To develop the items for possible inclusion on the Decision to Leave Scale (DLS), 40 items were chosen from factors suggested by theory and past research to be associated with the decision to stay in or leave a romantic relationship, as reviewed above. To enhance their ecological validity, items were also developed in consultation with treatment directors from four county-wide community agencies that provide assistance for individuals in violent relationships for both urban and rural residents of eastern Pennsylvania. (See Table 1.) To identify underlying dimensions of concern to women in making the decision to stay in or leave a relationship, the 40 original items from the DLS were subjected to both an exploratory factor analysis with Sample 1 (656 college women), and a confirmatory factor analysis with Sample 2 (430 college women and community women in crisis shelters). Subscales for the DLS were then formed from items with consistently high factor loadings in both samples.
To enhance their statistical power, psychometric examinations of the DLS subscales were conducted wherever possible with a combined sample of participants from Samples 1 and 2. Also, to enhance their ecological validity and possible application as a needs-assessment tool for women in violent relationships, psychometric examination of the DLS subscales always included a focus on women participants who reported receiving violence from their present partners. Internal reliability for the DLS subscales was measured with Cronbach's alpha calculated for the items within each subscale, using the combined sample of participants from Samples 1 and 2. Test-retest reliability for each DLS subscale was measured by calculating Pearson correlation coefficients for scores provided on two occasions one to three weeks apart, which was available only for participants from Sample 2. Discriminant validity for each DLS subscale was evaluated using stepwise multiple regression with the combined sample to examine whether the sub-scale score was associated with personal and situational variables as expected from theory, past research, and our domestic violence consultants (including self-esteem, relationship duration, children, family social support, and violence from the present partner). Criterion validity for each DLS subscale was evaluated using a one-way ANOVA with the combined sample to examine whether expected differences were found in DLS concerns reported by women with three levels of experience and risk for violence: college women who had “never” experienced violence from the present partner, college women who had experienced violence at least “once or twice” from the present partner, and women in community crisis shelters for victims of domestic violence. Finally, to provide a partial answer to the often-asked question of, “Why does she stay?,” participants from the combined sample who reported violence from their partners were used in t tests to compare the specific DLS concerns for women who decided to stay, and for those who decided to leave.

Participants were undergraduate students recruited from three campuses of Penn State University. Classes from a variety of disciplines were sampled (e.g., business, chemistry, criminal justice, economics, English, geology, German, history, kinesiology, mathematics, music, nursing, psychology, theater). Three recruitment procedures were used. For 35 classes in which instructors could allow use of class time, recruitment of participants and completion of the questionnaire were both conducted during class time and produced a response rate of 94%. For 16 classes in which instructors could not allow use of class time, recruitment of participants was conducted during class, questionnaires were completed outside of class and dropped off later in designated drop box locations on campus, with a response rate of 70%. Finally, as part of the participant pool procedure required at the central University Park location, posters were prepared that described the study, and interested students used a website to make an appointment for one of 14 meetings where they completed the questionnaires, with a response rate of 85%. For all recruitment methods, the study was described to participants as an investigation into predictors of the quality of romantic relationships in college students. After completing the anonymous questionnaires, participants sealed them in envelopes, dropped the envelopes into a large covered box, and selected a small thank-you gift (e.g., gum, candy, chips, pens).
Across all three campus locations, 1,014 students returned questionnaires (343 men, 671 women). Responses from students in the first classroom of 34 participants (19 men, 15 women) were used as pilot questionnaires to see if participants appeared to understand and complete all parts of the questionnaire. Although data were collected on both women and men, only data from the women are included in the present report. Of the remaining 980 participants, 656 were women (mean age = 21.6 [SD = 5.9]; 80.1% Caucasian, 9.5% African American, 3.2% Asian American, 3.1% Hispanic). Also, the mean relationship duration in months was 33.6 (SD = 57.7); the mean number of children was 0.2 (SD = 0.7); 13.6% of participants were married or shared a household with their partner; their mean number of hours of employment per week was 12.5 (SD = 12.7); and 100% of them had some college education.

The anonymous questionnaire asked participants to provide demographic information including age, ethnicity, marital status, and children. Participants were also asked to report whether they were now involved in a romantic relationship, how long they had been in the relationship in months (or years), and “right now, what is your decision about the future of your current romantic relationship?” (1 = stay, 2 = leave, 3 = don't know). Responses of “don't know” were later combined with responses of “stay.”
Participants were presented with the 40 original items from the DLS in a randomly selected order as shown in Table 1. Participants were asked to “rate how important each factor is when making your decision to stay in or leave a relationship. Use a five-point scale from 1 = not at all important to 5 = very important, or X = not applicable.” Responses of “not applicable” were later recoded as 1. Participants were asked to respond to the DLS items concerning the present romantic partner, or the past romantic partner if they did not have a present partner, or to leave them blank if they had no partner about whom to report.
The questionnaire also included questions designed to measure a number of personal and situational variables that would be used to evaluate the discriminant validity of the DLS subscales because of theory and past research suggesting their relationships with the decision to leave process. These variables included self-esteem, relationship duration in months, number of children, violence received from the present partner, and social support from the family. Self-esteem was measured with the 10-item Rosenberg Self-Esteem Scale (Rosenberg, 1965), with participants asked to rate each item with a four-point scale ranging from 1 (strongly disagree) to 4 (strongly agree), with numerical responses reversed for appropriate items and the sum of item responses then used as each participant's score. Violence from the partner was measured using the eight-item violence subscale of the Conflict Tactics Scale (Straus, Gelles, & Steinmetz, 1980), with participants asked to rate the frequency of each behavioral item using a three-point frequency scale (1 = never, 2 = once or twice, 3 = many times) and with the sum of the eight ratings then serving as each participant's score. The eight behavioral items included the following: threw something at the other person, pushed or grabbed or shoved the other person, slapped the other person, kicked or bit or hit with the fist, hit or tried to hit with something, beat the other person, threatened with a knife or gun, used a knife or gun. Social support from family was measured with the 20-item subscale from the Social Support from Family and Friends Scale (Procidano & Heller, 1983), with participants asked to check items that were true for them and with a tally of item responses indicating family support then serving as each participant's score.

Participants included 430 college and community women, including 347 college women from two campuses of Penn State University and 83 community women from two crisis shelters for victims of domestic violence from both urban and rural counties (mean age = 24.5 [SD = 8.6]; 68.4% Caucasian, 17.8% African American, 3.0% Asian American, 5.2% Hispanic). Also, the mean relationship duration in months was 46.8 (SD = 71.7); 24.8% of them were married or shared a household with their partner; the mean number of children was 0.8 (SD = 1.3); their mean number of hours of employment per week was 18.5 (SD = 15.1); and 4.7% had no high school degree, 16.8% had a high school degree, 72.0% had some college, and 6.5% had a baccalaureate or higher degree.
The college women were recruited in 25 classes from a variety of disciplines, with a response rate of 97%. To obtain a sample of college women that was independent of Sample 1, recruitment for Sample 2 was conducted one year after recruitment for Sample 1; similar classes were sampled to reduce the probability of approaching the same students, and students were asked to refrain from completing the questionnaire if they had already participated. To keep their questionnaire responses anonymous, but to permit calculation of test-retest reliability, students were asked to identify themselves only with a four-digit code of their choice. When the research team returned to their class one to three weeks later, they were asked to use the same four-digit code so that their two questionnaires could be matched. Students deposited their completed questionnaires into a closed box. To thank them for their participation, they were given a newly minted “gold” dollar coin.
The community women were recruited with the assistance of counselors and administrators from the crisis shelters, with a response rate of 51%. As with the college women, participants identified themselves only with a four-digit code of their choice, then sealed their completed questionnaires into envelopes provided, and were given a newly minted “gold” dollar coin to thank them for their participation. Because the timing of their return to the crisis shelter was uncertain, the community women participants completed the questionnaire only on one occasion.

As in Sample 1, the anonymous questionnaire asked participants to provide demographic information including age, ethnicity, marital status, children, whether they were now involved in a romantic relationship, and their present decision to stay in or leave that relationship. Participants were also presented with the 40 original items of the DLS, with the same instructions as those given participants in Sample 1, and also asked to rate each item for “how important is each factor when making your decision to stay in or leave a relationship,” using a five-point scale ranging from 1 (not at all important or not applicable) to 5 (very important). Measures also gathered the same personal and situational variables as in Sample 1, including self-esteem, relationship duration in months, number of children, violence received from present partner, and social support from family.

The underlying dimensions for the reasons women decide to stay in or leave a romantic relationship were examined for the college women in Sample 1. This initial factor analysis used varimax rotation, considered the 40 original items of the DLS, and selected only factors that emerged with at least three items loading .40 or higher. Of the 656 women in Sample 1, 631 women completed all items of the scale as required for inclusion in the analysis. Seven factors emerged for the concerns women report when making the decision to stay in or leave a relationship: Fear of Loneliness, Childcare Needs, Financial Problems, Social Embarrassment, Poor Social Support, Fear of Harm, and Hopes Things Change. (See Table 1.)

A confirmatory factor analysis was conducted for the college women and community women in crisis shelters of Sample 2. (See Table 2.) The factor analysis again used varimax rotation and again considered the 40 original items of the DLS in order to compare the factor structure produced to that produced by the exploratory factor analysis of Sample 1, and to select items for inclusion in the final version of the DLS. (Of the 430 women in Sample 2, 420 women completed all items of the scale as required for inclusion in the analysis, including 342 college women and 78 community women in crisis shelters.) Seven factors emerged from the factor analysis of Sample 2 that were similar to those produced by the factor analysis on Sample 1. Of the 40 original DLS scale items, 30 items that demonstrated consistently high loadings for similar factors in the exploratory and confirmatory factor analyses were selected to make up subscales for types of concerns women may have when making the decision to stay in or leave a relationship: Fear of Loneliness (eight items), Childcare Needs (six items), Financial Problems (four items), Social Embarrassment (four items), Poor Social Support (three items), Fear of Harm (three items), and Hopes Things Change (two items). A woman's score for each subscale was then calculated as the sum of her ratings for items within that subscale, using the five-point scale she used to indicate the importance of each item in making her decision to stay in or leave the relationship (ranging from 1 = not at all important or not applicable to 5 = very important).

Internal reliability measures for each DLS subscale were calculated with Cronbach's alphas for all 1,051 participants combined from Samples 1 and 2. Cronbach's alpha was.82 for Fear of Loneliness, .84 for Childcare Needs, .69 for Financial Problems, .71 for Social Embarrassment, .63 for Poor Social Support, .71 for Fear of Harm, and .70 for Hopes Things Change. The mean internal reliability for the seven DLS subscales was .73
To determine internal reliability of the DLS subscales for women in violent relationships, the above calculations were repeated for the 196 women from Samples 1 and 2 who reported receiving violence from their present partner at least “once or twice” on any of the eight behavioral items from the violence subscale of the Conflict Tactics Scale.[ 1] Internal reliability scores for these women reporting violence from present partners were .85 for Fear of Loneliness, .85 for Childcare Needs, .72 for Financial Problems, .68 for Social Embarrassment, .60 for Poor Social Support, .73 for Fear of Harm, and .64 for Hopes Things Change. The mean internal reliability for the seven DLS subscales for women reporting violence was .72.

Test-retest reliability was examined for each DLS subscale score for college women from Sample 2 who completed the questionnaire on two occasions one to three weeks apart (n = 221) and who had remained in the same relationship across that time (n = 94). Pearson correlation coefficients were calculated between scores produced at the first and second administration of the questionnaire and found to be .85 for Fear of Loneliness (p < .001), .78 for Childcare Needs (p < .001), .57 for Financial Problems (p < .001), .68 for Social Embarrassment (p < .001), .63 for Poor Social Support (p < .001), .82 for Fear of Harm (p < .001), and .55 for Hopes Things Change (p < .003). The mean test-retest reliability for the seven DLS subscales was .70.
To determine test-retest reliability of the DLS subscales for women in violent relationships, the above calculations were repeated only for the 21 women who were in the same relationship for the two occasions they completed the questionnaire, and who had reported on the first occasion that they received violence from the partner at least “once or twice” for any of the eight behavioral items from the Conflict Tactics Scale. Their test-retest correlations were .91 for Fear of Loneliness (p < .001), .87 for Childcare Needs (p < .000), .47 for Financial Problems (p < .04), .71 for Social Embarrassment (p < .001), .55 for Poor Social Support (p < .01), .93 for Fear of Harm (p < .001), and .41 for Hopes Things Change (p < .07). The mean test-retest reliability for the seven DLS subscales for women in violent relationships was .69.

Discriminant validity for each DLS subscale was evaluated with stepwise multiple regression that examined the sub-scale's relationship with a set of personal and situational variables expected from theory, past research, and our consultants to be associated with the decision to leave process. These variables included self-esteem, relationship duration, number of children, social support from family, and violence from the present partner. Such a multiple regression analysis was chosen because it allowed identification of which of the possibly inter-correlated predictor variables could explain a unique portion of the variance in each DLS concern. To make the examination of discriminant validity for the DLS concerns particularly relevant to women who had experienced violence, the multiple regression analyses were conducted only for the 196 women from Samples 1 and 2 who reported receiving violence from their present partner at least “once or twice” for any of the eight behavioral items of the Conflict Tactics Scale. For 34.7% of these women, if violence was received “once or twice” from the partner, it was also received “many times.”[ 2]
From Traumatic Bonding Theory it was anticipated that women with low self-esteem or little family support would report significantly more of the concerns about social relationships measured with the DLS dimensions of Fear of Loneliness, Social Embarrassment, Poor Social Support, and Hopes Things Change with their partners. From Investment Theory it was anticipated that women with greater investments in relationship duration or children would report significantly more of the practical concerns measured with Childcare Needs and Financial Problems. From Investment Theory it was also expected that when the cost of violence from their partners was high, women would report significantly more Fear of Harm.
As expected from Traumatic Bonding Theory, low self-esteem explained a significant portion of the variance in DLS concerns that women reported about Fear of Loneliness (beta = −.18, t = 2.52, df = 187, p < .02), Social Embarrassment (beta = −.41, t = 6.26, df = 187, p < .001), Poor Social Support (beta = −.14, t = 2.03, df = 187, p < .05), and marginally that Hopes Things Change (beta = −.14, t = 1.89, d f = 188, p < .07). As expected from Investment Theory, the investment of children accounted for a significant portion of the variance in DLS concerns women reported about Childcare Needs (beta = .51, t = 8.21, d f = 189, p < .001) and Financial Problems (beta = .36, t = 5.26, d f = 189, p < .001). Also as expected from Investment Theory, the relationship cost of frequent violence explained a significant portion of the variance that women reported in Fear of Harm (beta = .14, t = 1.98, d f = 189, p < .05). Relationship duration and family support as measured in the present study were not significantly associated with any of the seven DLS concerns once self-esteem, children, and violence were in the multiple regression equation.

Criterion validity for each DLS subscale was evaluated using a one-way ANOVA with all participants from Samples 1 and 2 to examine whether, as expected, significant differences were found in DLS concerns reported by three groups of women who vary in their experience and risk for violence: college women who reported “never” receiving violence from their present partners on any of the eight behavioral items from the violence subscale of the Conflict Tactics Scale (n = 510); college women who reported that they had received violence from their partner at least “once or twice” on any of the eight behavioral items (n = 167); and women who were in community crisis shelters to protect them from partner violence (n = 83). It was anticipated that women in shelters would report significantly more concerns about Fear of Harm, Childcare Needs, Financial Problems, complaints about Poor Social Support, and Social Embarrassment at needing to turn to a community crisis shelter to deal with these concerns. However, because women in shelters had crossed the decision threshold to “take action” to leave violence, it was expected that they would report significantly less Fear of Loneliness at the thought of leaving the violent partner, with fewer remaining Hopes Things Change for the better with the partner.
As expected, women in three levels of risk for violence reported significantly different concerns about Fear of Harm (F[ 2,739] = 6.41, p < .003), with women in crisis shelters reporting more concern than either college women with or without experience of partner violence (t = 3.20, d f = 240, p < .003; t = 3.53, d f = 576, p < .001; respectively). Also as expected, women in the three groups differed on concerns about Childcare Needs (F[ 2,739] = 29.36, p < .001), again with women in shelters reporting more concern than either college women with or without violence (t = 5.52, d f = 240, p < .001; t = 7.85, d f = 576, p < .001; respectively). Groups differed in concerns about Financial Problems (F[ 2,739] = 51.25, p < .001), with women in shelters reporting more concern than either college women with or without violence (t = 6.15, d f = 240, p < .001; t = 10.18, d f = 576, p < .001; respectively). Groups differed in complaints about Poor Social Support (F[ 2,739] = 59.65, p < .001), with women in shelters reporting more complaints than either college women with or without violence (t = 8.47, d f = 240, p < .001; t = 10.65, d f = 576, p < .001; respectively). Groups differed on concerns about Social Embarrassment (F[ 2,739] = 82.58, p < .001), with women in shelters reporting more concern than either college women with or without violence (t = 7.28, d f = 240, p < .001; t = 12.83, d f = 576, p < .001; respectively). Also as expected, the three groups differed in Fear of Loneliness (F[ 2,739] = 3.77, p < .03), with women in crisis shelters reporting less fear of being without the partner than did either college women with or without violence (t = 2.45, df = 240, p < .02; t = 1.79, df = 576, p < .08). However, although the groups differed significantly in Hopes Things Change (F[ 2,739] = 4.25, p < .02), women in crisis shelters did not differ significantly from college women either with or without the experience of violence. The only significant difference between college women with and without the experience of violence was that college women with violence reported significantly more Social Embarrassment (t = 3.43, d f = 662, p < .002) and Hopes Things Change (t = 2.97, d f = 662, p < .004).

Finally, the DLS subscales were used to approach the question of “Why does she stay?” by identifying concerns that appeared to be the most important perceived barriers for women facing the decision to stay in or leave a violent relationship. For 196 participants from Samples 1 and 2 who reported receiving violence from the present partner at least “once or twice” on any Conflict Tactics Scale items, t tests compared DLS subscale scores for those reporting the decision to stay in the relationship (n = 166), and for those reporting the decision to leave (n = 30, or 15.3%). As mentioned earlier, if violence was received “once or twice” it was also received “many times” for 34.7% of these 196 women. No information was available about whether women who reported that they would leave actually did so.
The decision to leave was associated with greater Fear of Harm (t = 2.12, d f = 189, p < .04), Childcare Needs (t = 2.75, d f = 189, p < .008), Poor Social Support (t = 3.82, d f = 189, p < .001), and marginally more concerns about Financial Problems (t = 1.91, d f = 189, p < .06). The decision to stay was associated with greater Fear of Loneliness (t = 2.23, d f = 189, p < .003). Hopes Things Change did not differ significantly between women who had experienced partner violence and decided to leave or those who decided to stay in the relationship (t = .88, d f = 189, p < .38).