Assessing Self-Efficacy Among Incarcerated Women: Scale Development and Psychemetric Properties

The increasing number of incarcerated women in the United States has created the necessity to better understand the needs of this population, so that successful correctional programs can be implemented. Programs that address these needs and are aimed at increasing relevant skills have been linked with program effectiveness. However, increasing skills alone is not enough. If these women leave prison without the necessary self-efficacy, it is likely that they will not attempt to perform these skills after release . The objective of this research was to develop a self-report questionnaire to measure selfefficacy in this population. The Self-Efficacy for Incarcerated Women (SIW) Scale yielded five coherent factors, representing different domains of self-efficacy: vocational, substance abuse treatment, relationship, release issues, and parenting . Ultimately, this scale could be used as one form of assessment for targeted intervention programs offered to female prisoners.


List of Tables
There has been a dramatic increase in the number of women imprisoned in the United States in the past two decades (Chesney-Lind, 1997;Phillips & Harm, 1998;Maeve, 1998). In fact, the rate of growth for female inmates has exceeded that of male inmates every year since 1981 (Singer, Bussey, Song, & Lunghofer, 1995). Furthermore, from a rehabilitative standpoint, the current system of incarceration does not appear to be effective. In a study conducted with two hundred incarcerated women, Singer et al. (1995) found that the average number of previous incarcerations for this population was 3.9. Simply locking these women up and hoping that they will learn their lesson does not appear to be the answer.

Treatment Needs
The soaring numbers of female inmates, along with their high recidivism rate, create the necessity to better understand the needs of this population so successful correctional programs can be implemented. However, women in prison represent a neglected population about which little is known (Coll, Miller, Fields, & Matthews, 1998;Phillips & Harm, 1998;Gray, Mays, & Stohr, 1995;Singer et al. 1995). There is currently very little information about what types of interventions are successful with female inmates (Koons, Burrow, Marash, & Bynum, 1997). The limited amount of attention given to the needs of these women is generally assumed to be a function of the fact that, despite their rapidly increasing numbers, women still constitute only a small fraction (5.7 %) of the total prison population (Phillips & Harm, 1998;Koons et al., 1997). For the most part, the facilities and services offered to female inmates are 1 primarily based upon research and experience with male inmates (Coll et al., 1998;Phillips & Harm;1998;Patterson, 1995). Programs available to this population have been "cloned from programs implemented for male offenders and provided to women offenders without consideration as to whether they were appropriate" (Koons et al., 1997, p. 517). There are undoubtedly program elements that need to be matched to offender characteristics unrelated to gender (Koons et al., 1997), such as structure or opportunity to practice learned social skills (Andrews, Bonta, & Hoge, 1990). However, there are needs that are unique to females which have important implications for the design of successful interventions with this population .
For example, women in prison have more serious problems with substance abuse than men do (Sheridan, 1996;Chesney-Lind, 1997;Prendergast, Wellisch, & Falkin, 1995). Miller (1984) found that female inmates were more likely than males to have used hard drugs and to have reported using drugs daily. Women were also more likely to have been under the influence of drugs when they committed the crime for which they are imprisoned, and more likely to have committed that crime to support their drug habit (Chesney-Lind, 1997;Prendergast et al., 1995). Female inmates are also more likely than males to be infected with HIV (Chesney-Lind, 1997), and to have suicidal thoughts (Coll et al., 1998). Their educational attainments, job skills, and work histories are lower than those of their male counterparts (Coll et al., 1998). In a study involving 566 female inmates from five exclusively women's jails, Gray et al. (1995) found that only 50% of those inmates had completed high school or had begun college. Furthermore, although 57% reported that they were supporting others at the time of their arrest, only 35% were employed. Family histories are also different: women are more likely to have been victims of sexual abuse as children (Sheridan, 1996;Snell & Morton, 1994), and this gender difference in abuse continues into adulthood, when more women than men report being victims of sexual, physical and emotional abuse (Snell & Morton, 1994). In addition to these demographic differences, women are often faced with the additional burden of parenting. Snell and Morton (1994) found that two thirds of incarcerated women had at least one child under the age of eighteen, and many women in prison are pregnant (Koons et al., 1997). Care and placement of children tends to be a more salient issue for women than for men, since a higher percentage of women are primary caregivers (Coll et al., 1998). In addition, many women feel guilt about being absent from their children's lives and worry that they will no longer have custody of their children when they are released (Bloom & Steinhart, 1993).
Rather than simply applying programs that were designed for male inmates, these differences suggest the importance of taking the unique needs of female inmates into account when designing interventions for them. There is an increasing effort among researchers and scholars to understand and document these needs, and to discover what characteristics make a program effective (Prendergast et al., 1995;Coll et al., 1998;Simon, 1991;Gray et al., 1995;Owen & Bloom, 1995;Singer et al., 1995;Sheridan, 1996;Koons et al., 1997;Chesney-Lind, 1997).
. In one of these studies, Austin, Bloom, and Donahue (1992) found that the most promising programs for female inmates are those that address specific women's needs (e.g., parenting, substance abuse, relationships, domestic violence) and those which are aimed at building skills. Koons et al. (1997) also found that the development of specific skills was considered, by both corrections experts and program participants, to be linked with effective programs. It is clear why this skill development is crucial for the success of these programs. When women are released from prison, they simply return to the same conditions that they faced before they were sentenced; without the necessary skills, it is not surprising that they often resort to the same survival tactics that resulted in their initial incarceration. Teaching these women relevant skills before they are released, therefore, is of utmost importance.

Importance of Self-Efficacy
However, simply having the skills is not enough-it is also important that these women have confidence that they are capable of performing these skills. As Bandura (1997) explains, "having knowledge and skills does not produce high attainments if people lack the self-assurance to use them well" (p. 80). This belief in one's own ability to perform an action is what Bandura (1977) labeled self-efficacy.
According to self-efficacy theory, people's beliefs about their ability to perform a behavior are directly related to the likelihood that they will attempt that behavior. People avoid activities that they believe exceed their coping capabilities, while they tend to undertake those that they judge themselves capable of handling (Bandura, 1977). In addition, perceptions of self-efficacy determine how long people will continue to attempt a task in the face of difficulty or adversity. When those with high self-efficacy are faced with obstacles, they tend to exert even greater effort. On the other hand, those with low self-efficacy are more likely to give up their attempt altogether (Bandura, 1982).
It is well documented that incarcerated women are faced with a variety of challenges when they re-enter the community, such as poverty, unemployment, lack of education and housing, drug and alcohol addictions, abusive partners, and young children 4 to care for. Even if these women learn the skills necessary to face these adversities successfully, if they leave prison without the confidence to perform these skills, it is quite likely that they will not attempt them at all. Unless people believe that their actions can produce desired effects, they have little incentive to act (Bandura, 1997). Therefore, it is crucial that these treatment programs increase self-efficacy as well as teach skills.
The objective of this study was to develop a self-report questionnaire that can be used to measure self-efficacy among this population. This questionnaire has the potential to eventually be used as one measure of effectiveness for targeted intervention programs.
Items for this purpose were developed as part of a federally funded project evaluating the Discharge Planning Program of women's medium and minimum-security facilities in Rhode Island. This secondary analysis examined the psychometric properties of the previously developed items, with the expectation that the women's self-report ratings would yield coherent, interpretable factors that would be correlated with one another. It was hypothesized that these factors would be consistent with the several content domains tapped in item construction: vocation/ education, substance abuse, parenting, relationship issues, and release issues.
The factor structure that emerged from the exploratory analyses was evaluated and compared with alternative models that represented different conceptualizations of the factor structure. The models tested were: it allows for relationships between the self-efficacy sub-scales. 5 3. Uncorrelated Model: This model also uses the factors suggested from the exploratory analyses, but proposes that these factors are independent of one another.
It was also hypothesized that these factor-based sub-scales would be related to other measures available in the data set dealing with perceived stress, alcohol and drug use, partner experiences, and various background variables (relationship status, whereabouts of children, education and job history , criminal history, age of first arrest, and length of current sentence).
Perceived stress was hypothesized to have a negative correlation with selfefficacy. According to Bandura (1997), stress reactions are governed largely by beliefs of efficacy (p. 262). It is not the objective properties of environmental threats and demands that trigger stress reactions; rather, it is a low sense of efficacy to exercise control over these threats and demands that leads to the experience of stress (Bandura, 1997). When we feel we can effectively deal with environmental stressors, they do not upset us, and therefore we do not experience them as stressful. However, we will experience stress when we feel that our coping capabilities are being overwhelmed.
Therefore, the women who report high levels of stress for a certain content domain will likely also report that they are not confident of their abilities in that same domain. For example, women who report high levels of stress regarding parenting will probably report low levels of self-efficacy for parenting as well.
It was assumed that previous alcohol and drug use would have a negative correlation with any self-efficacy factor(s) concerning substance use. The most influential source of perceived efficacy is enactive mastery experience; personal experiences of success builds a robust belief in one's efficacy, while failures undermine 6 one's feelings of efficacy (Bandura, 1997). Women who report high levels of drug or alcohol use have likely experienced failure at trying to overcome a substance abuse problem, and this failure would then undermine their confidence in overcoming this problem in the future.
Partner experiences were also believed to be related to self-efficacy. High levels of physical or emotional abuse from a primary partner were hypothesized to have a negative correlation with any self-efficacy factor(s) regarding relationship issues. If your primary relationship is marked by physical or emotional abuse, it is likely that you will experience this relationship as a failure, which will undermine your belief that you possess the skills necessary to have a successful relationship.
Several Background variables were also hypothesized to be related to selfefficacy. Women who had previously trained for a specific job or kind of work, women who had held a steady job prior to incarceration, and women who had gone to school for a long period of time were all expected to have higher self-efficacy regarding issues of vocational success. The job, training, and education would likely offer opportunities for enactive mastery experience, which would lead to increased self-efficacy. Women who had a husband or a steady partner were expected to have higher self-efficacy regarding issues of relationship success. The maintenance of a steady relationship would likely be experienced as one form of relationship success; therefore, the women who had attained this would be more likely to feel confident in this domain. Womenwho had been in prison previously were expected to believe that they do not have the skills necessary to stay out of prison. These women have had more experiences of failure in remaining out of prison than the women who are in prison for the first time. Both the age of first arrest 7 and the length of the current sentence were also expected to be correlated with selfefficacy regarding remaining out of prison. Women who begin committing crimes at a young age and women who commit more serious crimes were expected to lack the confidence that they can refrain from committing crime in the future.
Finally, women whose children had lived with them in the month prior to their arrest were expected to feel confident that they could take care of their children. While there are undoubtedly several factors that could cause children to be living with someone else, it is likely that one of the reasons is that the woman was having difficulty caring for them.

Method
Participants Subjects were 120 females who were incarcerated in the minimum and medium security women's prison in Rhode Island. The longitudinal design of the original study required that subjects be selected on the basis of their release date. Initially, only women who would be released by the third phase of that study were given the questionnaires.
Volunteers who would not be released by that time were eventually included, but were informed that they would not be involved in all phases of the study.
Demographic characteristics of this sample are consistent with other descriptive findings regarding incarcerated women. Most (50.8%) were 25-35 years old, with 34.2% older than that and 15.0% younger. The majority (61.7%) reported white as their race, followed by Hispanic-American (17.5%), African-American (15.0%), and Native American (2.5% ). Ninety-two percent reported English as their primary language; the remaining eight percent reported Spanish. About half (56%) had graduated from high 8 school, and 56. 7% did not have a job at the time of their arrest. The vast majority of these women (80.8%) were mothers, and of these women, 40% were living with their children before their arrest.

Procedures
Self-report questionnaires were administered by researchers (advanced undergraduates, graduate students, and faculty) within the residential areas of prison buildings in small group sessions, including interviews with each respondent. On average, these questionnaires took approximately one hour to complete . To account for the low literacy of some of the respondents, every woman had the option of being read the questionnaire (in English or Spanish) by the researcher.

Measures
Self-efficacy for Incarcerated Women (SIW). The scale that was investigated here was included as part of a packet of self-report questionnaires. Participants were asked to fill out these questionnaires at three time points: shortly after entering the prison, immediately before release, and three months after release. This research analyzed data from both the first and second phases of the study.
All of the 50 items on this self-efficacy questionnaire start with the question "How sure are you that you can ... ". Participants had five responses available: extremely sure, very sure, medium sure, not very sure, and not at all sure.
Since self-efficacy is conceptualized as a multifaceted phenomenon , rather than a global disposition (Bandura, 1997), the questionnaire covered various domains of skills .
The skills that were included were generated from various sources . First, an in-depth review of the existing literature was conducted . Second, female inmates were asked for their opinions during several small focus groups. The last source of items was interviews with the Discharge Planning Committee at the prison. These staff members used a Discharge Plan Form that provided important information about relevant content domains. According to Bandura (1997), "in developing efficacy scales, researchers must draw on ... expert knowledge of what it takes to succeed in a given pursuit", (p. 43) . It was assumed that the Discharge Planning Committee would offer this expert knowledge.
The generated items then went through a review stage, during which all of the members of the larger research project had a chance to revise them. The scale, as administered, is provided in Appendix A.
Background. There were a number of demographic and life history questions in the instrument battery . The questions relevant to this study were regarding children, previous incarcerations, age of first arrest, length of current sentence, education and job training history, and partner status. These questions are provided in Appendix B.
Stress. The Stress scale that was included in the instrument battery was designed to measure the level of perceived stress prior to arrest. Subjects were asked to rate how often they were worried about each of the 24 items in the month prior to their arrest. The items reflected a range of domains that could potentially cause stress. This scale was adapted from Cohen's (1983) Perceived Stress Scale, which was designed to measure stress as a unidimensional construct. Preliminary analyses suggest that this version is measuring six domains of stress: Parenting, Housing/ Basic Needs, Personal , Education/Vocation, Relationship Issues, and Health/ Safety.
The Stress Scale, along with factor loadings from the preliminary principal components analysis, is provided in Appendix C.
Alcohol and Drug Use. The Alcohol and Drug Use Scale that was included in the instrument battery was designed to explore the substance abuse history of these women.
Participants were asked about both their current (one month prior to arrest) and past frequency of alcohol and drug use. This scale was adapted from the substance abuse measure used by Harlow, Newcomb, and Bentler (1986), that was considered to have one latent factor. Preliminary analyses indicate that this version has three factors, Alcohol Use, Drug Use, and Teenage Substance Use.
The Alcohol and Drug Use Scale, along with the factor loadings from its preliminary principal components analysis, is provided in Appendix D.

Current or Recent Partner Experiences. The Current or Recent Partner
Experiences Scale was included in the instrument battery to investigate the nature of the relationships that these women have with their primary partners. Subjects were asked to rate the frequency of various types of emotional, physical, or sexual abuse that they may have experienced from this partner. This scale was adapted from Straus' (1979) Conflict Tactics Scale, which has four factors: reasoning, verbal aggression, violence, and "serious/lethal" violence. None of the reasoning items were included in this version, and several additional items were added. Preliminary analyses indicate that this adapted version has two factors: Physical/Severe Abuse and Psychological/Emotional Abuse.
The Current or Recent Partner Experiences Scale, with factor loadings from its preliminary principal components analysis, is provided in Appendix E.

Results
The SIW Scale was developed using three stages: exploratory analyses, confirmatory analyses, and construct validation.
Exploratory Analyses. The SIW Scale data collected during the first phase of the study was entered into SPSS for analysis. An exploratory principal components analysis (PCA) was conducted on the matrix of item intercorrelations generated from this data using pair-wise deletion (N = 120). Since it was hypothesized here that the factors would be correlated, an oblique (Promax) rotation was then used to simplify interpretation.
Items which did not load highly (above .50) on any factor, items which did not conceptually fit with the factor it loaded on, and complex items were deleted, which resulted in a 24-item scale. Five components were retained as a result of a PCA conducted on these 24 items. Only these five components had eigenvalues greater than 1.0, and the Scree Procedure (Cattell, 1966) indicated this solution as well. Together, these factors explained 69% of the total variance. Table 1 presents each variable along with its loading on the appropriate factor. Factor one, which accounted for 39.6% of the total variance, represents vocational self-efficacy. Its internal consistency, as measured by Chronbach's coefficient alpha (Chronbach, 1951), was .89. Factor two accounted for 10.1 % of the total variance and had an internal consistency of .87. It appears to be measuring self-efficacy for substance abuse treatment. Factor three accounted for 8.3% of the variance and had an internal consistency of .84. This component seems to be measuring relationship self-efficacy. Factor four represents self-efficacy for coping with release issues. It accounted for 5.9% of the variance and had an internal consistency of 12 .85. The fifth factor had an internal consistency of .86, accounted for 5.1 % of the variance, and appears to be measuring parenting self-efficacy.
All of the self-efficacy components were correlated with one another. Table 2 shows these factor intercorrelations.

Confirmatory Analyses
The SIW Scale data collected during the second phase of the study were entered into the EQS (Bentler, 1989) computer program for analysis. Maximum likelihood confirmatory factor analyses was conducted on these data to determine the plausibility of the factor structure suggested by the exploratory analyses. Alternative models representing different conceptualizations of the factor structure were also evaluated for comparison purposes. Figures 1-3 show the three models that were tested: one factor, five factor correlated, and five factor uncorrelated.
Several indices of overall model fit were computed and compared, as there is currently no clear agreement on a single optimal test (Maruyama, 1998). Three absolute overall fit indices, the maximum likelihood chi-squared test, the chi-squared to degrees of freedom ratio, and the average absolute standardized residual (AASR), were examined for each model. Absolute indices seek to determine whether the unexplained, or residual, variance remaining after model fitting is significant. In confirmatory factor analysis, an ideal chi-squared value has a non-significant probability value, which indicates that the matrix of residuals generated by the model are not significantly different than zero.
Unlike the chi-squared test, the chi-squared to degrees of freedom ratio allows for the comparison of alternative models by controlling for the differences in the number of their parameters . A model is considered to empirically "fit" the data when the chi-squared to 13 degrees of freedom ratio is less than 2.0. Both of these indices have the limitation that they are a direct function of sample size. The AASR is a measure of the difference between the predicted and observed matrices; a value of .06 or less is generally considered an acceptable measure of fit.
The Comparative Fit Index (CFI; Bentler, 1990), which is a relative index, was also calculated for each model. Relative indices seek to understand how well a model explains the observed data when compared with the worst fitting, or null model, which assumes that there are no common factors and that sampling error alone explains the covariances among the items. All relative indices range from zero to one, with higher values indicating better fit.
Neither the five factor uncorrelated model nor the one factor model had adequate fit of the data.
Another criterion used to determine model fit was the significance of individual parameters. Each factor loading between an observed variable and its underlying construct was examined. Ideally, each of these hypothesized parameters would be significant. This type of evaluation is very different from assessing overall fit. For example, it is possible for a good fitting model to have insignificant parameters in places where significance was expected, and it is also possible to find strong relationships between variables in a poorly fitting model (Maruyama, 1998). Each parameter of the five factor correlated model was significant (see Table 4).
14 In addition to overall fit indices and individual parameter significance, the five factor correlated model and the five factor uncorrelated model were also directly compared to one another using the chi-squared test of differences. This test can only be used when models are equivalent except for a subset of parameters that are free in one and fixed in another (they are "nested"). The two models tested here fit this criterionthey are equivalent except for the regression coefficients between the factors , which are free in the correlated model and set to zero in the uncorrelated model. Results shown that these parameters were significantly adding to model fit; chi-squared difference (df = 10) = 111.85, Q < .001. Table 5 shows these results.
The self-efficacy sub-scales were then examined for construct validity by examining their relationship with various other measures included in the questionnaire packet. Following the reasoning described in the hypotheses, a series of plausible relationships were explored.
The continuous variables were analyzed using Pearson-R correlations, which are shown in Table 5. The categorical variables were analyzed using one-way analyses of variance, and are shown in Table 6.

Discussion
The Self-Efficacy for Incarcerated Women Scale yielded five coherent, interpretable sub-scales that represent domains of critical importance for incarcerated women. The findings are promising and suggest that this scale can be used in future studies as a way to measure self-efficacy in this population.
There was strong support for conceptualizing the first factor as vocational selfefficacy. Each of the items included in this sub-scale is measuring one aspect of gaining or keeping employment. In addition, this factor was related to previous employment, education, and job training.
The second factor is slightly more difficult to interpret, but there seems to be empirical support for conceptualizing it as substance abuse treatment self-efficacy. Some of the items that make up this factor are directly related to getting and staying in treatment, while others are measuring different skills that are helpful when trying to get or remain sober, such as making friends who do not have bad habits and knowing things that are temptations. This factor's negative correlation with both alcohol and drug use adds additional support for this interpretation.

17
It is interesting to note the items that contribute to the Relationship factor. At first glance, it appears that these items do not relate to each other in a meaningful way.
However, we believe that, for these women, having a house to live in and having money to pay bills are experienced as aspects of a successful relationship. Support for this reasoning can be seen in the fact that this factor was marginally correlated with the scale that measured stress regarding housing and basic needs.
The fourth factor is clearly measuring release issues. Each of the four items that make up this factor is directly related to one aspect of the release process. The interesting result of this factor was that its relationship to background variables did not tum out as hypothesized. It turned out that women who had more time left in their current sentence and women who had been incarcerated previously had higher scores on this sub-scale.
One possible explanation for the first result is that women who are not very close to their release date may have a strong belief that they can make the changes in their life necessary to remain out of prison. However, as their release date approaches they begin worrying about just how they will make these changes, and their confidence decreases. It is possible that the second result can be explained by thinking of the women with previous incarcerations as having had more opportunity to "learn their lesson" or to rehabilitate inside the prison. Further research is needed to investigate the plausibility of these explanations.
The four items contained within the final factor are clearly all aspects of parenting. Women whose children were living with them prior to their arrest had higher scores on this sub-scale , offering additional support of this conceptualization. 18 There is certainly room for additional psychometric study of the SIW. For example, a major limitation of this research was its small sample size. Future research conducted on a larger sample could offer stronger support for this factor structure. In addition, the generalizability of this study may be limited due to the fact that data was collected at a single site in Rhode Island. However, the demographics of our sample are not out of line with those for other similar settings .
Another limitation to this study is the fact that many women whose data were used during the first, or exploratory, phase of study were also used in the second, or confirmatory phase . Ideally, these two samples should have been completely separate.
In addition, there are limitations to using self-report data with incarcerated populations. For example, it is likely that a positive response bias existed in the data.
Although the women were informed that their responses were to be kept confidential, the nature of the prison environment makes it possible the women did not believe that this was true. It is possible that the women believed their responses would affect the treatment they received in prison, or even the length of their sentence. However, there was still enough variability in the responses to generate interpretable factors and correlations with other variables.
Future research is needed to investigate the relationship between these selfefficacy sub-scales and outcome variables such as recidivism and adjustment variables after release (such as holding down a job or staying sober). The SIW also has the potential to be used to measure any immediate effects of programs offered within the prison, such as parenting classes or vocational readiness interventions. The most useful potential application of this scale will be its ability to be predictive of successful outcomes. Ideally, programs for incarcerated women should be implemented which address the domains measured by the SIW Scale, and these programs will increase selfefficacy. This increased self-efficacy, in tum, will lead to the successful adjustment of these women into the community upon release.
20      Note. Boldface indicates ANOV As used for construct validity purposes. TRAIN = Did you ever train for a specific job?; PARTNER = Do you have a husband or regular partner?; FIRST= Is this the first time you have gone to prison?; CHil..1) = In the month before you were arrested, were any of your children living with you? *p < .05. **p < .01.

Confidence In Skills
When you complete your prison stay you will move back into the community. We would like to know how confident you feel right now that you can handle the challenges you will face after you're released . CIRCLE the answer closest to your true feelings.
How SURE are you that you can: In the month before you were arrested, were any of your children living with you?
How long have you gone to school?
Did you ever train for a specific job or kind of work?
In the last five years before you were arrested, how much did you hold a steady job?
Is this the first time you have been sentenced for a crime or have gone to prison?
How old were you the first time you were arrested?
How much longer do you expect to be here for this sentence? Factor I: Drug Use During the last month you used drugs, how often did you shoot drugs with a needle that had already been used by someone else? In the month before you were arrested, how often did you use drugs? At the time in your life when you used drugs the most, how often did you use them? During the last month you used drugs , did your main partner use them with you? Factor 2: Teenage Substance Use How old were you when you first used any of these drugs? During your teen years (before age 18), how often did you use drugs? During your teen years (before age 18), how often did you have a drink of beer, wine, or liquor? How old were you when you first used alcohol (other than just tasting it)?