Statistics with ANOVA

I n 2002 the World Health Organization released its World Report on Violence and Health, in which former South African president Nelson Mandela wrote of violence against women as one example of a legacy of suffering, “a legacy that reproduces itself, as new gener- ations learn from the violence of generations past. . . . But [we are

not] powerless against it.”1

Each year, 1 million women in the United States report being physically abused by their intimate partners; half of these victims report a physical injury.2 So great is the problem that Healthy People 2010, the report of the U.S. Department of Health and Human Services, specified a 20% reduction in the occurrence of intimate-partner abuse as a national health objective to be achieved by 2010.3

Nursing and other professional organizations acknowledge the health consequences of intimate-partner violence and the need for routine assess-

Judith McFarlane holds the Parry Chair in Health Promotion and Disease Prevention at the Texas Woman’s University College of Nursing in Houston, where Ann Malecha and Julia Gist are assistant professors. Iva Hall is undergraduate coordinator and Sheila Smith is an instructor in the Department of Nursing at Lamar University in Beaumont, TX. Kathy Watson is a statistician at Baylor College of Medicine and Elizabeth Batten is a bilingual caseworker at the Harris County District Attorney’s Office, both in Houston. Contact author: [email protected].

This project was supported by Grant No. 2000-WT-VX-0020, awarded by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice. Points of view in this document are those of the authors and do not necessarily represent the official position or policies of the U.S. Department of Justice. The authors wish to thank the Family Criminal Law Division of the Harris County District Attorney’s Office for assistance in the collection of data; they also acknowledge the 149 women who maintained contact with the investigators for 18 months. The authors have no sig- nificant ties, financial or otherwise, to any company that might have an interest in the publication of this educational activity.

Increasing the Safety-Promoting Behaviors

of Abused Women

Increasing the Safety-Promoting Behaviors

of Abused Women In this study, a telephone intervention for

victims of intimate-partner violence showed

efficacy for 18 months.

OVERVIEW Despite an epidemic of intimate- partner violence against women, and general agreement that women should be screened for it, few assessment and intervention proto- cols have been evaluated in controlled stud- ies. To test a telephone intervention intended to increase the “safety-promoting behavior” of abused women, 75 women received six tele- phone calls over a period of eight weeks in which safety-promoting behaviors were dis- cussed. A control group of 75 women re- ceived usual care. Women in both groups received follow-up calls to assess safety- promoting behaviors at three, six, 12, and 18 months after intake. Analysis showed that the women in the intervention group prac- ticed significantly (P < 0.01) more safety- promoting behaviors than women in the con- trol group at each assessment. On average, women in the intervention group practiced almost two more safety-promoting behaviors than they had at time of intake and nearly two more than women in the control group; the additional behaviors were practiced for 18 months. This nursing intervention requires only 54 minutes to complete (six nine-minute telephone calls) and can be integrated into any health care setting. Because less than one hour of professional nursing time is involved, the cost of the intervention is mini- mal. Future research should determine whether the adoption of safety-promoting behaviors by abused women averts trauma and its subsequent health care costs.

Key words: abuse; intimate-partner violence; safety intervention; clinical trial; women’s health

By Judith McFarlane, DrPH, RN, FAAN, Ann Malecha, PhD, RN, Julia Gist, PhD, RN, Kathy Watson, MS, Elizabeth Batten, BA, Iva Hall, PhD, RN, and Sheila Smith, PhD, RN,C

CE2.5Continuing Education HOURS

40 AJN ▼ March 2004 ▼ Vol. 104, No. 3 http://www.nursingcenter.com

ORIGINAL RESEARCH

ment of all women.4-8 There are several published protocols, specific to health care settings, for the identification of and assessment for such violence; prevention protocols have been published as well.9, 10

But there has been little evaluation of these proto- cols in controlled studies. Although a vast amount of research into intimate-partner abuse documents the type and extent of victims’ health-related prob- lems—including acute trauma, chronic pain and other conditions,11, 12 and increased use of health care services13—few interventions have been tested.14, 15

We recently completed a clinical trial that tested a nursing intervention designed to increase the num- ber of “safety-promoting behaviors” practiced by abused women—that is, actions they take to pro- mote their own safety. The first published report based on this clinical trial evaluated the interven- tion’s effectiveness at three and six months after the intervention was completed.16 In this report we evaluate the intervention’s effectiveness at 12 and 18 months. Two conceptual models guided this 18- month clinical trial: Walker’s three-phase “cycle of violence”17-19 and Curnow’s “open window phase” of help-seeking and self-help behaviors, which builds on Walker’s theory.20

CONCEPTUAL MODELS Walker’s cycle of violence delineates three distinct phases of abuse. In phase one, the tension-building phase, the abuser becomes moody, hostile, and crit- ical of the woman, who usually attempts to calm the abuser by becoming nurturing or compliant or staying out of his way. In phase two, the acute, violent-incident phase, the abuser physically and psychologically assaults the woman. The victim usually feels extremely frightened or threatened. In the calm, or “honeymoon,” phase, which occurs shortly after the assault, the abuser typically ex- presses sorrow for his actions; behaves in a loving, charming, or contrite manner; and promises that the violent behavior will not be repeated. The calm phase gives the woman hope that her partner’s behav- ior is going to change. According to Walker, char- acteristics commonly seen in the abused woman throughout much of this cycle include denial of the partner’s abuse and of the extent of her own injuries, as well as a belief that there are no alterna- tives. Walker describes a dramatic change, however, in the abused woman from the end of phase two, the violent period, into phase three. During this transition, the abused woman may assess her situa- tion realistically, acknowledge her inability to con- trol or stop the abuser’s violence, and express a desire to stop being a victim.

Curnow’s open window phase confirms this transition, during which avoidant and dependent behaviors and denial are absent and victims seek

help. Curnow wrote that in this period of clarity and seeking help—which she called the “open win- dow phase”—an abused woman • realizes she is a victim and is not able to stop the

violence. • is most likely to reach out for help. • will learn whether there are alternatives to

violence. • is most receptive to intervention.

Abused women most often seek help from the jus- tice system—through protection orders and from law enforcement personnel.2, 12 Many women prefer to obtain a protection order because it does not involve the filing of criminal charges, arrest, jail, and the posting of bond. (See Court Orders of Pro- tection, above.) Making police contact or applying for a protection order usually occurs immediately after or within 48 hours of a violent incident—in

[email protected] AJN ▼ March 2004 ▼ Vol. 104, No. 3 41

Court Orders of Protection More information on the process.

Aprotection order is a court order that restricts theaccess of one person (usually a male abuser) to another (usually an abused woman), as well as her property, children, family, or live-in friends for a speci- fied period. The term restraining order is used in some jurisdictions.

Protection orders, both temporary and permanent, publicly document that abuse has occurred; if such an order is violated, the suspect is subject to criminal prose- cution. A civil protection order offers the abused person legal action when she doesn’t want the abuser charged criminally or jailed for an offense. But choosing this action does not preclude other civil or criminal action.

Unlike arrest proceedings, the purpose of a civil pro- tection order isn’t to punish past conduct, but to prevent future harm. To make protection orders more accessible and enforceable, the 1994 Violence Against Women Act (VAWA) provided that no filing fees or service costs be charged to obtain protection orders; and in 2000 a Congressional revision of the 1994 VAWA included a full faith and credit provision, which means that law enforcement protection is assured in any state, regard- less of where the protection order was issued, including tribal territories.21 Additionally, the VAWA revision estab- lished penalties for persons who cross state lines to abuse an intimate partner, making interstate domestic abuse and harassment a federal offense.

The criteria necessary to obtain a protection order are mandated by individual state statutes, which vary from state to state. The criteria usually include the applicant’s having been a victim of intimate-partner violence and a likelihood of such violence in the future.

other words, during the transition from phase two to phase three of Walker’s cycle of violence and in Curnow’s open window phase. We hypothesized that abused women who contacted the justice system and underwent a telephone counseling intervention aimed at increasing safety-promoting behavior would report adopting such behaviors at 12 and 18 months more often than would abused women who received usual care.

METHODS Design. This study used a two-group, repeated- measures design. (For details of the study design, see More on Methods, page 43.)

Setting. The study was conducted at a special family-violence unit of a large urban district attor- ney’s office that serves an ethnically diverse popula- tion of 3 million. The family-violence unit processes civil protection orders for abused women and offers

42 AJN ▼ March 2004 ▼ Vol. 104, No. 3 http://www.nursingcenter.com

SAFETY-PROMOTING BEHAVIOR CHECKLIST Ask the woman to answer “Yes,” “No,” or “Not Applicable.”

Have you ever . . .* ¿En alguna occasión usted ha . . .* Yes Si No NA

. . . hidden money?

. . . escondido dinero?

. . . hidden an extra set of house and/or car keys?

. . . escondido un juego extra de las llaves a la casa y/o al coche?

. . . established a code with family or friends?

. . . establecido un código para usar con su familia o con sus amigos?

. . . asked neighbors to call police if violence begins?

. . . pedido a sus vecinos que llamaran a la policia si empiece la violencia?

. . . removed weapons?

. . . quitado armas?

Have you ever had available . . . ¿En alguna occasión usted ha tenido a su disposición . . .

. . . Social Security number (yours, his, children)?

. . . el número de Seguro Social (de usted, del abusador, de sus hijos)?

. . . rent and utility receipts?

. . . los recibos de la renta o de la luz, el agua, y el gas?

. . . birth certificates (yours and children)?

. . . los actas de nacimiento (de usted y de sus hijos)?

. . . ID or driver’s license (yours and children)?

. . . el ID o la licencia para manejar (de usted y de sus hijos)?

. . . bank account numbers?

. . . los números de las cuentas bancarias?

. . . insurance policies and numbers?

. . . los números y pólizas de aseguranza?

. . . marriage license?

. . . el acta de matrimonio?

. . . valuable jewelry?

. . . unas joyas de valor?

. . . important phone numbers?

. . . los números de teléfono importantes?

. . . hidden bag with extra clothing?

. . . una bolsa escondida con ropa extra?

*After the first visit, change the opening phrase to: “Since the last time we talked, have you . . .” (“¿Desde la ultima vez que completamos esta cuestionario, usted ha . . . ”)

counseling on domestic violence and referrals to each applicant. All applicants are given the name and phone number of their assigned intake case- worker at the district attorney’s office and encour- aged to telephone the caseworker for further assistance. There are no fees for the services to the applicant. The office is open from 8 AM to 5 PM, Monday through Friday. Appointments are not taken, and everyone is assisted on a first-come, first- served basis.

Sample. All English- or Spanish-speaking women who applied and qualified for a protection order against an intimate partner were invited into the study by one of the six investigators (five of the investigators were RNs; the sixth was a caseworker at the district attorney’s office). An investigator was present at the unit each day. We systematically allo- cated women to either the treatment group or the control group, depending on the week of enrollment in the study (systematic allocation to groups by week of enrollment is unlikely to introduce any sampling bias.) Sampling with this method contin- ued for 28 consecutive business days, until there were 75 women in the control group and 75 in the treatment group. (A total of 154 women qualified for the study and were invited to participate; four women refused.) One woman committed suicide three weeks into the study. All of the remaining 149 women completed the three-, six-, 12-, and 18- month follow-up interviews, for a study retention rate of 99% (see How Is Sample Size Determined? page 44).

Instruments. At intake, we completed a demo- graphic-data form for each participant, in order to document age, education, self-identified race or eth- nicity, and relationship to the abuser. This form also included the names, addresses, and phone numbers of “safe contacts”—family members, friends, neigh- bors, and employers—which would help us com- plete follow-up interviews.

We used a questionnaire—a 15-item checklist— to determine each woman’s experience with taking actions that could protect her in the future (see Safety-Promoting Behavior Checklist, page 42). Read aloud by the investigator, all questions began with the phrase, “Have you ever . . .” followed by specific safety-promoting behaviors. In subsequent telephone calls, investigators asked the same series of questions, with the opening phrase modified: “Since the last time we talked, have you . . . ?” (This checklist was first described in “Abuse During Pregnancy: A Protocol for Prevention and Inter- vention”24 and has been used to help pregnant women who are abused.22, 25)

To ensure that the Spanish version of the demo- graphic-data form and the questionnaire matched the English, forward-and-backward translation was done. That is, one translator translated the instru-

ments from English into Spanish, and then another translated the Spanish version back into English; the original and newly translated English versions were then compared.

Not all 15 items on the checklist were applicable to each woman (for instance, an unmarried woman wouldn’t have a marriage license); therefore, the scoring of the checklist was adjusted for purposes of interpretation and comparison (the adjusted total fell within a 0-to-15 range of behaviors performed). The following equation expresses the relationship of the number of applicable behaviors performed to the adjusted total number of behaviors: a/b = x/15, where a is the number of behaviors performed, b is

[email protected] AJN ▼ March 2004 ▼ Vol. 104, No. 3 43

More on Methods Determining sample size and analyzing results.

The sample size was determined using a poweranalysis (for more on power analysis, see How Is Sample Size Determined?, page 44). Based on a previous study we conducted that measured the effec- tiveness of a safety-promoting intervention on a group of pregnant women who were abused,22 which found a moderate treatment effect, we pre- dicted a similar effect size in the current study. To detect a moderate effect size of 0.45 with 80% power (that is, a significant treatment effect would be demonstrated in eight out of 10 repetitions of the study), 60 women were needed in each group.23 To allow for 25% attrition, 75 women were recruited for each group. Because there is little research in this area and a moderate treatment effect was predicted, the power calculations were purposely conservative.

Descriptive statistics (means, standard deviations, frequencies, and percentages) were calculated for all study variables. In the analysis of the data from this study, means, standard deviations, frequencies, and percentages were used to describe the women. Differ- ences in demographic characteristics among the inter- vention and control groups were investigated using independent t tests and chi-square tests of independence. Characteristics exhibiting significant group differences, such as age, were included in subsequent analyses. To determine whether the adjusted safety-promoting behavior totals changed over time, data were ana- lyzed using a repeated-measures analysis of covari- ance with one between-groups factor (that is, in which group the subject was enrolled, control or intervention). Statistical assumptions, such as normality of the data and homogeneity of variance, were tested and met. To control for inflated type I error resulting from multiple testing and maintain the experiment-wise error rate of 0.05, Bonferroni’s method of adjustment was used and the α level was spread over the five tests conducted, for a significance level of 0.01 (0.05/5 = 0.01).

the total number of applicable behaviors, and x is the adjusted total. If a and b are known, the ad- justed total number of safety-promoting behaviors performed can be calculated by solving for x (by cross-multiplying the two fractions).

For example, if a woman was single and didn’t have an insurance policy, a bank account, or a weapon in her home (thus making her answer on four questions “NA”) and practiced all but one applicable safety-promoting behavior, the number of applicable behaviors performed would be 10 out of 11 (10/11); cross-multiplying 10/11 and x/15 (or dividing 150 by 11) yields 13.63, the adjusted

total number of behaviors performed. In compari- son, consider a second woman who also performed 10 safety-promoting behaviors, but to whom all but one behavior was applicable. Her adjusted total number of behaviors performed would be calcu- lated by cross-multiplying 10/14 and x/15 (150 divided by 14), which yields 10.7. Higher adjusted scores indicated a higher number of safety-promoting behaviors performed.

Procedures. Data collection began after we received approval from the institutional human sub- jects review board and consent of the district attor- ney’s office. At the district attorney’s office, an

44 AJN ▼ March 2004 ▼ Vol. 104, No. 3 http://www.nursingcenter.com

How does a researcher know how large thesample needs to be or how many subjects to include in a study to ensure a statistically signifi- cant result? This is an important question. Having too few subjects can lead to false conclusions, particularly if no significant differences are found between the control and intervention (experimen- tal) groups. Proper estimation of sample size is crucial to the success of a study.

McFarlane and colleagues’ study found that the differences between the control and intervention groups in scores on the safety-promoting behavior checklist were statistically significant—meaning that it’s unlikely that the differences between the groups were chance findings. However, suppose for a moment that these differences were shown not to be statistically significant: the researchers would con- clude the intervention wasn’t effective. But would that be the correct conclusion? Perhaps they simply didn’t study enough women to be able to detect real differences. (Concluding that no difference exists when one really does exist is called a type II error. Such an error can result in the rejection of interven- tions that really are effective—all because the researchers didn’t include enough subjects to demonstrate a real effect.)

The ability to detect an effect, if one exists, is called statistical power. To make sure a study has sufficient statistical power, researchers conduct a power analysis, which determines how many sub- jects must be included in the study to find an effect of treatment, if one exists (see More on Methods, page 43). To conduct a power analysis, researchers first must speculate on how large the effect of the intervention on the outcome variable will be (in this case, how much of an effect the telephone- counseling intervention will have on women’s safety-promoting behaviors).

The magnitude of the effect of treatment is

How Is Sample Size Determined? Statistical power, power analysis, and effect size.

called effect size. By convention, an effect size of 0.2 indicates a small effect; 0.5, a moderate effect; and 0.8 or greater, a large effect. Researchers’ estimates of effect size are usually based on what prior research suggests the impact of the intervention will be. McFarlane and col- leagues were purposely conservative in their power analysis because there was little research on interventions such as theirs. However, their own previous study that measured the effectiveness of a safety intervention on a group of pregnant women who were abused, which found a moderate treat- ment effect, served as a guide in predicting a simi- lar effect size in this study. Therefore, they speculated that the telephone-counseling interven- tion would have a moderate effect (0.45) on the intervention group’s safety-promoting behaviors. This determined the study’s sample size of 75 par- ticipants. If the researchers had anticipated a small effect, meaning that the intervention would have been less effective, they would have needed more subjects because the effect of the intervention would have been harder to detect; had they pro- jected a large effect of the intervention, they would have needed fewer subjects because the effect of the intervention would have been easier to detect.

Why not always anticipate a small effect size, which means more subjects will be included, to ensure that a real difference between the control and intervention groups won’t be missed? Unneces- sarily large samples cost more and waste time and effort collecting unneeded data. There can also be ethical concerns about exposing more people than necessary to untested, potentially harmful interven- tions or withholding what proves to be effective interventions from control groups. These points under- score the importance of sample size estimation or power analysis when designing a clinical study. —Diana J. Mason, PhD, RN, FAAN, editor-in-chief

investigator escorted each woman invited to partic- ipate in the study to a private room and explained the study’s purpose, potential risks and benefits, protocol, instruments, administration time, and fol- low-up schedules. Each woman who agreed to par- ticipate signed an informed-consent form, and the investigator administered the two study instru- ments (demographic-data form and checklist), offering a choice of English or Spanish. Because all participants received telephone calls—those in the intervention group received six intervention calls and four follow-up calls, and those in the control group received four follow-up calls—the safety of all was ensured by establishing times for the calls that would be convenient and private.

Women in the control group received the usual services of the family-violence unit of the district attorney’s office, which included counseling on pro- moting safety and information on social services and legal resources. All applicants received infor- mation on protection orders and the court proceed- ings involved. The caseworker provided a session of counseling and referrals, which took about an hour. Depending on the applicant’s needs, the caseworker also performed various follow-up activities (for example, phone calls to the applicant and to social service agencies). An appointment for the date and time of the first follow-up data-collection telephone interview, which would occur in three months, was also made for each woman in the control group.

Women in the intervention group received the usual services of the district attorney’s office and were also given a day and time for the first inter- vention telephone call.

The six intervention calls. The protocol used is part of McFarlane and Parker’s abuse-prevention protocol.9, 24 Women who were assigned to the inter- vention group received six “safety intervention” telephone calls. The investigator who enrolled the participant in the study completed all six interven- tion calls as well as the four data-collection follow- up calls (at three, six, 12, and 18 months after intake). The first intervention call occurred between 48 and 72 hours of intake; the remaining were made at one, two, three, five, and eight weeks afterward. Each call began with the safety-promoting behavior checklist, noting behaviors adopted since the previ- ous call. During the intervention calls, the investiga- tor discussed specific safety-promoting behaviors with the participant, suggesting, for example, that she make extra keys, apply for a driver’s license, and obtain copies of documents such as a Social Security card and marriage license. The investigator would also suggest hiding money and documents—in an empty tampon container, for example—or entrust- ing them to a friend, neighbor, or relative. Discussion also involved creating a code for use with family and friends as a signal for the need for

help, removing weapons from the home, and enlist- ing neighbors to call the police if they overhear an altercation. The safety-intervention telephone calls lasted from three to 25 minutes, with a mean of nine minutes per call. The safety intervention ended with the sixth telephone call, eight weeks after entry into the study.

The four follow-up calls. At three, six, 12, and 18 months after intake, investigators made data- collection calls to all participants; they assessed whether the women were still performing the safety- promoting behaviors they’d performed previously and whether they had adopted others. The investi- gators asked women in both the intervention and the control groups the questions on the checklist during each follow-up call, but no further informa- tion on safety was provided. If any participant wanted to discuss the abuse or sought information, she was referred to her intake caseworker at the dis- trict attorney’s office.

RESULTS Demographic characteristics of the intervention and control groups appear in Table 1 (page 47). The mean age of women in the intervention group, 30.3 years, was significantly lower (t148 = 2.964, P = 0.003)—almost five years lower—than the mean age of the control group, 34.6 years. For this reason, age was treated as a covariate in further analysis. It was not found to affect the outcome measure of the study, and therefore, only unad- justed results from a repeated measures analysis of variance (RM ANOVA) are reported. There were no other significant demographic differences between the two groups.

Safety-promoting behaviors among all partici- pants. Means and standard deviations for the adjusted total number of safety-promoting behaviors performed by the women in the intervention and

[email protected] AJN ▼ March 2004 ▼ Vol. 104, No. 3 45

Safety-promoting behaviors

include enlisting neighbors

to call police if they overhear

an altercation.

McFarlane and colleagues were interested inincreasing, through the use of a specific inter- vention, the average number of safety-promoting behaviors performed by a cohort of abused women. There were two groups of study subjects, those who received the intervention and those who didn’t (the control subjects). There was little vari- ability among the groups; for example, all were women, all had been abused, all had sought help at the district attorney’s office. The application of the intervention may theoretically be viewed as imposed variation. Should it be effective, the researchers would expect to see a sizable treat- ment effect. In this case, there was indeed a signif- icant increase in the average number of safety- promoting behaviors performed in the intervention group, as compared with the control group.

The inherent variability in any sample of study participants is one of the greater challenges for the investigator (and for the statistician). Consider how blood pressure can vary in a group of sub- jects, even when they’re of comparable age and health status. The researcher has to figure out which differences are reflective of the natural vari- ation of variables (such as sex and socioeconomic or smoking status) and which are attributable to the study intervention. It’s important to understand that differences among study subjects may be attributed to both within-group variation and between-group variation—in other words, there may be variation within either the control group or the study group that needs to be examined. There are a number of statistical tests that can help tease out the sources of variation. Analysis of variance (ANOVA) is one of them.

ANOVA and the paired t test. ANOVA is essen- tially a direct extension of the paired t test, which determines whether a difference between two study groups is greater than the variability within the groups. To use a paired t test, you need two groups of subjects and two points in time for assessment. Paired t tests are frequently used for studies with before-and-after (or pretest–posttest) designs.1 If the intervention is effective, you would expect that changes among subjects in the treatment group, from baseline to the final measurement, would be of greater magnitude than those observed in the con- trol group. The paired t offers some real advantages in statistical power. Paired observations of the same individual before and after an intervention cuts down on the number of subjects one needs to study. Measuring the same subject at two points in time

ANOVA Concept Oh, those variable variables.

reduces within-group variance, as each subject serves as his own control.

While a paired t test can only be conducted using two groups of subjects, ANOVA may be applied to two or more groups. In addition, ANOVA enables the investigator to study factors other than the intervention that might be influential.

RM ANOVA and RM AN(C)OVA. ANOVA can assess an outcome at a single point in time or over a series of points in time. The repeated-measures (RM) aspect of the McFarlane study design adds a level of sophistication to the understanding of changes in safety-promoting behaviors in this cohort of women over time. If one perceives learning and behavior to be dynamic processes (in other words, changing over time), it makes sense to observe a skill or behavior at more than one point in time. In this study, subjects in both the intervention and control groups were assessed at five distinct time points. Investigators were able to demonstrate that while the numbers of safety-promoting behaviors increased over time in both groups of women, they increased most dramatically (and significantly) in the intervention group.

Sometimes two groups in a study aren’t per- fectly matched; in the McFarlane study, for instance, the mean baseline ages in the two groups were different. In such cases, an analysis of covariance (AN[C]OVA) is used to adjust for the difference. AN(C)OVA can be viewed as a “special circumstance” ANOVA. The differing baseline variable (in this case, age) is treated as a “covariate,” which, if not adjusted for, could confound the results. Using the AN(C)OVA, you can determine the impact of age on behaviors while still also assessing the impact of the inter- vention (because the results of this analysis showed that age did not have a significant impact on the effectiveness of the intervention, only the RM ANOVA results are reported). These analyses enabled the investigators to determine the influence of the intervention, the influence of time, and the interaction between these two vari- ables on the practice of safety-promoting behav- iors.—Ilene Wilets, assistant professor at Mount Sinai School of Medicine, New York City, and research subject advocate at its General Clinical Research Center

REFERENCE 1. Campbell DT, Stanley JC. Experimental and quasi-experimental

designs for research. Boston: Houghton Mifflin Company, 1963.

46 AJN ▼ March 2004 ▼ Vol. 104, No. 3 http://www.nursingcenter.com

control groups at intake, and at three, six, 12, and 18 months after intake, are presented in Figure 1 (page 48).

The RM ANOVA results showed • a significant main effect for group—indicating

that the number of safety-promoting behaviors performed increased significantly in the interven- tion group (F1,147 = 23.72, P < 0.001).

• a significant main effect over time—meaning that the number of safety-promoting behaviors per- formed increased significantly in the intervention group over time (F4,144 = 5.45, P < 0.001).

• a significant group-by-time interaction—mean- ing that the changes in the numbers of safety- promoting behaviors performed over time between the two groups were significantly differ- ent (F4,144 = 2.81, P = 0.028). To determine where the significant differences

across the five time periods (group-by-time interac- tion) were, follow-up analyses were conducted. One involved t tests of the simple effects between the intervention and control groups at each time period (intake and three, six, 12, and 18 months after in- take) as well as an RM ANOVA of differences within each group (intervention and control) over time. In other words, we first looked at the differ- ences between the intervention and control groups

at each testing time, then looked at the control group’s changes across all time periods and the intervention group’s changes across all time periods.

Tests of simple effects between groups showed significant (P < 0.01) differences between the inter- vention and control group scores at three, six, 12, and 18 months. Results from the RM ANOVA within each group showed that only in the interven- tion group were the differences significant (P < 0.01) over time—their intake scores were significantly lower than subsequent scores determined at three, six, 12, and 18 months after intake.

Effect size is an estimate of the strength of an intervention—in this case, a measure of the inter- vention’s ability to affect the women’s safety- promoting behavior: the higher the number, the stronger the effect of the intervention. Our find- ings indicate that the average number of safety- promoting behaviors women in the intervention group practiced increased by two from intake to three months and that an average increase of nearly two was sustained over 18 months. The effect size demonstrates the differences between the interven- tion and control groups. The effect size was large at three months (0.91) and remained moderate at 12 months (0.50) and 18 months (0.56). (See Figure 1, page 48.)

[email protected] AJN ▼ March 2004 ▼ Vol. 104, No. 3 47

Table 1. Characteristics of Women Studied Characteristic Intervention group Control group

(n = 75) (n = 75)

Age in years (±SD)* 30.25 (±7.87) 34.61 (±9.91) Education in years (±SD) 11.35 (±2.99) 12.20 (±2.55)

Race or ethnicity, n (%)

African American 23 (30.7) 26 (34.7) White 19 (25.3) 21 (28) Latino 33 (44) 28 (37.3)

Relationship to abuser, n (%)

Spouse or common-law spouse 40 (53.3) 41 (54.7) Ex-spouse or ex–common-law spouse 14 (18.7) 11 (14.7) Girlfriend 5 (6.7) 7 (9.3) Ex-girlfriend 16 (21.3) 16 (21.3)

Language spoken, n (%)

English 62 (82.7) 64 (85.3) Spanish 13 (17.3) 11 (14.7)

Note: One subject from the control group died; therefore, analysis was performed on 149 subjects.

*t148 = 2.98, P = 0.003

DISCUSSION Relying on Curnow’s open window theory, we tested whether a safety-promoting intervention offered to abused women when they were seeking help from the justice system would be effective in significantly increasing and maintaining the number of safety- promoting behaviors they perform. The intervention was effective. Adoption of safety-promoting behav- iors significantly increased over time among women in the intervention group, and they continued to practice these behaviors for 18 months. On average, women in the intervention group reported practicing almost two additional safety-promoting behaviors for 18 months after intake.

Each safety-promoting behavior can require great effort to perform and entail significant risk. For example, removing weapons from the home and

hiding a bag containing clothing and essential med- ications and documents can be dangerous. And in order to copy a house or car key, an abused woman has to obtain the key (many abusers keep house and car keys on a ring attached to a belt worn at all times); locate a place to duplicate the key and, often, pay for transportation to get there (several women had to either learn bus routes or find someone they could confide in and request help with transporta- tion from); and then return the key to its original location—all without the abuser’s knowledge. The women were eager to share stories of their success; one woman told of confiding in a neighbor who agreed to phone the police if the abused woman closed a kitchen curtain that’s usually open.

The fact that the use of safety-promoting behav- iors seemed to level off at six months and remained

48 AJN ▼ March 2004 ▼ Vol. 104, No. 3 http://www.nursingcenter.com

Figure 1. Adoption of Safety-Promoting Behaviors over 18 Months

Control group Intervention group

M ea

n- A

dj us

te d

To ta

l N um

be rs

o f

Sa fe

ty -P

ro m

ot in

g Be

ha vi

or s

Pr ac

tic ed

( ± SD

)

9.6 (±3.1)

9.9 (±2.8)

10.4 (±2.2) 10.6 (±2.5) 10.5 (±2.6)

10.4 (±2.8)

12.5 (±2.9)

12.0 (±2.5) 11.9 (±2.7) 12.0 (±2.7)

Intervention group, n = 75; control group, n = 75. One subject from the control group died; therefore, analysis was performed on 149 subjects.

the same at 12 and 18 months indicates a need for a “booster” intervention. We didn’t ask the women why additional safety-promoting behaviors weren’t adopted, although many reported feeling safer after going to the authorities and wanted to forget the past and start over. Some women who had moved to new residences said that they decided not to inform their new neighbors about past violence or ask for help in the event of an altercation. It’s important to note that the increase in safety-promoting behaviors seen in women in the control group may be attribut- able to their exposure to the safety-promoting behav- iors checklist during the repeated-measurement process, which may have had an effect similar to that of intervention.

Future research must test intervention models designed to maintain the use of safety-promoting behaviors by abused women. How long does the open window stay open? If a woman discloses abuse during routine assessment in a health care set- ting and this intervention is offered, can a similarly large effect be expected? We believe that the inter- vention can be effective in such settings when a vio- lent episode has occurred recently. But it’s not known how long after an episode of abuse the results found in this study can be expected to be duplicated. Replication of this research in a variety of clinical settings (prenatal, emergency, primary, and long-term care, for example) is urgently needed to measure the effectiveness of this intervention; effectiveness of the intervention relative to the most recent violent episode must also be studied. Future research must also consider correlates such as rela- tionship status (current versus former partner); type, frequency, and severity of violence; and previous help-seeking actions taken. Future research should also determine whether the adoption of safety- promoting behaviors by abused women averts trauma and its subsequent health care costs.

The high retention rate in the study is attribut- able primarily to the systematic use of the contact information the women supplied at intake. Inves- tigators telephoned alternate contacts when a woman’s telephone number or address had changed. Additionally, if all telephone and written contacts with the women failed, the researchers used the field community tracking strategies out- lined by Block and colleagues,26 as well as the recently published report by the National Institutes of Justice on recruitment and retention strategies in research of intimate partner violence.27 Using these published strategies, we were able to contact each woman for each of six intervention calls and four follow-up calls.

To our knowledge, this is the first clinical trial of a safety-promoting intervention for abused women. These findings clearly demonstrate that an interven- tion to increase the number of safety-promoting

behaviors practiced by abused women is highly effective when offered after an abusive incident has taken place. The effectiveness of the intervention remains substantial for 18 months. The average total time required to apply the intervention was only 54 minutes (six nine-minute phone calls), less than one hour of nursing time. The low intensity of this intervention means that it could feasibly be inte- grated into a variety of urban and rural health care settings. Additionally, the cost of the intervention is minimal and the impact potentially great.

Because evidence-based nursing practice enables nurses to track outcomes consistently and determine whether specific nursing care is effective, nurses and other health care providers must go beyond assess- ment and documentation of abuse to the applica- tion of tested intervention protocols that can interrupt ongoing violence, prevent future trauma, and promote the safety and health of women. ▼

REFERENCES 1. World Health Organization. World report on violence and

health. Geneva: World Health Organization; 2002. p. ix. 2. Rennison C, Welchans S. Intimate partner violence.

Washington, DC: Office of Justice Programs. Bureau of Justice Statistics. U.S. Department of Justice; 2000. http://www.ojp.usdoj.gov/bjs/pub/pdf/ipv.pdf.

3. U.S. Department of Health and Human Services. 15. Injury and Violence Prevention. In: Healthy people 2010. Conference Edition. Washington, DC: U.S. Deptartment of Health and Human Services: For sale by the U.S. G.P.O. Supt. of Docs.; 2000. p. 15-34. http://purl.access.gpo.gov/ GPO/LPS8595.

4. American Nurses Association. Position paper on violence against women. [Web site]. 2000. http://www.nursingworld. org/readroom/position/social/viowomen.pdf.

5. Family violence: an AAFP white paper. The AAFP Commission on Special Issues and Clinical Interests. Am Fam Physician 1994;50(8):1636-40, 44-6.

6. Violence against women. Relevance for medical practition- ers. Council on Scientific Affairs, American Medical Association. JAMA 1992;267(23):3184-9.

7. American College of Obstetricians and Gynecologists. Domestic violence. ACOG Technical Bulletin 209. Washington DC: The College; 1995.

8. Paluzzi PA, Houde-Quimby C. Domestic violence. Implications for the American College of Nurse-Midwives and its members. J Nurse Midwifery 1996;41(6):430-5.

9. McFarlane J, et al. Abuse during pregnancy. A protocol for prevention and intervention. 2nd ed. New York: National March of Dimes Birth Defects Foundation; 2001.

10. Warshaw C. Identification, assessment and intervention with victims of domestic violence. In: Warshaw C, editor. Identification, assessment and intervention with victims of domestic violence: improving the healthcare response to domestic violence. 2nd ed. San Francisco: Family Violence Prevention Fund; 1996. p. 49-85.

11. Campbell JC. Health consequences of intimate partner vio- lence. Lancet 2002;359(9314):1331-6.

[email protected] AJN ▼ March 2004 ▼ Vol. 104, No. 3 49

Complete the CE test for this article by using the mail-in form available in this issue or by going to Online CE at www.ajnonline.com.

12. Tjaden P, Thoennes N. Extent, nature and consequences of intimate partner violence. Findings from the National Violence Against Women survey. Washington, DC: Office of Justice Programs. Bureau of Justice Statistics. U.S. Depart- ment of Justice; 2000. http://www.ncjrs.org/pdffiles1/nij/ 181867.pdf.

13. Wisner CL, et al. Intimate partner violence against women: do victims cost health plans more? J Fam Pract 1999;48(6): 439-43.

14. Chalk RA, et al. Violence in families: assessing prevention and treatment programs. Washington, DC: National Academy Press; 1998.

15. Chalk R. Assessing family violence interventions. Linking programs to research-based strategies. Journal of Aggression Maltreatment and Trauma 2000;4(1):29-53.

16. McFarlane J, et al. An intervention to increase safety behav- iors of abused women: results of a randomized clinical trial. Nurs Res 2002;51(6):347-54.

17. Walker L. The battered woman. 1st ed. New York: Harper and Row; 1979.

18. Walker L. Battered women: sex roles and clinical issues. Professional psychology 1981;12(1):81-9.

19. Walker L. The battered woman syndrome. 2nd ed. New York: Springer; 2000.

20. Curnow SA. The Open Window Phase: help-seeking and reality behaviors by battered women. Appl Nurs Res 1997;10(3):128-35.

21. Violence Against Women Act of 1994, PL 103-322. 108 Stat. 1902.

22. McFarlane J, et al. Safety behaviors of abused women after an intervention during pregnancy. J Obstet Gynecol Neonatal Nurs 1998;27(1):64-9.

23. Lipsey M. Design sensitivity. New York: Sage; 1990. 24. McFarlane J, Parker B. Abuse during pregnancy. A protocol

for prevention and intervention. New York: National March of Dimes Birth Defects Foundation; 1994.

25. Parker B, et al. Testing an intervention to prevent further abuse to pregnant women. Res Nurs Health 1999;22(1): 59-66.

26. Block C, et al. Beyond public records databases: field strate- gies for locating and interviewing proxy respondents in homicide research. Homicide Studies 1999;3(4):349-66.

27. McFarlane J. Intimate partner violence victims in urban health care and justice settings. In: Dutton M, editor. Recruitment and retention in intimate partner violence research. Washington, DC; 2003. http://www.ncjrs.org/ pdffiles1/nij/201943.pdf.

50 AJN ▼ March 2004 ▼ Vol. 104, No. 3 http://www.nursingcenter.com

GENERAL PURPOSE: To provide registered professional nurses with an opportunity to review a study that tests the effectiveness of a telephone intervention protocol on the safety-promoting behavior of abused women.

LEARNING OBJECTIVES: After reading this article and taking the test on the next page, you will be able to • discuss the key findings of previous research in the

area of domestic violence, especially as they apply to this study.

• describe the methodology used in this study. • outline the results of this study and its implications

for nursing practice.

To earn continuing education (CE) credit, follow these instructions:

1. After reading this article, darken the appropriate boxes (numbers 1–15) on the answer card between pages 48 and 49 (or a photocopy). Each question has only one correct answer. 2. Complete the registration information (Box A) and help us evaluate this offering (Box C).* 3. Send the card with your registration fee to: Continuing Education Department, Lippincott Williams & Wilkins, 345 Hudson Street, New York, NY 10014. 4. Your registration fee for this offering is $16.95. If you take two or more tests in any nursing journal published by Lippincott Williams & Wilkins and send in your answers to all tests together, you may deduct $0.75 from the price of each test.

Within six weeks after Lippincott Williams & Wilkins receives your answer card, you’ll be notified of your test results. A passing score for this test is 11 correct answers (73%). If you pass, Lippincott Williams & Wilkins will send you a CE certificate indicating the number of contact hours you’ve earned. If you fail, Lippincott Williams & Wilkins gives you the option of taking the test again at no additional cost. All answer cards for this test on Increasing the Safety-Promoting Behaviors of Abused Women must be received by March 31, 2006.

This continuing education activity for 2.5 contact hours is provided by Lippincott Williams & Wilkins, which is accredited as a provider of continuing nursing education (CNE) by the American Nurses Creden- tialing Center’s Commission on Accreditation and by the American Association of Critical-Care Nurses (AACN 11696, category O). This activity is also pro- vider approved by the California Board of Registered Nursing, provider number CEP11749 for 2.5 contact hours. Lippincott Williams & Wilkins is also an approved provider of CNE in Alabama, Florida, and Iowa, and holds the following provider numbers: AL #ABNP0114, FL #FBN2454, IA #75. All of its home study activities are classified for Texas nursing continu- ing education requirements as Type 1. *In accordance with Iowa Board of Nursing administrative rules governing grievances, a copy of your evaluation of this CNE offering may be submitted to the Iowa Board of Nursing.

CE2.5Continuing Education HOURS