Social cohesion

There is no single, universally accepted definition of social cohesion; rather, scholars have articulated the concept in multiple ways, reflecting diverse theoretical frameworks and contextual interpretations. Generally, social cohesion refers to the strength and quality of relationships that bind individuals and groups within a society, enabling cooperation, trust and a shared sense of belonging. It encompasses both horizontal ties, i.e., the connections between individuals and communities, and vertical ties that link people to institutions and the state [6]. The literature broadly agrees on five interrelated domains of social cohesion: trust, sense of belonging, willingness to participate, willingness to help and acceptance of diversity [4,7].

Social cohesion is strongly linked to sustainable peace. Sustainable peace involves fostering meaningful connections among diverse social groups, promoting inclusive participation in governance processes and ensuring that traditionally marginalized groups have a voice in decision-making [8]. Evidence from both high-income countries and LMICs shows that the most successful social cohesion interventions are those that address both structural inequalities and inter-group relations [4,7,9,10]. For example, in contexts affected by displacement, policies that ensure equitable access to basic services, early inclusion of host communities in planning processes and joint economic development opportunities reduce exclusion and foster shared interests [11–13]. Infrastructure such as inclusive public spaces and shared service facilities play a vital role in building trust and a sense of belonging [14]. However, such policies and infrastructure are often lacking in LMIC contexts shaped by internal divisions and where political approaches to displaced populations are often exclusionary [15].

At the individual and communal level, interventions rooted in contact theory, which involves fostering positive, cooperative interactions between members of different social or identity groups, ideally under conditions of equality and a supportive institutional environment, have generally shown small positive effects on social cohesion [4]. These interventions include modalities such as sports teams, collaborative projects and facilitated dialogues that bring people from different groups together [4]. In the Southwest Asia and North Africa (SWANA) region, contact theory-based interventions implemented among conflict-affected communities have yielded mixed results. In Iraq, a mixed-religion soccer intervention demonstrated positive impacts on Christian players’ attitudes towards Muslim teammates but not on inter-group cohesion more broadly [16]. In Lebanon, participation in a national youth volunteering program had a positive impact on Lebanese youths’ sense of belonging and tolerance [17]. However, two other training-based interventions found, respectively, no impact on inter-group prejudice between Lebanese and Syrian refugees, and potentially some backlash among Lebanese participants [18], and no impact on Jordanian and Lebanese host communities’ outgroup bias but some small positive impacts among Syrian refugees [19]. Explanations for these divergent findings may include variation in program design and the short duration of most interventions, the different groups involved (in the context of Lebanon’s multiple social cleavages), and the different measures used to capture social cohesion.

Community center approaches

Interventions grounded in contact theory highlight the importance of structured opportunities for positive, cooperative interaction between groups. Community centers represent a promising but underexplored modality for operationalizing this approach in fragile and refugee-hosting contexts. By providing shared spaces that facilitate equal access, repeated interaction and collective activities, community centers offer an institutionalized infrastructure for fostering neighborhood-level social cohesion, particularly in marginalized urban areas [20].

Evidence from high-income settings demonstrates the role of community spaces in strengthening social ties and rebuilding fragmented social fabrics. The urban planning literature emphasizes that “space matters”: the provision of open and inclusive community spaces can increase social interaction, build social capital and enhance sustainability [20]. Similar findings have been reported in regeneration projects in Europe, including community center models in Switzerland designed to enhance participation and belonging in diverse neighborhoods [21] and in more recent work linking community centers to social entrepreneurship and collaborative innovation [22].

Community centers have also shown promise in displacement contexts. Refugee resettlement programs in the United States highlight the role of such spaces in creating cross-group engagement and supporting integration [21]. In the SWANA region, programming for Syrian refugees that provided spaces that facilitate cross-group engagement and shared service access, thereby operating in ways comparable to community centers, also suggests the potential of this approach to build sense of belonging and stronger inter-group ties [23,24].

Despite the promise of community centers as participatory, community-led spaces for building social cohesion, they remain largely absent from the literature in LMICs and are rarely subject to evaluation. To the best of our knowledge, no previous studies in the SWANA region have evaluated a community center-based approach to building social cohesion. The only systematic review of social cohesion interventions in LMICs to date also did not identify any such evaluations [4]. Given their adaptability and grounding in participatory, community-led approaches, this highlights an imperative gap in the literature regarding effective approaches for fostering social cohesion and stability in fragile contexts.

Setting

Tripoli, Lebanon’s second-largest city and the capital of the Northern Governorate, has long grappled with structural marginalization, deteriorating infrastructure and chronic underinvestment. These conditions have been exacerbated by the economic collapse Lebanon has experienced since 2019, with poverty rates in Tripoli exceeding 50% and unemployment surpassing 35% by some estimates [25,26]. The city also hosts large numbers of Syrian and Palestinian refugees, concentrated in impoverished urban neighborhoods. The resulting pressure on basic services and economic opportunities has intensified existing socio-economic vulnerabilities and intergroup tensions between host and refugee populations [25,26]. Within this fragile urban landscape, this study focuses on two adjacent but deeply divided neighborhoods: Bab al-Tabbaneh (predominantly Sunni) and Jabal Mohsen (predominantly Alawite), separated by the symbolic and contested Syria Street. These neighborhoods are nationally emblematic of entrenched sectarian fault lines and have experienced what observers have described as “the deadliest proxy war in Lebanon” [27]. Between these two neighborhoods, long-standing tensions escalated with the spillover of the Syrian conflict, culminating in recurrent rounds of armed clashes between 2008 and 2015 that left dozens dead, injured hundreds and displaced families on both sides [28,29]. Despite modest improvements in local relations and apparent signs of reduced clashes, Bab al-Tabbaneh and Jabal Mohsen remain heavily stigmatized and continue to stand as potent reminders of the fragility of Lebanon’s social cohesion and the lingering volatility of urban communal relations [29].

The women-led community center in Tripoli and the evaluated activities

Situated between the divided neighborhoods of Bab al-Tabbaneh and Jabal Mohsen, the women-led community center known as Beit el Salam was established in 2021 as a neutral and inclusive space for dialogue, collaboration and peacebuilding. The center was created with the support of Threads of Peace (commonly known as House of Peace, or HOPe), a local peace-building NGO in Lebanon. HOPe is a politically neutral, non-sectarian organization whose core model, the Social Peace Process, brings together community members from diverse backgrounds to engage in dialogue, build trust, jointly analyze challenges and design initiatives that address real needs. The process aims to empower participants as leaders of change and fosters resilience, inclusion and social cohesion within their communities.

The establishment of Beit Al Salam center resulted from the coming together of four previously independent, women-led initiatives established through HOPe’s Social Peace Process. These initiatives consist of two cooking groups, an upcycling project and a wellness care initiative. Although each had distinct goals, the idea of a shared space emerged as both a practical solution—allowing them to lower operational costs—and a symbolic act of unity. In its early stages, the center operated primarily as a shared platform through which the initiatives continued their own activities, but in a collective and more visible way. The simple act of women from both communities working side by side and being seen together in public service became a symbol of cooperation across divides. Over time, the center has evolved into a broader hub for women-led activities and local engagement. While the women themselves lead and manage these efforts, HOPe provides technical support.

The intervention evaluated in this study marked the first structured, multi-session program to be designed and delivered entirely by women from the center. Developed in collaboration with HOPe, the intervention brought participants together in fixed groups that met weekly over the course of seven sessions of about two and a half hours each. As a pilot under evaluation for potential scale-up, the program combined trainings by the center’s four initiatives with structured dialogue and group-building components. Each session was grounded in a shared agreement to ensure a safe, respectful environment, and opened with a group check-in to foster emotional connection. Sessions were facilitated by the volunteer women who had previously launched the community-led initiatives that gave rise to the center itself—an intentional effort to deepen their leadership and sustain their role as agents of change within their community. Women were not compensated for attending the sessions. Transportation costs to the community center were not covered by the program, but transportation was facilitated through a shared bus service with group pick-up points in the neighborhoods.

Study design

The study was implemented as a randomized controlled trial (RCT). The RCT protocol was registered on the American Economic Association RCT Registry (ID: AEARCTR-0013423). Community center volunteers conducted an intensive outreach campaign to recruit women for a series of sessions to be held at the center. They distributed digital flyers via the center’s social media platforms that provided information on the center’s activities and a link for interested women to fill a registration form. Volunteers also conducted in-person recruitment in neighborhoods, filling the recruitment form for those who expressed interest in the activities. The recruitment form collected contact and basic demographic information from potential participants, including information needed to screen for eligibility. The eligibility criteria for participation were that the individual be a woman, aged 18 or above, of Lebanese, Palestinian or Syrian nationality, living in the target communities and who had not previously attended activities at the community center. Recruitment was conducted between April 29, 2024 and July 31, 2024.

A total of 522 unique individuals filled out the recruitment form. After applying eligibility criteria, 362 women were identified as eligible to participate (Fig 1). These individuals were contacted via WhatsApp by trained women volunteers from the community center, who conducted an initial welcoming call. During the call, volunteers explained the program’s objectives, the nature of participation and logistical considerations such as the absence of compensation or transportation coverage. Following these calls, 206 women confirmed their interest in joining the sessions and agreed to be contacted regarding the evaluation study. This was considerably below our initial target sample size of 340 participants, based on sample size calculations using monitoring data that HOPe had collected from other workshops, albeit with different populations. Transportation emerged as a significant challenge to participation; although the program subsequently facilitated a shared bus with group pick-up points, it remained a barrier for many eligible participants who lived too far from the pick-up locations to join. The center’s initiatives were also concerned about their ability to offer the sessions to such a large number of women given space constraints.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Study flowchart.

https://doi.org/10.1371/journal.pone.0336198.g001

Randomization was conducted using Stata by taking a simple random sample, without replacement, of 50% of the de-identified participant list. A small number of participants who registered late were randomized separately using the same process. Women who were selected in the sample were assigned to the intervention group and invited to participate in the community center trainings. Women who were not selected were assigned to the control group and were informed that they would be offered a space in the training sessions in the fall of 2024. Due to delays related to the war in Lebanon in fall 2024, women in the control group were offered the intervention sessions in spring 2025. The randomization results were communicated to the participants after the baseline survey was completed so as not to bias survey responses.

Mobility is a challenge for women in the study context. After randomization was completed, several participants contacted the implementation team to inform them that they could only attend the sessions if they could commute to the center with a friend or relative who had been randomized to the other group. To avoid losing women from the center and the study, these requests were accommodated. Four women were therefore moved from their original control group assignment to the intervention group and two women from the intervention group to the control group (Fig 1).

Data collection

Data collection was conducted at baseline (pre-intervention) and endline (post-intervention) using structured, self-administered surveys completed via an online form. Based on HOPe’s previous experience and given the sensitive nature of the topic, self-administered questionnaires were preferred to reduce response bias based on social desirability and (perceived) characteristics of the interviewer. The surveys captured respondents’ basic sociodemographic information and outcomes related to social cohesion, civic engagement, self-efficacy and subjective wellbeing. Questions on intervention attendance and satisfaction were added for the intervention group at endline.

All recruited participants were contacted by the evaluation team by phone prior to being informed of the results of the randomization. Using an approved recruitment script and consent form, female undergraduate student volunteers from a local university obtained oral informed consent to participate in the study, emphasizing that participation was voluntary, confidential and would not affect participants’ involvement in the program. The volunteers documented whether each woman consented to participate in the evaluation study by writing ‘yes’ or ‘no’ on an Excel-based study tracking sheet that was overseen by one of the authors. Oral consent was approved for the study by the overseeing Institutional Review Board (IRB).

Participants who consented were subsequently sent the baseline survey link via WhatsApp. While the survey was self-administered, participants with limited literacy or digital access were supported by the same student volunteers, who conducted the survey verbally and recorded responses on their behalf following standardized protocols. A total of 202 participants completed the baseline survey between June and July 2024 (Fig 1). All surveys were collected in the Levantine dialect of Arabic on the KoBo Toolbox platform.

One month following completion of the intervention, a link to the endline survey was sent to all study participants via WhatsApp. A total of 177 participants completed the endline survey between October and November 2024 (Fig 1; 12.4% attrition). There was no evidence of selective attrition. Allocation to the intervention vs. control group was not significantly associated with study attrition, nor was nationality or neighborhood of residence (Table A1 in S1 Appendix). Of collected sociodemographic characteristics, only younger age was significantly associated with attrition. Given these results and the fact that the endline data collection was conducted during the 2024 war on Lebanon, attrition appears to have been largely random and related to the difficult circumstances in the country.

Outcomes

The study examined outcomes in three areas: social cohesion, civic engagement and wellbeing. To the extent possible, we employed validated outcome measures from existing literature. Other measures were developed or adapted to the context given the paucity of related measures that had been developed and tested in a Middle Eastern context. Due to the sensitivity of the topic, “prefer not to answer” options were added for all outcome questions. These responses were coded to missing at the analysis stage, leading to incomplete data for some outcomes.

Our primary outcome was generalized trust, one of the core components of social cohesion. We measured trust through a single-item question used, among other studies, in the global World Values Survey program: “Generally speaking, would you say that most people can be trusted or that you need to be very careful when dealing with people?” [30,31]. The option choices were “Most people can be trusted,” “You need to be very careful” and “Don’t know.” The small number of “Don’t know” (n = 6 at endline) responses were recoded as “You need to be very careful” for analysis. Additionally, a three-item social trust scale was constructed using questions about whether the respondent trusts 1) people she meets for the first time, 2) people from different religions and 3) people from different nationalities. Responses were collected on a 4-point Likert scale (1 = Do not trust at all, 2 = Do not trust very much, 3 = Trust somewhat, 4 = Trust completely) and averaged across the three items to yield a continuous score between one and four, with higher values indicating greater trust.

We also used the social relationships scale developed by Wilson and Secker [32] in the United States. The scale was chosen to capture the sense of belonging dimension of social cohesion. The eight-item scale assesses perceived inclusion, safety and social integration in the community. Items captured respondents’ perceived usefulness to the community, feelings of being appreciated by others, geographic mobility and cultural exposure, social and cultural participation and experiencing prejudice and safety concerns. Responses to each item were recorded on a 4-point scale ranging from Not at all = 1 to Yes, definitely = 4. We added a “prefer not to say” option that was recoded to missing at the analysis stage. To ensure consistent interpretation across items, we reverse coded negatively phrased items at the analysis stage so that higher values always indicate more positive perceptions of belonging. We then calculated a continuous, composite score by summing all items into a scale with a raw score range from 8–32. The reliability of the scale was acceptable, although not optimal, with an alpha of 0.71. To facilitate interpretation, we standardized the scale to have a mean of zero and a standard deviation of one so that the results can be interpreted in terms of change in standard deviations.

Our other measure of social cohesion assessed diversity of social ties. Respondents were asked how many close female friends they had and, of those friends, how many were of a different religion and how many were of a different nationality. As responses were clustered at the lower end of the distribution, we created a categorical outcome for each question coded as 0 friends, 1–2 friends, 3–4 friends and 5 or more friends.

We measured civic engagement, which overlaps with the social cohesion domain of willingness to help, using the Civic Engagement Scale developed by Doolittle and Faul [33] in the United States. The scale captures both attitudinal commitments—such as social responsibility and civic duty—and behavioral expressions of helping, volunteering and community participation. The scale includes two subscales measuring civic engagement attitudes and behaviors, respectively. The attitudes subscale was comprised of eight items capturing participants’ beliefs about social responsibility, civic duty, volunteering and involvement in the community. Responses were recorded on a 7-point Likert scale (Disagree = 1 to Agree = 7), with higher scores indicating stronger civic values. We calculated a continuous civic engagement attitudes score by summing all eight items (scale range 7–56). The subscale had acceptable reliability, with an alpha of 0.79. The civic engagement behaviors subscale included five of the original six items in the subscale, which assesses actual civic actions such as volunteering, helping others, donating to charity and engaging in community discussions. Responses were similarly recorded on a 7-point scale and summed to create a civic engagement behaviors score (range 5–35). The subscale also had acceptable reliability, with an alpha of 0.76. We then derived a total civic engagement score by summing the attitudes and behaviors subscales (range 12–91). Alpha of the overall scale in our study population was 0.82. We standardized all three scales to have a mean of zero and a standard deviation of one.

Our measure of subjective wellbeing was the WHO-5, a short, simple tool with demonstrated validity across a wide range of contexts and interventions [34]. The WHO-5 captures participants’ self-reported positive emotional states over the previous two weeks. Respondents rate the frequency of experiencing five states, such as “I have felt cheerful and in good spirits” on a 6-point scale from 0 (“At no time”) to 5 (“All of the time”). The five items were summed (range 0–25) and multiplied by four to yield a final score ranging from 0 to 100, with higher scores indicating better wellbeing [34].

Finally, we measured generalized self-efficacy (GSE), which captures an individual’s global belief in her ability to manage new or difficult tasks or to cope with adversity [35]. To facilitate respondent understanding of the self-efficacy questions, we adopted a 3-item General Self-Efficacy Scale (GSE-3) [36] despite the fact that this scale has not been previously well-validated in Arabic (unlike the longer, original GSE scale). The GSE-3 consists of three statements “I can rely on my own abilities in difficult situations,” “I am able to solve most problems on my own,” and “I can usually solve even challenging and complex tasks well.” Response choices were on a 5-point Likert scale ranging from 1 (“Does not apply at all”) to 5 (“Applies completely”). Responses across the three items were averaged to create a single, continuous measure with a range from one to five [36]. Alpha of the scale in our study population was 0.72.

Analysis

We present descriptive analysis of the characteristics of the intervention and control groups at baseline, demonstrating a balanced evaluation sample. We also present descriptive analysis of compliance with the randomized assignment, focusing on attendance of the community center sessions among women who were assigned to the intervention group. Given the lack of literature on social cohesion measures among our study population, we then present a descriptive analysis of the outcome measures at baseline. For all descriptive analyses, statistical significance tests are conducted using a chi-squared test for categorical variables and a t-test for continuous variables.

Our preferred estimate of the intervention impact is based on an intent-to-treat (ITT) analysis, i.e., we consider women’s assignment in the RCT regardless of whether they complied with that assignment. An ITT analysis is appropriate in RCTs, such as our study, in which compliance was imperfect because the ITT estimate avoids reintroducing selection bias based on who does and does not comply with their treatment assignment [37,38]. In the case of this study, the ITT estimate is also of interest because uptake of the intervention is key to its success [37]. In other words, whether people are willing to commit to an uncompensated, community-based program is an important aspect of whether such programs can be effective and sustainable. On the other hand, the ITT may yield an underestimate of the impact of the intervention if not all of the individuals assigned to the treatment actually receive it and/or individuals assigned to the control group inadvertently do receive (part of) the treatment. As a sensitivity analysis, we therefore also estimate the Local Average Treatment Effect (LATE), which estimates the impact of the intervention on those who comply with their assignment. The LATE estimator relies on stronger assumptions than the ITT, namely monotonicity and the exclusion restriction [37,38]. These assumptions are generally reasonable for this study.

Thanks to the randomized design, the ITT estimates are straightforward and entail a simple regression of the outcome on treatment status.

(1)

Where is the outcome of interest for individual Y and X is equal to 1 if the individual was in the intervention group and 0 otherwise. For continuous outcomes we estimate Ordinary Least Squares (OLS) regression models. For binary outcomes, we estimate logistic regression models and for ordinal outcomes (number of friends) ordinal logistic regression models. The LATE estimates use an instrumental variable approach, where receipt of treatment is instrumented by treatment assignment [37,38]. In this analysis, women who were assigned to the treatment but did not attend any intervention sessions are considered not treated and women who were assigned to the control group but reported visiting the center during the period between the evaluation baseline and endline were considered treated. All analyses were conducted using Stata 18.

Ethical approval

Ethical approval for the study was obtained from the American University of Beirut Institutional Review Board (Protocol SBS-2023–0360).

Characteristics of participants

The randomization was successful, with no significant differences in the sociodemographic characteristics of the intervention and control groups (Table 1). Participants were on average 38.5 years old and around two-thirds were between ages 35 and 64. Two-thirds of participants were Lebanese and the remainder were Syrian. No significant differences were found in the religious makeup of neighborhoods where participants resided (p = 0.23), although the percentage of women in the control group who lived in mixed neighborhoods was somewhat higher. Overall, 28.6% of participants lived in primarily Sunni neighborhoods, 25.6% in primarily Alawite neighborhoods and 45.7% in mixed neighborhoods, reflecting the success of the recruitment strategy in reaching women from different communities.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Balance between intervention and control groups at baseline (all observations at baseline).

https://doi.org/10.1371/journal.pone.0336198.t001

Three-quarters of participants were married and the large majority were not working. Just over half reported having heard of the community center prior to the recruitment drive for this intervention, again with no significant difference (p = 0.54) between women allocated to the intervention and control groups. Balance on all assessed characteristics of the intervention and control groups was maintained at endline despite attrition from the study (Table A2 in S1 Appendix).

Take-up of the intervention

Of the 92 women in the intervention group with complete survey observations, 70 (76.1%), attended at least one of the seven intervention sessions (Table 2). This leaves a notable proportion, 22 (24%), who never attended any session (of the women in the intervention group who did not complete the endline survey, 9 of 12 also had not attended any intervention sessions). While participants had been informed prior to the start of the intervention that no compensation or transportation support would be provided, when contacted to ask why they did not attend the sessions difficulty commuting to the center was one of the more common reasons. Other reasons included personal or family health issues, caregiving responsibilities and return to Syria. No sociodemographic characteristics were associated with never attending the sessions, although the estimates are very imprecise due to the small sample size (Table A3 in S1 Appendix).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Intervention take-up and compliance with treatment assignment.

https://doi.org/10.1371/journal.pone.0336198.t002

Looking more closely at session attendance, the data shows considerable variation in engagement levels among those who attended. Only about a third of women in the intervention group attended all sessions. On the other hand, among women assigned to the control group and who were successfully surveyed at endline, 16 (20.0%) reported having visited the community center since the baseline survey was run. As the intervention sessions were closed and center volunteers monitored attendance, it is likely that these visits were for other reasons, e.g., open community events or social gatherings. Thus, while these control group participants may have received some of the “treatment” of engaging with the community center, this was likely at a considerably lower intensity than among women who attended the intervention sessions.

Social cohesion, civic participation and wellbeing among study participants

At baseline, participants demonstrated low levels of generalized trust, with nearly 90% reporting that one must be too careful in dealing with people (Table 3). Scores on the other measures were somewhat higher. On the three-item social trust scale, the mean score was 2.5 (out of a maximum of four). On the social relationships scale, the mean score was 21.7, somewhat above the middle of the possible range of 8–32. Nearly all respondents reported having close female friends, with a mean of 4.9. On average, of these approximately five close friends, 1.9 were of a different religion and 1.6 of a different nationality. However, nearly half (49.2%) had no friends of a different religion and 42.3% had no friends of a different nationality, indicating considerable variation in the diversity of participants’ social networks.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Descriptive analysis of intervention outcomes at baseline.

https://doi.org/10.1371/journal.pone.0336198.t003

Turning to civic engagement, participants scored on average 43.5 on the attitudes subscale, well above the midpoint, and 20.8 on the behaviors subscale, right at the midpoint, suggesting more positive attitudes towards civic engagement than realized behaviors (Table 3). The average score on the overall scale was 64.3. The mean score on the WHO-5 was low, at 33.3, well below the midpoint. The average score on the GSE-3 was 3.5 out of five. There were no significant differences in any of the outcomes by treatment status, indicating balance between the intervention and control groups at baseline.

Impact of the intervention

Table 4 presents the intention to treat estimates of the intervention impact on measures of social cohesion. Intervention participants had 12% higher odds of saying that in general people can be trusted, but the result was not statistically significant. There was no impact on participants’ reported number of friends or on the three-item social trust scale. There was a positive impact on the social relationships scale, with participants in the intervention group scoring 0.60 standard deviations higher than those in the control group at endline (p < 0.001). There was a substantial amount of missing data on this variable, indicating that some respondents refused to answer some of the scale questions. Among women with complete observations, 17% did not answer at least one scale question at endline. In most of these cases, the respondent refused to answer 1−2 questions on the scale; there was only one respondent who did not answer any of the eight questions. Missingness on the scale at endline was not very highly correlated with missingness at baseline (correlation coefficient = 0.37), indicating that it was not always the same respondents who declined to answer all scale questions in each survey round. While the coefficient was somewhat attenuated (0.51–0.52 standard deviations; p < 0.001), the positive result remained robust to different recodings of the refused responses that kept these respondents in the analytical sample (Table A4 in S1 Appendix).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Intervention impacts on measures of social cohesion, intention to treat estimates (logistic regression, ordinal logistic regression and OLS regression results)^a.

https://doi.org/10.1371/journal.pone.0336198.t004

Turning to the other outcome measures, the intervention had a significant impact on civic engagement attitudes and behaviors (Table 5). Participants in the intervention group scored 0.33 standard deviations higher on the attitudes subscale at endline (p < 0.05) and 0.26 standard deviations higher on the behaviors subscale (p < 0.1), for an increase of 0.33 standard deviations (p < 0.05) on the overall civic engagement scale. No impacts were observed on generalized self-efficacy or subjective wellbeing.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Intervention impacts on measures of civic engagement and wellbeing, intention to treat estimates (OLS regression results) ^a,b.

https://doi.org/10.1371/journal.pone.0336198.t005

The local average treatment effect estimates were of slightly greater magnitude than the intention to treat estimates, but substantively identical in terms of which outcomes were and were not affected by the intervention (Table A5 in S1 Appendix). This is consistent with the effects being primarily driven by women in the treatment group who did attend the sessions, whereas women in the control group who visited the center likely had relatively little exposure to its approaches and activities.