prior to or during the injury as indicative of self-protective acts. Those who engage in self-protective behavior only after being injured should be included in the analysis and classified as not having used self-protective actions.

De to inconsistent results across studies and the limitations of the available data, we suggest caution in making recommendations about the advantages or disadvantages of using self-protective behavior during a physical assault. While we believe that the primary prevention of assaults remains the most promising means of minimizing injuries from physical assaults and that future research is needed to determine which primary prevention strategies are most effective (15), we also believe that more research is needed to shed light on what women should do if they are assaulted. Future studies on the victimization of women need to include questions regarding injuries, the use of self-protective behaviors, the types of self-protective behaviors used, and the temporal sequencing of the self-protective behavior and the injury outcome. Optimally, the design of study questions should allow for analysis of forceful and nonforceful self-protective behaviors separately while taking into account temporal sequencing. Further, large samples are needed to ensure enough statistical power to examine whether self-protective behaviors reduce the risk of sever as well as minor injuries. In this way, research findings can be better utilized to inform women regarding injury prevention strategies to use during an assault.

Acknowledgements

The authors wish to thank Dr. Polly Marchbanks for her comments on a previous draft of this paper. They also thank Tracy James Mattingly with the US Census Bureau for her calculation of the design effects and advice on providing a rationale for its use.

References

1. Mercy JA, Resenberg ML, Powell KE, et al. Public health policy for preventing violence. Health Af (Millwood) 1993; 12:7-29
2. Bureau of Justice Statistics. Criminal victimization in the United States, 1994. Washington, DC: US Department of Justice, 1997. (Publication no. NCJ-162126).
3. Grisso JA, Schwartz DF, Miles CG, et al. Injuries among inner-city minority women: a population-based lingitudinal study. Am J Public Health 1996;86:67-70.
4. Bachman R, Carmody DC. Fighting fire with fire: the effects of victim resistance intimate versus stranger perpetrated assaults against females. Fam Violence 1994;9:317-31.
5. Block R, Skogan WG. Resistance and nonfatal outcomes in stranger-to-stranger predatory crime. Violence Vict 1986;1:241-53.
6. Marchbanks PA, Lui KJ, Mercy JA. Risk of injury from resisting rape. Am J Epidemiol 1990;132:540-9.
7. Cook P. The relationship between victim resistance and injury in noncommercial robbery. Legal Stud 1986;15:405-16.
8. Ullman SE, Knight RA. Fighting back: women's resistance to rape. J Interpersonal Violence 1992;7:31-43.
9. Sorenson SB, Upchurch DM, Shen H. Violence and injury in marital arguments: risk patterns and gender differences. Am J Public Health 1996;86:35-40.
10. Bachman R, Saltzman L. Violence against women: estimates from the redesigned survey. Bureau of Justice Statistics Special Report. Washington DC: US Department of Justice, August 1995. (Publication no. NCJ-154348).
11. Kindermann C, Lynch J, Cantor D. Effects of the redesign on victimization estimates. Bureau of Justice Statistics National Crime Victimization Survey. Washington DC: US Department of Justice, April 1997. (Publication no. NCJ-164381).
12. Craven D. Female victims of violent crime. Burear of Justice Statistics Selected Findings. Washington DC: US Department of Justice, December 1996. (Publication no. NCJ-162602).
13. Rand M, Strom K. Violence-related injuries treated in hospital emergency departments. Bureau of Justice Statistics Special Report. Washington DC: US Department of Justice, August 1997. (Publication no. NCJ-156921).
14. Bachman R, Saltzman L. The consequences of victim resistance during a violent crime victimization: Do the costs outweigh the benefits? Presented at the American Society of Criminology, Chicago, Illinois, November 1996.
15. Chalk R, King P, eds. Violence in families: assessing prevention and treatment programs. Washington DC: National Academy Press, 1998.