Freedom Project

Freedom Project trains prisoners in nonviolent communication and meditation. Two complementary studies of its effects are reported in this article. The first study is correlational; we found decreased recidivism rates among prisoners trained by Freedom Project compared with recidivism rates in Washington state. The second study compared trained prisoners with a matched-pair control group and found improvement in self-reported anger, self-compassion, and certain forms of mindfulness among the trained group. Ratings of role-plays simulating difficult interactions show increased social skills among the group trained by Freedom Project than in the matched controls.

The number of people incarcerated in the United States has significantly increased in the past three decades. In 1980, approximately 330,000 people were incarcerated (Beck & Gilliard, 1995), but by the end of 2010, there were over 1.6 million people in federal and state prisons (Guerino, Harrison, & Sabol, 2011), greater than the population of Idaho (U.S. Census, 2010). At the end of 2010, 4.1 million people were on probation and 841,000 were on parole (Bonczar & Glaze, 2011) and more than 700,000 state and federal prisoners return to civilian society every year (Guerino et al., 2011). Nearly two thirds of released prisoners are rearrested for a new crime within 3 years of their release and half are sent back behind bars (Urban Institute, 2009). The cost in human suffering and the sheer enormity of the problem has fostered a surge of interest in the topic of prisoner training and reentry (Bucklen & Zajac, 2009). Given the public health and public safety issues associated with reentry, developing policies and programs that promote successful reentry and prevent reoffending are indispensable for a healthy community (Wheeler & Patterson, 2008).
Several studies (e.g., Laub & Sampson, 2003;Tripodi, 2010) have identified events, such as marriage and forming strong ties to work, which can alter criminal trajectories. Other studies (e.g., Giordano, Cenkovich, & Rudolph, 2002;Maruna, 2001;van der Knaap, Alberda, Oosterveld, & Born, 2011) have found evidence that psychological changes are even more important than keystone events in stopping or slowing down the rate of offending. Interventions that target antisocial attitudes, poor decision-making, problem-solving skills, self-control or self-regulation skills, association with criminal friends, and substance abuse can reduce recidivism by upwards of 20 percentage points compared with programs that ignore these needs (Andrews & Bonta, 2003). Bucklen and Zajac (2009) concluded, "The greatest problem for parolees was managing themselves in a prosocial manner while facing demands from their environment" (p. 239).
Communication can be a critically important element of prosocial behavior. A prime example, as described in the General Aggression Model (C. A. Anderson & Bushman, 2002), is the use of verbal aggression, a catalyst to physical assault among violent criminals (Infante & Rancer, 1996;Toch, 1969). The use of verbal aggression is significantly correlated with the extent of resorting to physical assault (C. M. Anderson & Rancer, 2007;Infante & Wigley, 1986), felony offenses (Spillane-Grieco, 2000), and physical aggression in correctional centers (Cornell, Peterson, & Richards, 1999;Mejovšek, Buðanovac, & Šućur, 2000). A number of researchers have found that a deficiency in verbal skills is a major cause of verbal aggression (e.g., Bandura, 1973;Infante, Chandler, & Rudd, 1989;Roberto, 1999;Toch, 1969). It follows that teaching communication skills is an important intervention aimed at reducing aggression and violence in society.
Compassionate social communication can be taught according to the concept of Nonviolent Communication (NVC), originated by Marshall Rosenberg (1999), which involves noticing others' behaviors, examining the accompanying feelings, making requests, and acknowledging the needs that have been met/unmet. In this way, during conflicts, there is no blame or fault assigned to either party. NVC has been extensively used to train educators, managers, mental health care providers, gangs, lawyers, military officers, clergy, government officials, individuals, and families (Rosenberg, 2003); it has also been used to train prisoners in the United States and other countries (Niebuhr, 2001). Correctional facilities in Washington (Freedom Project, 2009), Oregon (Center for Compassionate Living, 2012), California (Bay Nonviolent Communication, 2008), British Columbia, Denmark, and Sweden (Bryson, 2000) have maintained their support and use of NVC for many years There is a small but growing body of academic literature, including research, surrounding the efficacy and implications of using NVC within various contexts and settings (Branscomb, 2011;Cox & Dannahy, 2005;Dougan, 2011;Fullerton, 2009;Hulley, 2006;R. Jones, 2005;S. Jones, 2009;Nonviolent Communication Experimental Project, 1999;Savic, 2005;Steckal, 1994;Young, 2011).
Two of the best-designed studies were carried out by Donna Riemer (2007Riemer ( , 2009, who researched the outcome of interventions involving NVC at maximum and medium security forensic psychiatric units in Wisconsin. In both settings, Reimer found remarkable decreases in violence, use of restraints, and involuntary seclusion on the ward when compared with pre-NVC-training levels. A more recent study involving parolees showed that NVC training was associated with significant increases in empathy . The NVC process consists of four basic skills: (a) observing a situation without evaluation, (b) acknowledging the accompanying feelings, (c) understanding how those feelings are a result of a met/unmet need, and (d) clearly requesting concrete actions and exploring how to honor all parties' needs in a flexible and creative manner, without demands (Rosenberg, 1999).
These basic skills require mindfulness for their implementation. Mindfulness meditation has been shown to ameliorate aggression, anxiety, recidivism in prisoners, and drug use (Alexander, Walton, Orme-Johnson, Goodman, & Pallone, 2003;Hawkins, 2003;Parks et al., 2003;Perkins, 1998). Meditation and mindfulness have also been used successfully to help nonprison populations overcome intractable conflict (Horton-Deutsch & Horton, 2003), to help parents whose children behave in particularly challenging ways (Koren-Karie, Oppenheim, & Getzler-Yosef, 2008), and for counselors who work under stress (Christopher, Christopher, Dunnagan, & Schure, 2006). Mindfulness has been described as "the awareness that emerges through paying attention on purpose, in the present moment, and nonjudgmentally to the unfolding of experience moment by moment" (Kabat-Zinn, 2003, pp. 145-146). Mindfulness meditation can improve the capacity to regulate emotion, combat emotional dysfunction, improve patterns of thinking, and reduce negative mind-sets (Siegel, 2007, p. 6). Samuelson, Carmody, Kabat-Zinn, and Bratt (2007) offered a vivid example of the beneficial effects of mindfulnessbased training in correctional facilities; this study examined 1,350 inmates trained in mindfulness-based stress reduction. The authors found significant pre-to post-training improvements in hostility, self-esteem, and mood disturbances.
The body of research on mindfulness and the proven effects of NVC lead to the theory that mindfulness can support NVC by increasing awareness, sharpening observational skills, and creating time and space to formulate a response; in turn, NVC can support mindfulness by decreasing violent outbursts that interfere with presence of mind.
The combination of NVC and meditation has been taught by Freedom Project in some of Washington State's correction facilities. Freedom Project is a program for inmates run mostly by former inmates and volunteers. It has offered prison programs since 1999; there have been anecdotal reports of success and increased participation, but no formal studies of the program's effectiveness. Freedom Project provides introductory and advanced trainings in NVC and mindfulness practices such as NVC study groups, support groups for inmate facilitators, and multi-day theme-based workshops on anger, empathy/self-empathy, reconciliation, connecting couples, and parenting. Inmates' participation in the trainings is always voluntary, never obligatory (Freedom Project, 2009).
The purpose of the set of studies presented here is to explore the effects of the training offered by Freedom Project. The first study examines the relation between training and recidivism rates. The second study compares trained and untrained inmates in their use of nonviolent communication skills during simulated and challenging social interactions; self-reports of mindfulness, self-care, and anger management are also compared. These two studies are presented as a unit because together they illuminate the effects of Freedom Project training for society and for individuals. The conclusions are strengthened by including different modes of measurements, such as official records, behavioral observations, and self-report. The combined findings allow for the exploration of whether the training is associated with improvement during incarceration and upon release, as well as which specific skills and attitudes change.

General Training Method
Freedom Project's training of inmates is based on Marshall Rosenberg's nonviolent communication approach that has been detailed elsewhere (Leu, 2003;Rosenberg, 1999). The materials and specific activities used during the training have been compiled and published in the book Nonviolent Communication (NVC) Toolkit for Facilitators (Gill, Leu, & Morin, 2009).
Prisoners of the Monroe Prison Complex are offered 2-day workshops every third weekend of every month. In addition, every month there are three, 3-hr evening classes in both NVC and mindfulness. All classes are offered within the prison compound. Workshops are either introductory or organized around themes such as conflict, empathy, healing, or couples and family relationships; all workshops include mindfulness. Teaching styles vary from trainer to trainer and according to the workshop theme. Twice a year, specific mindfulness meditation workshops are offered as part of the theme-based workshop series.
Class sizes range from 15 to 25 participants, and workshop participant numbers range from 20 to 45. All training is done in groups and participation is always voluntary and may be repeated. Freedom Project training is one of several training options offered to inmates in the Washington State prison system. Prisoners who have received NVC basic training are given preference for workshop participation; some prisoners with long sentences have accumulated many hours of training. Freedom Project programs are open to all prisoners who choose to enroll. There are no early release or special treatment incentives offered for participation.
The program was not designed or conducted as a formal research study, and limitations imposed by administrative constraints and the need to adapt to differences among the various prison settings necessitated some modifications. The training circumstances themselves are not optimal: for example, the training rooms are usually small, cramped, hot in summer, cold in winter, and with windows that do not open.

Study 1
In the first study, we considered the relationship between Freedom Project training and recidivism. We compared the percentage of recidivism in trained inmates with the general recidivism rates in Washington State for the same period. We hypothesized that inmates who had received training would have a lower recidivism rate than those who were not trained. In addition, we proposed that the number of hours of training in NVC and mindfulness might predict whether trainees would reoffend or not when they reentered the community.

Method
Participants. Training data were collected on 1,315 inmates of the Monroe Correctional Complex in Monroe, Washington, including the Minimum Security Unit; the Twin Rivers Unit; and the Washington State Reformatory. Crimes leading to incarceration included sex offenses, drug crimes, burglary, armed robbery, assault, kidnapping, and murder. The Washington State Department of Corrections (DOC) supplied information about release and reoffenses for most of the sample inmates, except for 136 for whom data were not reported and 20 for whom some identifying information did not match the training data. In addition, 274 of the trained inmates were not yet released from custody; recidivism could not be studied for them. The final sample was composed of 885 inmates who were trained by Freedom Project volunteer staff; complete release and reoffense data were known on these inmates.
The 885 trainees included in this analysis were all men, who averaged 37.2 years of age (SD = 9.8, range = 57.9) on the date of their final training. Of the 885 trainees, 654 were White (74%), 178 were Black (20%), 27 were North American Indian (3%), 17 were Asian/Pacific Islanders (2%), 1 was Eskimo, 6 were of other ethnicities, and 2 were of unknown ethnicities. Comparable demographic data for the general prison population used as a comparison group were available for those released in 2007: White (63.6%), Black (18.6%), Hispanic (11.1%), North American Indian (3.9%), Asian/Pacific Islanders (2.5%), and Other/Unknown (0.3%; Evans, 2011). The trained group were more likely to be White and less likely to be Hispanic than the general population, but otherwise were of comparable ethnicities. The inmates were trained an average of 35.2 hr (SD = 66.1, range = 1,057.8). Those released as of 12/31/2008 (n = 667) had been released an average of 39.4 months (SD = 29.4, range = 118.6).
To compare recidivism rates of trained and untrained inmates, the data for all prisoners released in 2000 by the Washington State DOC were corrected by eliminating the Freedom Project trainees released in that same year (N = 30) from the totals. A chi-square analysis revealed a highly significant trend, with fewer Freedom Project trainees returning to prison than expected and more from the general population returning to prison than expected, χ 2 (1) = 84.2, p < .0001. Of the 697 trained inmates who remained released on December 31, 2008, 175 (23%) had been released for greater than 5 years, 76 (17%) had been released for between 4 and 5 years, 82 (13%) had been released for between 3 and 4 years, 89 (12%) had been released for between 2 and 3 years, 116 (11%) had been released for between 1 and 2 years, and 159 (25%) had been released for less than 1 year.
Because approximately 30% of the Freedom Project trainees were sex offenders, and sex offenders may have a lower rate of recidivism for any crime than nonsex offenders (Hanson & Bussière, 1998), the data were reanalyzed to determine if recidivism remained lower for the Freedom Project trainees than for the general population of prisoners once the sex offenders were eliminated from the analysis. First, the trainees were coded as sex offenders if any of the crimes for which they were incarcerated in a Washington State prison was a sex crime (including "Rape 1," "Rape 2," "Rape of a child 1," "Other violation child sex," "Nonviolent sex with child," "Indecent liberties," and "Sex Crime, other"). The 268 released trainees coded as sex offenders reoffended at a rate of 11.2%, compared with a recidivism rate of 25.6% for the nonsex offenders. The chi-square was then recalculated, eliminating the 268 sex offenders from the 885 released trainees and comparing the recidivism rates of the remaining 617 trainees. There was still a highly significant trend with fewer Freedom Project trainees returning to prison than expected and more from the general population returning to prison than expected, χ 2 (1) = 31.6, p < .0001.
A one-tailed t test assuming unequal variances was performed to determine if the number of hours of training differed between the subgroup of the 885 trained inmates who committed a reoffense that returned them to prison (n = 182) versus the subgroup who were still released on December 31, 2008 (n = 685). First, the training hours data were inspected for outliers, using the definition of outliers as scores that were greater than the mean plus three standard deviations for the subgroup. Five outliers were identified and removed for the subgroup that had been returned to prison, leaving data for 177 trainees. Sixteen outliers were identified and removed for the subgroup of those who were still released on December 31, 2008, leaving data for 669 trainees. Those who returned to prison had significantly fewer hours of training (M = 20.6, SD = 10) than those who were still released (M = 26.3, SD = 19.7), t(566) = 5.3, p < .0001. Because the training-hours data for both groups were positively skewed, a nonparametric test of differences was also calculated that translates the training hours data (again with outliers removed) to ranks for each group. Median training hours for the group who did not reoffend was 19 hr, for the group that reoffended 18.3 hr, Mann−Whitney U = 70,971.5, n 1 = 697, n 2 = 182, p = .013.

Discussion
The first hypothesis of Study 1 was that inmates who had received training would have a lower recidivism rate than those who were not trained. This hypothesis was supported. We chose to focus on recidivism data reported by Washington State for 2008 because we used 12/31/08 as the date by which trainees had to be released to study their reoffenses for this study. By eliminating the counts of trained prisoners from the 2008 recidivism data, we had a recidivism rate for prisoners who were not trained. This general prison population comparison group had fewer members who claimed a White ethnicity and more who identified as Hispanic than the Freedom Project group. Despite the demographic differences, both the percentages of those reoffending (37% vs. 21%) and the statistical analysis of frequencies led us to reject the null hypothesis that training made no difference in recidivism. Recidivism as a measure has been criticized. First, the associations between crime, arrest, and imprisonment vary (cited in Gehring, 2000); therefore, reimprisonment as a sole measure of an individual's successive criminal activities can be inaccurate. Race and socioeconomic status affect whether or not imprisonment occurs (Blair, Judd, & Chapleau, 2004). Racial profiling might have allowed the Hispanic members of the general prison population to be reimprisoned at a higher rate than the White members of that comparison group. Second, outside-of-state convictions are generally not incorporated into within-state rates of recidivism. Finally, different states and different entities within a state often use different definitions of recidivism (Evans, 2010;Gehring, 2000). In the Freedom Project study, the first two limitations apply; however, the same Washington State DOC office produced the recidivism data for the trained and untrained prisoner groups. While the proportion of sex offenders in the trained group was high (30%), even when that subgroup was removed from the trained group data, recidivism rates still differed significantly between the trained group and the state-reported data for the general population of prisoners. The second hypothesis that trainees who reoffended would have fewer hours of training than those who did not reoffend was also supported. The nonnormal shape of the training hours distributions led us to rely on a nonparametric approach to conclude that the distributions of training hour differed significantly between those trainees who reoffended and those who did not.

Study 2
Given the relationship between lower recidivism rates and Freedom Project participation, this second study considered what changes might have been brought about by the Freedom Project training in inmates. In other words, we were interested in knowing how the training worked and what effects it had on individual inmates.
We hypothesized that inmates who underwent the NVC and mindfulness training, in comparison with untrained inmates, would show increased mindfulness, increased use of nonviolent communication skills when asked to address difficult interpersonal interactions, increased self-compassion, and decreased anger.

Method
Participants' characteristics. This is a matched-pairs study that compared inmates trained by the Freedom Project with untrained inmates matched on key demographic variables.
The 26 participants were male prison inmates at the Monroe Correctional Complex in Monroe, Washington. At the time of their evaluation, their ages ranged from 24 to 66 years old. All participants were Caucasian. To ensure that the participants understood all the questions and instructions correctly, only inmates fluent in English were invited to participate. No inmate was ruled out due to participation in any other training program available in the prison. The crimes the inmates were convicted of included rape, rape of a child, murder, indecent liberties, and assault.
Sampling procedures. The study used matched-pair sampling to look at the relationship between Freedom project training, and mindfulness and anger. The Trained group consisted of 13 inmates selected because they had already undergone at least one full cycle of training in mindfulness and NVC with Freedom Project, and because they had agreed to take part in this study. The hours of training provided by the Freedom Project ranged from 100 to 1,591. The Untrained group consisted of inmates matched and paired with the Trained group by crime category and age. The Washington State DOC selected matched untrained inmates from its files. The selected men were invited by Freedom Project to volunteer for the study. When there was a pool of untrained inmates who matched the same trained participant, only one of the untrained men was included in the study. Untrained participants were randomly selected from the matched pool to form comparable matched pairs. The final sample was composed of 26 Caucasian male participants who were matched on age (range = 24 to 66 years old, or 42 years) and crime category (rape, rape of a child, murder, indecent liberties, or assault).
Procedure. This experiment was approved by the Washington State DOC's Institutional Review Board and Antioch University Seattle's Institutional Review Board. In accordance with the Institutional Review Boards' agreement, no identifiable data were provided back to the Washington State DOC regarding an individually identifiable performance, completion of participation, or if an inmate declined participation. Each inmate who showed up for the testing session received a thank you letter and a chocolate bar, regardless of their completion of the testing protocol. Some participants did not complete all the assessments, in which case the data from both members of the matched pair were eliminated from the study. In one case, there was a tape recorder malfunction, limiting the number of role-plays available for scoring and analysis. A total of 26 men took part in the study.
Fifty-five inmates were initially screened using the Symptom Checklist-90-Revised (SCL-90-R; Derogatis, 1983) to eliminate participants with psychotic features; none were excluded from the study based on that criterion. Each testing session took less than an hour and was conducted by trained volunteers. The paper and pencil questionnaires were completed by participants in a group assessment session; the role-plays were done individually and privately.
The SCL-90-R (Derogatis, 1983) is a self-report psychological symptom checklist. It is composed of 90 questions that measure psychological distress. The checklist is based on a 5-point Likert-type scale where a lower score indicates less symptom severity. The items are rated on a scale of 0 (not at all) to 5 (extremely).
The MAAS is a self-report 15-item questionnaire developed by Brown and Ryan (2003) that is based on a Likerttype scale where a higher score indicates increased mindfulness. Items are rated on a scale of 1 (almost always) to 6 (almost never). The MAAS has good convergent, discriminant, and construct validity; internal consistency across samples (explicit measures α = 0.85; state mindfulness α = 0.92); and excellent test−retest reliability (r = .81, p < .0001; Brown & Ryan, 2003). The MAAS indicates the presence or absence of attention to and awareness of what is occurring in the present; it does not necessarily indicate attributes such as acceptance, trust, empathy, or gratitude. Some sample MAAS items include "I find it difficult to stay focused on what's happening in the present" and "I do jobs or tasks automatically, without being aware of what I'm doing." The Self-Compassion Scale (Neff, 2003) measures the ability to recognize a broader human context and lessen selfjudgment. It consists of a 26-item self-report questionnaire based on a Likert-type scale where a higher score indicates more self-compassion. The items are rated from 1 (almost never) to 5 (almost always). The internal consistency of the Self-Compassion Scale equals .92; it has good construct, convergent, and discriminant validity; and it has a test−retest reliability of .93 (Neff, 2003). According to Neff's conceptualization and psychometric construction of the test, self-compassion implies mindfulness and the Self-Compassion test includes a Mindfulness subscale. The Mindfulness subscale particularly measures over-identification with what is occurring. Some examples of items from the Mindfulness subscale include "When something upsets me I try to keep my emotions in balance," "When I'm feeling down I try to approach my feelings with curiosity and openness," and "When I fail at something important to me I try to keep things in perspective." The BAAQ (Maiuro et al., 1987) includes items that address assault, indirect hostility, irritability, negativism, resentment, and verbal hostility. The BAAQ is a self-report 6-item questionnaire based on a 5-point Likert-type scale where a higher score indicates greater anger. The items are rated on a scale of 0 (extremely unlikely) to 4 (very likely). It has an internal consistency of .82 and reliability of .84. It provides a good general measure of anger and dyscontrol behavior (Maiuro et al., 1987). Examples of items include "When I really lose my temper I am capable of hitting or slapping someone" and "When I get mad I say threatening and nasty things." The role-play protocol is reported in Table 1 and scoring categories in Table 2. The protocols were adapted, with permission, from Rosenberg's Nonviolent Communication: A Language of Compassion (1999). Six interviewers were trained in how to instruct the participants on the task and introduce the role-plays. The interviewers worked one-onone with each participant. A training scenario and two scorable role-plays were serially presented to each participant. Participants were allowed to respond to the protocol questions at will. All the role-play responses were tape recorded, transcribed, and coded by trained coders blind to participants' group memberships. The coders were two graduate students trained in the scoring protocol via education about NVC concepts and scoring of mock responses until establishing an inter-rater reliability of 75%.
Based on the transcription, each phrase uttered by a participant was rated for the presence or absence of each of the four response criteria (Table 2); in other words, each phrase was rated dichotomously four times. The number of "yes" answers in each category represents the number of times the participants said a phrase that met that NVC evaluative criteria.

Results
A paired-sample t test for the SCL-R-90, which was used as a pre-screening test, revealed no significant differences between the groups (M trained = 52.2, SD = 39.6; M untrained = 59.8, SD = 42.8), t(24) = 0.47, p = .64, indicating similar levels of mental health symptomatology in both groups.
To look at the relationship among Freedom Project training and mindfulness and anger, we conducted a MANOVA  A. Identifying and expressing own feelings  Adjectives or adjectival clauses describing psychological or emotional state and  Use "I" or "we" or  An assessment or description of "how I feel" vs. "how I think." or  Don't follow "feel" with ^that, like, as if, I, you, he, she, they, it ^names or nouns ^passive voice or  Conditional feelings-would, might, may, could-followed by feeling word B. Taking responsibility for own feelings  "Because" must be followed by "I" or "we" and  Don't blame others and  Do not use "it" or "that" as the subject noun and  Not conditional feeling C. Note. If there is a "yes" and "no" within one category, the final decision is "yes." with one independent variable with two levels (training and control) and three dependent variables (MAAS, Mindfulness subscale from the Self-Compassion Scale, and BAAQ). Of significance were the omnibus test of significance (λ = .8086), F(1, 24) = 5.69, p = .02, and the effects of training on the Mindfulness subscale, F(1, 24) = 4.55, p = .04, and BAAQ, F(1, 24) = 4.30, p = .04; the effect on the MAAS was not significant, F(1, 24) = 0.09, p = .76. A Pearson Correlational Analysis of the BAAQ and the number of hours of training (excluding those who received no training) showed a significant correlation (r = −.368, p = .02). The more training hours completed, the lower the anger score. A histogram revealed that the number of hours of training was not linear with the Mindfulness subscale.
Role-play data analysis. A total of 90 phrases were uttered by all participants, which gave a grand total of 360 coded responses. Fifty-three out of the 64 responses (83%) coded as meeting a criterion of NVC communication were given by Freedom Project trained participants. Out of the 296 responses, 135 (46%) that were neutral or did not meet the NVC criteria were given by the trained participants. A chisquare analysis was highly significant, with more responses given by Freedom Project trainees meeting the NVC criteria, χ 2 (1) = 29.19, p < .01.
As most of the phrases rated as "Identifying and expressing own feelings" were also rated as "Taking responsibility for own feelings" and these two rating criteria might not have been independent, the statistical analysis was repeated, excluding the first category; in other words, a total of 90 responses (22 "yes" counts and 68 "no" counts) were dropped from the analysis, leaving 270 ratings. The chi-square analysis remained highly significant, with more responses given by Freedom Project trainees meeting the NVC criteria, χ 2 (1) = 16.46, p < .01.

Discussion
Although mindfulness training is already well established as an intervention for inmates, we found that different measures of mindfulness tapped different concepts and resulted in different effects. The Freedom Project training had a significant effect on the mindfulness component of the Self-Compassion scale, which focuses on maintaining a balanced view of a situation, in other words, on equanimity. No effects were found on the MAAS, which is a mindfulness measure more concentrated on attentional focus.
Substantial effects were noted on the BAAQ, indicating reduction in anger among inmates who had received Freedom Project training. The number of hours of training was significantly related to a reduction in anger.
According to the role-play analysis, inmates who had been trained in the Freedom Project approach were significantly more likely to use nonviolent communication skills than their untrained counterparts. Inmates who had received training were much more likely to identify and express their own feelings, to take responsibility for them, to be able to explain their own needs or make requests without imposing demands, and to express empathy. In other words, participants trained in the Freedom Project skills gave responses that were not commonplace among the general prison population.

Summary and Concluding Discussion
The aim of this article was to explore the effects of Freedom Project training. The first study investigated whether the training in mindfulness and nonviolent communication affected the recidivism rate. Results indicated significantly less recidivism among former inmates trained by Freedom Project than among those who were not. Even among trained men, those who were more likely to return to prison had significantly fewer hours of training. Extrapolating, if 37% of returnees are expected to reoffend, but only 21% of Freedom Project trainees did so, and if the cost of keeping a man in prison is U.S. $98 per day, training by Freedom Project has saved the state of Washington US$5,065,320.00 per year (Table 3).
We also explored specific behavioral and attitudinal outcomes of training that may prevent reoffending. The results of the second study found significant differences between the group of inmates trained by the Freedom Project and a matched pair group of inmates with no training. The results are based on a small, matched-pair sample and are preliminary. The study found that self-reports of inmates trained by Freedom Project showed significantly improved self-compassion and mindfulness when compared with matched controls as measured by the Self-Compassion Scale, but not as measured by the MAAS. The measurement of anger (BAAQ) showed improvement: the more hours of training, the less anger expressed by participants. A replication with a larger sample is highly recommended. Behavioral observations demonstrated very striking differences in the communications styles of the trained and untrained groups, with Freedom Project trained inmates being significantly better able to use the basic NVC skills of identifying and expressing their own feelings, taking responsibility for them, and explaining their own needs without imposing demands. In spite of the limitations of research based on pre-existent interventions, the evidence is strengthened by the use of a wide variety of measurement modes, including long-term follow-up effects, self-report measures, and behavioral observations. We believe the results from these studies are suggestive of behavioral improvement and promising of highly beneficial social impact of the training. We hope to see further research continued. Recommended further research would include a replication of this study with a larger population, placebo-controlled studies of recidivism after training, process research including correlates with physiological measures and analysis of the relationship between improvements in anger and nonviolent communication skills on the one hand, and incidence of disciplinary action while incarcerated on the other. In addition, the generalizability of the second study can be improved by including diverse cultural and ethnic participants. Random-assignment control trial studies are also recommended, because of the problem of self-selection into the training group.

Acknowledgments
Special thanks to all Freedom Project volunteers, particularly Jay Jackson and Dow Gordon; to the Department of Corrections employees, particularly Marjorie Petersen, Bruce Stegner, and David Daniels; to the volunteers from Antioch University Seattle, including Deborah Sweet; to Richard Horn for consultation on data analysis; and to Marshall Rosenberg and his publisher PuddleDancer Press, for permission for the adaptation of their material for this research.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) received no financial support for the research and/or authorship of this article.