Introduction

Globally, over a quarter of woman have experienced physical and/or sexual violence by a partner during their lifetime [1]. This violence can have serious consequences for women’s physical and mental health, the health and wellbeing of their children, and their participation in social and economic activity [2–4].

Recent years have seen a growth in research on violence against women (VAW), with the evidence base now showing that VAW is preventable through a range of interventions. One model that has been shown to be effective in multiple well-conducted evaluations is combined economic and social empowerment interventions for women [5–8].

The IMAGE intervention in South Africa, combining microfinance and group-based gender training sessions for women, was the first such intervention shown to reduce women’s experience of physical and sexual intimate partner violence (IPV) in a cluster randomised trial [9]. Since then, evidence from other contexts has supported the potential for similar interventions to reduce IPV [10, 11].

The rationale for interventions that combine economic and social empowerment components has long been that bigger and broader impacts can be achieved by simultaneously tackling multiple strands of structural disadvantage faced by women [9, 12]. At the same time as challenging inequitable gender attitudes and norms, building women’s confidence and fostering communication and conflict resolution skills, they also address women’s economic dependence on men. They thereby enhance women’s power and resources to be more discriminate in partner choice, negotiate the parameters of new or existing relationships, or leave abusive relationships [13]. There is now mounting evidence that economic interventions combined with gender transformative interventions are more consistently associated with reductions in IPV than are economic interventions alone [5, 14, 15]. The question, however, of whether gender transformative interventions alone can prevent IPV remains under-researched. Are gender training interventions sufficient in their own right to reduce IPV in contexts where women have limited access to and control over economic resources?

The MAISHA study, comprising two cluster randomised trials of a 10-session gender training intervention to prevent IPV, offers an opportunity to explore this question—one trial conducted among women in pre-established microfinance groups [16] and the other among women not receiving microfinance in Mwanza city, Tanzania [17]. When delivered to women in established microfinance groups (CRT01), the MAISHA intervention was associated with a reduction in women’s past-year experience of physical IPV (aOR 0.63, 95% CI 0.41–0.98) [18]. The same intervention, delivered to women in newly formed neighbourhood groups not receiving microfinance (CRT02) had no impact on past-year physical IPV (aOR 0.98, 95%CI 0.72–1.33) [19]. Attendance at MAISHA sessions was high in both trials, particularly so in CRT02 (82% attended 7 or more of the 10 sessions in CRT02, versus 67% in CRT01) [18, 19]. The question thus arises as to why the same gender training intervention led to a reduction in past-year physical IPV among women in pre-established microfinance groups, but not among women in newly formed groups not receiving microfinance.

This paper attempts to unpick how and among whom the MAISHA intervention worked to prevent physical IPV among women in pre-established microfinance groups (CRT01). We then seek to assess the extent to which those same pathways were present, absent or modified among women in CRT02.

First, we explore whether intervention effectiveness in each trial varied between different sub-groups of women. We discuss whether any observed differential effects between sub-groups in CRT01 might explain the null impacts on physical IPV in CRT02 given differences in the underlying characteristics of women enrolled in the two trials. Second, we identify potential mediators through which MAISHA may have impacted on physical IPV in CRT01, and assess alternative explanations as to why the intervention did not reduce past year physical IPV in CRT02: a) the relevant mediators didn’t change; b) mediators changed but did not translate into reductions in IPV; c) certain mediators changed in an unintended direction.

Materials and methods

Study setting and study design

The MAISHA study, comprising two cluster randomised trials (CRT01 and CRT02) was conducted in Mwanza city in north-western Tanzania. IPV is very common in Tanzania; overall, 44% of ever-married women report having experienced physical or sexual violence from their husband or partner [20]. Patriarchal norms are prevalent, over half of women and 40% of men reporting attitudes accepting of a man’s use of violence against his wife [20]. Drivers of IPV also include early marriage, with more than a third of women married before the age of 18. Young brides are more likely to drop out of school and to begin childbearing early, resulting in limited economic opportunities and subsequent economic dependence on male partners [21]. Overall, among women (aged 15–49 years), 62% report being currently married or living together, while 10% report being divorced or separated [20].

The MAISHA study was implemented by the Mwanza Intervention Trials Unit, the Tanzania National Institute for Medical Research, and the London School of Hygiene & Tropical Medicine in close collaboration with local leaders and community members. The trials are described in full elsewhere [18, 19].

The CRT01 trial recruited 66 existing microfinance loan groups located in three wards in Mwanza city. Of these, 33 groups (544 women) were randomised to receive the MAISHA gender training intervention, and 33 groups (505 women) were waitlisted to receive the intervention after trial completion (control). The CRT02 trial was also conducted in Mwanza city but in different wards where BRAC (the microfinance provider in CRT01) was known to be less active. The trial team worked in collaboration with local leaders to form 66 neighbourhood groups of women who were not engaged in a formal microfinance loan scheme. Of these, 33 groups (627 women) were randomised to receive the MAISHA gender training intervention, and 33 groups (638 women) were waitlisted (control). Details of the randomisation process are described elsewhere [16, 17]. CONSORT flow diagrams for the two trials are presented in Fig 1 (CRT01) and Fig 2 (CRT02).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. CONSORT flow diagram for CRT01.

https://doi.org/10.1371/journal.pgph.0002497.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. CONSORT flow diagram for CRT02.

https://doi.org/10.1371/journal.pgph.0002497.g002

The MAISHA intervention

The MAISHA curriculum was developed by EngenderHealth (an international non-profit organisation focusing on gender equity and reproductive health) in collaboration with the research team. It draws upon other published curricula, including Sisters for Life from the IMAGE study [9], and was designed to be participatory and reflective. The MAISHA intervention aimed to empower women, prevent IPV, and promote healthy relationships by increasing women’s knowledge and awareness (eg, of the consequences of normative attitudes to gender and IPV), developing their relationship skills (eg, communication and conflict resolution), and improving their social capital and peer support networks.

In both the CRT01 and CRT02 trials, the 10-session MAISHA intervention (outlined in Fig 3) was delivered to the 33 intervention arm groups on alternate weeks over a 20-week period. Each session lasted between 1.5 and 2 hours. Venues were selected to be convenient to participants, with sessions generally taking place at the group leader’s house or in a quiet area of a local café or guesthouse. Sessions were delivered by trained female facilitators, following the Wanawake Na Maisha curriculum. Facilitators were recruited and trained by the research team and EngenderHealth on curriculum content, materials and facilitation skills. Intervention delivery was monitored by the trial coordinator and senior research team members.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Outline of the Wanawake Na Maisha curriculum.

https://doi.org/10.1371/journal.pgph.0002497.g003

Conceptual framework

The pathways through which the MAISHA intervention is hypothesised to impact on physical IPV, and participant characteristics that might modify impact, are laid out in Fig 4. This conceptual framework was developed based on the MAISHA theory of change (S1 Fig), insights from the MAISHA qualitative data [27], broader IPV risk factor and prevention literature [6] and the availability of data items from the study questionnaires.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Conceptual framework—MAISHA pathways of impact on physical IPV.

https://doi.org/10.1371/journal.pgph.0002497.g004

Outcome

The outcome for this analysis is women’s reported experience of past year physical IPV at follow-up. This was measured among all ever-partnered women, including women who had (had) intimate partners with whom they did not cohabit. Details of questionnaire items used to construct this outcome are presented in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Questions used to construct physical IPV outcome.

https://doi.org/10.1371/journal.pgph.0002497.t001

Results

Response rates were high in both trials, with 89% of participants in each completing both baseline and follow-up interview. Mean group size was also similar in both trials, with a mean of 19.7 (SD = 5.03) women per group in CRT01 and 19.2 (SD = 4.7) in CRT02. Attendance was higher in CRT02 compared to CRT01 (82% attended 7 or more sessions in CRT02, versus 67% in CRT01). In both trials, intervention and control arm women were similar with respect to a range of baseline characteristics, as reported in the main trial papers [18, 19] (S2 Table). Although control arm women in CRT01 had slightly higher levels of education and lower levels of sexual IPV than intervention arm women, these imbalances were adjusted for in relevant analyses of intervention impact.

Characteristics of women in the two trials are presented in Table 2. On average, women in CRT01 were older than women in CRT02 (mean age 39.6 years versus 33.1 years). They were slightly less likely to be currently married/living with a man as if married (73% versus 81%), reflecting greater levels of separation and widowhood. Women in CRT01 were less likely than women in CRT02 to have no/incomplete primary education, though only a fifth of women in either trial were likely to have progressed beyond primary level education. Across both trials, the vast majority of women (95%) had at least one child for whom they were responsible.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Baseline individual- and relationship-level characteristics of women in the MAISHA trials, presented separately for women in CRT01 and CRT02.

https://doi.org/10.1371/journal.pgph.0002497.t002

Several indicators suggest that women in CRT01 had more financial resources and independence than women in CRT02. They were more likely to have personally earned money in the past 12 months (97% versus 80%), and to have a higher income if they worked. 34% of women in CRT01 reported that they would be able to look after their family on their (the woman’s) income alone, versus 20% of women in CRT02. Their households too were better off, with CRT01 women less likely than CRT02 women to report that their households had experienced financial hardship in the past year (47% versus 62%).

Though baseline attitudes on gender roles and the acceptability of IPV did not vary between women in the two trials, women in CRT01 were less likely than women in CRT02 to have experienced physical, sexual, emotional or economic IPV in the past 12 months. Further analysis shows this was largely accounted for by their older age (S3 Table).

Subgroup analyses

The subgroup analyses of intervention effects among CRT01 women suggest that intervention impacts on physical IPV differed according to study participants’ characteristics (Table 3). Effects appeared greater in women under 35 years than in women 35 years or older. They were also greater among people who had never experienced physical IPV at baseline (preventing new onset of IPV) than they were among those with a history of physical IPV at baseline (preventing continuation).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Adjusted odds ratios of intervention impact on past year physical IPV among different sub-groups of women in CRT01.

https://doi.org/10.1371/journal.pgph.0002497.t003

Though evidence from the LRT tests of interaction was weak, the point estimates of adjusted odds ratios suggest that effects may also have been stronger among women with above median income, and those who could look after their family on their income alone.

There was no evidence of any interactions in CRT02, with no intervention effects on physical IPV observed in the overall sample or any of the subgroups (S4 Table).

Intervention impacts on potential mediators

Woman’s attitudes and behaviours.

CRT01. The intervention was associated with reductions in the acceptability of IPV and the belief that a woman should be obliged to have sex with her husband (Table 4). Of these, attitudes accepting of IPV were strongly associated with increased risk of IPV (Fig 5), making reduced acceptability of IPV a potential mediator of intervention effect on physical IPV. There was less evidence that the intervention was associated with changes to broader gender attitudes around a woman’s obligation to obey her husband or a man’s role as primary provider for the family.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Association between individual- and group-level potential mediators and physical IPV risk.

https://doi.org/10.1371/journal.pgph.0002497.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Intervention impact on potential mediators (at follow-up), presented separately for CRT01 and CRT02.

https://doi.org/10.1371/journal.pgph.0002497.t004

In keeping with these results, the addition of attitudes accepting of IPV to the model of intervention impact on physical IPV led to a slight attenuation of effect (11% reduction in aOR), while the addition of the other attitudes did not do so (Table 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Intervention impact on women’s past year experience of physical IPV, with and without adjustment for potential mediators—CRT01.

https://doi.org/10.1371/journal.pgph.0002497.t005

CRT02. All attitudes relating to IPV and gender roles were impacted on in a progressive direction in the intervention group compared to the control group (Table 4). However, none of these attitudes were strongly associated with IPV risk (Fig 5).

Help-seeking/support to women experiencing IPV/social participation.

CRT01. Women in the intervention arm were more likely to say they would feel very comfortable seeking help if they themselves were experiencing IPV, and a higher percentage of people in their microfinance groups said that they would feel confident to intervene if someone they knew were experiencing IPV (Table 4). Only the latter, group confidence to intervene, was associated with lower risk of IPV for the respondent (Fig 5). Though this association was not statistically significant, when group confidence was added to the model of intervention impact on physical IPV there was a sizeable attenuation of effect (16% reduction in aOR) (Table 5).

The intervention does not appear to have impacted on women’s participation in meetings in the community (Table 4), which anyway was not associated with risk of physical IPV (Fig 5).

CRT02. Impacts on help seeking and group confidence to intervene were also observed in CRT02, though from a baseline of slightly lower confidence to seek help/support others. Again, only group confidence to intervene was associated with lower IPV risk, but as in CRT01 the association was not statistically significant (Table 4).

As with CRT01, the intervention did not impact on women’s levels of participation in meetings in the community, with lower overall participation among women in CRT02 compared to CRT01 (Table 4). Interestingly, there was some suggestion that this indicator was anyway associated with increased risk of IPV in this population (though the confidence interval included unity). Therefore, even if participation in meetings had increased in the intervention arm it would likely not have led to a decrease (and may even have led to an increase) in IPV risk among these women.

Relationship dynamics.

CRT01. The intervention was not associated with improvements in communication with the partner in CRT01, nor was it associated with an increase or decrease in arguing with the partner over a variety of topics related to gender roles/perceived transgression of gender roles (Table 4). Better communication is, however, associated with lower IPV risk, and arguing over gender roles associated with increased IPV risk, suggesting that intervention efficacy in preventing IPV could be enhanced if greater improvements in communication within the relationship were achieved. The intervention was associated with a slight (not statistically-significant) reduction in ‘any arguing’, and when this mediator was included in the model of intervention effect on IPV, there was a small attenuation in effect size (9% reduction in aOR) (Table 5).

Women in the intervention group were more likely than their control counterparts to be confident to assert an opinion different to their partner’s (Table 4), but this confidence was not associated with decreased IPV risk (Fig 6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Association between relationship-level mediators and physical IPV risk.

https://doi.org/10.1371/journal.pgph.0002497.g006

Women in the intervention arm were slightly more likely to report that their partner was often suspicious that they were unfaithful, though this difference was not statistically significant (Table 4). In turn, greater suspicion was associated with increased risk of IPV (Fig 6). When partner suspicion was accounted for in the model of intervention effect on physical IPV, the effect size increased suggesting that intervention impacts on IPV would have been greater in the absence of increased partner suspicion (Table 5).

CRT02. In CRT02, the intervention was similarly not associated with general improvements in communication with the partner. It was, however, associated with increased arguing about the woman not fulfilling her role, and weakly associated with increased arguing about the woman disobeying her partner and the man being disrespectful (Table 4). From a baseline of less frequent arguing about gender roles in CRT02, the intervention appears to have increased arguing among intervention arm women to similar levels to those observed in CRT01 (Table 4). In turn, arguing was strongly associated with increased IPV risk (Fig 6).

There was less evidence that women in the intervention group were more confident than control women to assert an opinion different to their partner’s (Table 4) (an indicator which was associated with reduced IPV risk in this trial) (Fig 6).

There was no strong evidence of intervention impact in either direction on partner’s suspicion that the woman was unfaithful.

Separation.

CRT01. In CRT01, there was no evidence that women in the intervention arm, partnered at baseline, were more likely than women in the control arm to separate from their partner during the course of the trial (an indicator anyway not associated with IPV risk).

CRT02. In CRT02, there was similarly no association observed between the intervention and relationship dissolution (Table 4). There was also no suggestion that separation was associated in either direction with IPV risk (Fig 6).

Discussion

The two MAISHA trials have offered us insights into the contexts in and mechanisms through which a gender training intervention can work to prevent IPV, as well as potential barriers to the effectiveness of such interventions. Among women receiving microfinance (CRT01), the MAISHA intervention reduced women’s experience of physical IPV, with greater impacts observed among younger women (<35 years versus 35+), those in non-violent relationships at baseline, and (though evidence was weaker) among women with greater financial independence. In CRT02, however, despite higher levels of attendance to the intervention, no impacts were seen on physical IPV either overall or within specific sub-groups. While positive impacts were seen on individual- and group-level potential mediators in both trials—reduced acceptability of IPV, confidence to seek help if experiencing IPV (hypothetical scenario), and group-level confidence to intervene in cases of IPV—relationship-level mediators such as communication within the relationship do not appear to have been positively impacted on in either trial. Indeed, in CRT02 it appears that arguments with partners over perceived transgressions of gender roles by women may actually have increased. Evidence also suggests that reduced acceptability of IPV may have been less strongly linked to reduced IPV risk among women in CRT02 than it was among women in CRT01. The intervention did not appear to lead to increased rates of separation in either trial.

The differential impacts seen among different sub-groups of women in CRT01 are instructive for the potential targeting of these types of interventions and the identification of potential barriers to intervention effectiveness. In CRT01, we observed greater impacts among younger women compared to older women, and among women with no prior experience of physical IPV at baseline compared to those who had already experienced physical IPV at baseline. These findings suggest that the intervention was more effective at preventing the new onset of physical IPV where it wasn’t previously occurring, than it was at preventing the continuation of violence in already violent relationships. It is perhaps logical that an intervention targeted to women only might be more effective at preventing women from entering into violent/less gender-equitable relationships, or at fostering women’s communication and negotiating skills that prevent currently non-violent relationships from escalating into conflict, than it is at changing the nature of relationships in which a man’s use of violence against his partner is already an established dynamic.

Findings from the sub-group analysis may also help explain why we didn’t observe impacts on physical IPV in CRT02. Although women in CRT02 were on average younger than women in CRT01, and we might therefore expect greater impacts in CRT02 than those observed in CRT01, women in CRT02 were also more likely to be experiencing IPV at baseline. This is the subgroup for whom the intervention was demonstrably less effective in CRT01.

In CRT01, we also observed stronger impacts among women with greater financial independence at baseline, in line with arguments that structural factors such as poverty and women’s financial dependence on men can trap women in relationships and act as barriers to the success of violence prevention initiatives [5, 29, 30]. Qualitative evidence from women enrolled in CRT01 also points towards synergies between the gender training and microfinance, with women discussing the importance of their financial security and independence in allowing them to stand up for themselves [27]. Women in CRT02 were less likely than women in CRT01 to have their own income, and those who did earned on average considerably less than their counterparts in CRT01. They were also less likely to report that they could look after the family on their income alone. It is important to note that it is not possible on the basis of our data to infer whether women in CRT01 were in a stronger economic position as a result of their participation in microfinance, or whether their participation in microfinance was driven or facilitated by greater pre-existing levels of financial resources/independence. Indeed, it is likely a bidirectional association. Regardless of the reason for this difference, it is plausible that greater poverty and/or lack of financial autonomy prevented women in CRT02 from using their MAISHA training to enact change at the relationship-level.

In both trials, the intervention impacted on several factors that we hypothesised to be potential pathways to IPV prevention. In particular, there were large impacts on attitudes. In both CRT01 and CRT02, the intervention was associated with significant reductions in the acceptability of IPV. However, these attitudes were not uniformly related to IPV risk in the two trials: Non-acceptability of IPV was linked to lower IPV risk in CRT01, thereby making it a potential mediator of intervention effect in CRT01; but not in CRT02, indicating that attitudinal change may not be sufficient to bring about reduced risk of IPV in this population of women. Since a woman’s attitudes can only impact indirectly on her own risk of experiencing IPV—through her choice of partner, her decisions on whether or not to remain with an abusive partner, or her motivation to negotiate revised relationship parameters and dynamics with her partner—it is not surprising that the association between attitudes and IPV might be modified by context. It is plausible that a shift to more progressive attitudes can only translate into changes in IPV risk, if a woman also has sufficient bargaining power within the relationship, engendered by other factors such as access to and autonomy over financial resources.

The intervention also impacted positively on broader gender attitudes in both trials, with even larger impacts observed in CRT02 than CRT01. These attitudes, however, were less strongly linked to reduced IPV risk in CRT01, than were attitudes explicitly related to the acceptability of IPV. They were also unrelated to IPV risk in CRT02. These findings are in line with those from the SASA! study, a trial of a community mobilisation VAW prevention intervention in Uganda, which found that attitudes on the acceptability of IPV were more influential mediators of intervention effect than were broader gender attitudes [31]. They support the idea that broader transformations in gender norms and attitudes may not be sufficient in themselves to prevent violence, if specific attitudes around the acceptability of violence against women are not also directly addressed and challenged. The finding that attitudes towards a woman’s right to refuse sex are not strongly linked to physical IPV risk, also suggests that sexual coercion within intimate relationships, though often reinforced by physical violence, is a distinct phenomenon that transcends relationships with and without physical violence [32–34].

In both trials, there was a weak association between the intervention and women’s reported comfort seeking help in the hypothetical scenario that they were experiencing IPV. However, comfort in seeking help was not associated with reduced IPV risk in either trial. It is possible that a bi-directional relationship between comfort seeking help and IPV risk may have led to this net zero association. Though help seeking might reduce IPV risk in the long term, those with experience of IPV might be more familiar and comfortable with the process of seeking help and respond accordingly in relation to the hypothetical scenario. Conversely, women with no experience of IPV might envisage more reticence in telling others. If this is the case, we may have underestimated the potential role that increased comfort seeking help might have as a mediator of longer-term intervention effect on IPV.

Confidence to intervene in the hypothetical case that another woman was experiencing IPV also increased, with higher group-level percentages of women confident to intervene in intervention groups compared to control groups (8.8% higher in intervention versus control groups in CRT01; 10.3% higher in intervention versus control groups in CRT02). Group-level confidence to intervene was associated with reduced IPV risk in both trials (not statistically significant), and when added to the model of intervention effect on physical IPV in CRT01 led to a moderate attenuation of effect. This evidence, in accordance with an evaluation of the scaled-up IMAGE intervention in South Africa [35], is suggestive of the role that improved social support networks may have in mediating intervention effect on individual women’s experiences of physical IPV.

Participation in community meetings did not increase in either trial. There was also evidence that participation in meetings was associated with increased rather than decreased IPV risk, particularly among control arm women in CRT02. This suggests that interventions with men and others in the community are necessary in order to increase partner- and community-level acceptance of women taking on more active roles outside of the house and within the community [36].

Though good communication between a woman and her partner, as indicated by their discussing each other’s daily activities and worries, was associated with reduced IPV risk in both trials, there was no evidence from either trial of improved communication among intervention arm women. This runs counter to qualitative evidence in which women who had participated in the intervention cited more positive communication with their partners and better conflict resolution skills [27]. It is possible that the lack of impact on communication observed here is artefact of the crude measure of communication used in this analysis (based on topic and frequency rather than quality of discussion). Interestingly, women’s confidence to assert an opinion different to their partner’s did increase, particularly in CRT01, in line with findings from the qualitative research in which women talked about greater confidence to express themselves about their own perceptions of situations and dynamics within the relationship that they wished to change [27]. Increased confidence was in turn associated with reduced IPV risk in CRT02 but not in CRT01.

Arguments between a woman and her partner were a strong risk factor for IPV in both trial populations. In neither trial did the intervention lead to a decrease in arguing, and in CRT02 there was some suggestion that arguments over perceived transgression of gender roles by the woman may actually have increased among women in the intervention arm. It is plausible that individual-level changes to a woman’s attitudes, confidence and behaviours—in the absence of any intervention with the male partners—has led to increased tension in relationships that from the outset were characterised by greater gender inequality [37]. This, in turn, could have cancelled out the effect of other pathways we might have expected to lead to reduced IPV. The importance of involving men in IPV prevention programming, including through couples-based interventions, has gained traction in recent years, and CRTs of several such interventions have demonstrated their effectiveness in preventing IPV [25, 26, 38–40]. Indashyikirwa, for example, a 21-session curriculum delivered to couples recruited from Village Savings and Loans Associations in Rwanda, led to substantial reductions in women’s reported experience of physical and/or sexual IPV, as well as reduced relationship conflict and improved communication more broadly [25].

In CRT01, the data also point to a possible increase in women reporting that their partners often suspected them of infidelity. In turn, the mediation analysis suggests that intervention impacts on physical IPV may have been even greater in the absence of this suspicion. Again, this unintended consequence (if real and not a spurious finding) could plausibly arise from the lack of any intervention with male partners—with greater intervention impacts possible if men’s attitudes linking infidelity to the acceptability of women refusing sex or seeking work outside the home were addressed. Indeed, in the SASA! trial of a community intervention that involved entire communities—both women and men—to address community-level norms underpinning high levels of VAW, levels of male partners’ suspicion over infidelity did fall. This fall came alongside progressive shifts in both women’s and men’s attitudes around IPV and women’s right to refuse sex. The intervention led to community-level reductions in IPV, with changes in men’s attitudes and reduced suspicion over infidelity emerging as potentially important mediators of this impact [31].

Finally, although we might hypothesise that interventions such as MAISHA could impact on violence by motivating and better equipping women to leave abusive relationships, MAISHA does not appear to have made women more likely to leave their partner in either trial. It is therefore through individual-level change and improvements to relationship dynamics, some of which have been described above and others of which have likely not been captured in this analysis, that MAISHA has reduced women’s risk of physical IPV in CRT01. A similar pattern has been observed in relation to other IPV prevention interventions such as IMAGE in South Africa [13] and SASA! In Uganda [31].

Conclusions

In summary, this study provides important insights to the field of IPV prevention research. The analysis not only sheds light on potential pathways of effect through which the MAISHA intervention has impacted on violence in CRT01, and that future violence prevention interventions might exploit, but also identifies contextual factors which might impede intervention effect and act as barriers to success at different points along the prevention pathway. Importantly we find that the MAISHA intervention is more effective in younger women with no history of IPV at the outset of the intervention, and potentially more effective among women with higher levels of financial independence. This might in part explain the lack of impact on physical IPV in CRT02.

These findings point to the importance of addressing structural factors such as poverty and women’s economic dependence on men that trap women in potentially abusive relationships. This might be achieved through incorporating economic strengthening components into IPV prevention interventions, or delivering IPV programming alongside pre-existing development activities which seek to improve women’s access to and control over financial resources. It is important to note that enhanced intervention efficacy must be balanced against the risks sometimes inherent in microfinance based IPV prevention interventions, where attendance can be adversely affected by participants’ difficulties keeping up with loan repayments and the demands of running a business [41, 42]. However, recent reviews have highlighted the effectiveness of IPV prevention interventions that combine economic strengthening (such as cash transfers, village savings and loans associations or microfinance) and gender transformative components [5, 6, 15]. As the global scale and reach of social protection schemes such as cash transfer programming increases from an already large base, this may provide expanding opportunities and platforms for scaled-up delivery of gender transformative IPV prevention interventions [15].

Finally, we see more impact on individual-level rather than relationship-level mediators in both trials, and attitudes less strongly linked to IPV risk in CRT02 compared to CRT01. Our findings also illustrate the potential for IPV prevention programmes to impact on specific relationship-level mediators in the unintended direction, for example increasing arguments between intimate partners. These perverse consequences with respect to specific mediators have the potential to counteract other positive effects of an intervention. An awareness of the potential for adverse consequences should be incorporated into intervention programming, and pre-emptive steps taken to ameliorate these risks. It should also inform larger strategic conversations around IPV prevention, such as the importance of involving and targeting men as well as women, including with couples-based and community-level interventions [6, 36]. This may be particularly important with respect to older women where (potentially violent) relationship dynamics may already be deeply entrenched. In order to shift such patterns, not only must male norms and behaviours be targeted, but community-level responses to VAW also strengthened. The SASA! community mobilisation intervention in Uganda, for example, through addressing risk and protective factors at multiple levels of the social ecology, was effective in promoting the cessation of violence within previously abusive relationships, in addition to preventing the new onset of abuse [31]. Through addressing these broader structural and socio-ecological factors, we can more effectively empower women and enhance IPV prevention interventions.

Supporting information

Acknowledgments

We wish to thank the MAISHA study participants for their time and commitment. We also thank the MAISHA research team for their contribution and dedication to implementing the MAISHA study, and the administration teams at the Mwanza Intervention Trials Unit and London School of Hygiene and Tropical Medicine for their support.