Challenges to pre-migration interventions to prevent human trafficking: Results from a before-and-after learning assessment of training for prospective female migrants in Odisha, India

Background Awareness-raising and pre-migration training are popular strategies to prevent human trafficking. Programmatic theories assume that when prospective migrants are equipped with information about risks, they will make more-informed choices, ultimately resulting in safe migration. In 2016, India was estimated to have 8 million people in modern slavery, including those who migrate internally for work. Work in Freedom (WiF) was a community-based trafficking prevention intervention. This study evaluated WiF’s pre-migration knowledge-building activities for female migrants in Odisha to prevent future labour-related exploitation. Methods Pre- and post- training questionnaires were administered to women (N = 347) who participated in a two-day pre-migration training session. Descriptive analysis and unadjusted analyses (paired t-tests, McNemar’s tests, Wilcoxon signed ranks tests) examined differences in women’s knowledge scores before and after training. Adjusted analyses used mixed effects models to explore whether receiving information on workers’ rights or working away from home prior to the training was associated with changes in scores. Additionally, we used data from a household survey (N = 4,671) and survey of female migrants (N = 112) from a population sample in the same district to evaluate the intervention’s rationale and implementation strategy. Results Female participants were on average 37.3 years-old (SD 11) and most (67.9%) had no formal education. Only 11 participants (3.2%) had previous migration experience. Most participants (90.5%) had previously received information or advice on workers’ rights or working away from home. Compared to female migrants in the population, training participants were different in age, caste and religion. Awareness about migration risks, rights and collective bargaining was very low initially and remained low post-training, e.g. of 13 possible migration risks, before the training, participants named an average of 1.2 risks, which increased only slightly to 2.1 risks after the training (T(346) = -11.64, p<0.001). Changes were modest for attitudes about safe and risky migration practices, earnings and savings. Before the training, only 34 women (10.4%) considered migrating, which reduced to 25 women (7.7%) post-training (X2 = 1.88, p = 0.169)—consistent with the low prevalence (7% of households) of female migration locally. Women’s attitudes remained relatively fixed about the shame associated with paid domestic work. Survey data indicated focusing on domestic work did not correspond to regional migration trends, where women migrate primarily for construction or agriculture work. Conclusion The apparent low effectiveness of the WiF short-duration migration training may be linked to the assumption that individual changes in knowledge will lead to shifts in social norms. The narrow focus on such individual-level interventions may overestimate an individual’s agency. Findings indicate the importance of intervention development research to ensure activities are conducted in the right locations, target the right populations, and have relevant content. Absent intervention development research, this intervention suffered from operating in a site that had very few migrant women and a very small proportion migrating for domestic work—the focus of the training. To promote better development investments, interventions should be informed by local evidence and subjected to rigorous theory-based evaluation to ensure interventions achieve the most robust design to foster safe labour migration for women.

away from home. Compared to female migrants in the population, training participants were different in age, caste and religion. Awareness about migration risks, rights and collective bargaining was very low initially and remained low post-training, e.g. of 13 possible migration risks, before the training, participants named an average of 1.2 risks, which increased only slightly to 2.1 risks after the training (T(346) = -11.64, p<0.001). Changes were modest for attitudes about safe and risky migration practices, earnings and savings. Before the training, only 34 women (10.4%) considered migrating, which reduced to 25 women (7.7%) posttraining (X 2 = 1.88, p = 0.169)-consistent with the low prevalence (7% of households) of female migration locally. Women's attitudes remained relatively fixed about the shame associated with paid domestic work. Survey data indicated focusing on domestic work did not correspond to regional migration trends, where women migrate primarily for construction or agriculture work.

Conclusion
The apparent low effectiveness of the WiF short-duration migration training may be linked to the assumption that individual changes in knowledge will lead to shifts in social norms. The narrow focus on such individual-level interventions may overestimate an individual's agency. Findings indicate the importance of intervention development research to ensure activities are conducted in the right locations, target the right populations, and have relevant content. Absent intervention development research, this intervention suffered from operating in a site that had very few migrant women and a very small proportion migrating for domestic work-the focus of the training. To promote better development investments, interventions should be informed by local evidence and subjected to rigorous theory-based evaluation to ensure interventions achieve the most robust design to foster safe labour migration for women.

Migration awareness raising & effectiveness
Globally there are an estimated 40.3 million people are victims of modern slavery, with approximately 16 million people in forced labour in the private economy [1]. Modern slavery is an umbrella term for various situations where a person is exploited by others for various forms of gain, and includes bonded or forced labour, debt bondage, sexual exploitation, worst forms of child labour, domestic servitude and human trafficking [2].
Preventing modern slavery, a contested concept [3], has become a salient policy concern. Over the past decade, there has been substantial investment in awareness-raising and premigration training as a preferred strategy to prevent modern slavery, including human trafficking [4]. This programmatic approach is likely to be related to the limited implementation risks, large potential population reach, low cost and the political acceptability of awareness raising and training interventions [4]. Recently, 37 States endorsed the UN Call to Action to end forced labour, modern slavery and human trafficking, in which they commit to "develop and publish national strategies that raise awareness and improve understanding of the issues amongst the general public and amongst communities considered vulnerable to exploitation" [5]. Awareness-raising activities operate under the programmatic assumption that if district, located in south-eastern Odisha, which has not featured in broader debates in the state about labour migration [20]. Known primarily for male out-migration in textiles, diamond factories and ship-breaking industries in Gujarat, very little is known about women's outmigration from Ganjam.

Work in Freedom intervention
Work in Freedom (WiF) is a large multi-component intervention managed by the ILO and funded by the Department for International Development (DFID). WiF was implemented in India, Nepal and Bangladesh as source countries, and Lebanon and Jordan as destination countries, between 2013 to 2018. One component of the WiF intervention is the communitybased activities that aimed to prevent labour trafficking by enhancing women's autonomy, fostering adoption of 'safe migration' practices and assertion of migrant workers' rights [15]. The India component of WiF focussed on prospective in-country migrants only. The implementing partner (ILO) selected Ganjam as the intervention site.
The WiF community-based interventions in India were conducted via direct outreach to women in local households who were offered pre-migration training, which was delivered by peer educators and NGO staff. The community-based intervention aimed to empower women and adolescent girls to avoid what were considered to be 'risky' labour migration practices and labour arrangements, focusing on the paid domestic work sector [23]. Community training sessions were designed to shift attitudes and practices among prospective migrants, household members, and the wider community by delivering key messages about the importance of women's economic and social contributions, recognizing domestic work as a valid form of work, and highlighting risks that arise during the migration process [24].
Direct outreach by partner organizations to recruit participants for the training targeted prospective women workers and their family members (including husbands or parents), who were invited to participate in at least two 60-90 minute conversations in their home or other convenient location. Conversations focussed on safe versus risky migration practices, as well as broader issues around labour, workers' legal and human rights, migration and gender equality. Women did not need to express an interest in or history of migrating for work to be included in these "pre-decision" sessions. This was a deliberate strategy intended to minimise the stigma attached to programme participation, and to maximise the audience receiving programme messages. Direct outreach sessions were unstructured to permit flexible engagement with women and their families [24].
Women who expressed an interest in migrating for work during direct outreach sessions were invited to attend a two-day pre-migration training. Sessions aimed to support women to make informed decisions about whether, and how, to migrate for work, including modules on self-care, financial literacy, use of technology to ensure wellbeing, and safety in transit and at destination. Sessions were organised at the Gram Panchayat (village assembly) level and held once approximately 30 women had signed up per session (multiple sessions were held with different groups of approximately 30 women). Overall, a total of 347 participated in the training and completed pre and post training interviews. The training was implemented by the Self Employed Women's Association (SEWA) and AAINA, local NGOs with experience in gender, and labour mobilisation and organisation among women [24].

Data
We used data collected by AAINA, the Centre for Women's Development Studies and the London School of Hygiene and Tropical Medicine as part of the Study on the Work in Freedom Transnational (SWIFT) programme evaluation in Odisha, India. Pre and post training questionnaires to measure differences in attitudes, beliefs, plans, intentions and practices were administered by trained interviewers. Preferred selection criteria for interviewers included fluency in local dialects and Odiya language. The survey instrument was designed by the research team with one member (MD) working closely with local partners to appropriately phrase questions. The instrument was translated to Odiya language, and further revised through pilot testing and reviewed after back-translation to English. Vignettes included in the survey instrument were designed with local context in mind, e.g. asking participants what they would do if a recruitment agent came to their village recruiting for work in Mumbai, but that they had heard agents can deceive people. We interviewed all women participating in the WiF training in Ganjam district between January and March 2016. Women were interviewed immediately before and after the training using the structured survey instrument in S1 File. Women were interviewed in the Odiya language or in local dialects where requested.
Additionally, we used data from a representative household survey with female migrants in the same region to describe the context of implementation and WiF's target group. Female migrants were identified through a census of 4,671 households in 20 probabilistically sampled villages in two blocks of Ganjam District, Odisha between November 2015 and February 2017. We used a multistage sample design, with stratification and clustering.

Ethics
Informed written consent was obtained for all participants, whom were read participant information sheets by trained field interviewers in Odiya language. Where low literacy prevented participants from signing their name, consent was indicated via thumbprint, a common and acceptable practice in the study setting. Participants were assured that data would remain confidential and anonymised, that participation was entirely voluntary, and that they were free to refuse participation and could withdraw from the study at any time. Informed consent was emphasized as an ongoing process, with no adverse consequences for withdrawal from the study at any time.
Ethics approval was obtained from the London School of Hygiene and Tropical Medicine (LSHTM) (reference no. 8840) and the Centre for Women's Development Studies, Indian Council of Social Science Research.

Variables
Awareness scores. To develop 'awareness scores', women were asked a single question relevant to a construct. For example, awareness of the benefits of migrating for work was asked as follows: "What would you say are the main benefits of moving away from home to take up work somewhere else?" Women answered free form in their own words, i.e. interviewers did not read out item lists. Items that were included in each construct are shown in Table 1. We summed the number of items women mentioned in their response for each construct, before and after the training, and report mean scores in Table 3. In some instances, vignettes were used to prior to eliciting women's awareness on particular constructs (please see Fig 1 -further examples are listed in S1 File, which includes the overall survey instrument). Benefits of using the vignette approach include the potential to tailor questions to specific local contexts, and depersonalization that encourages participants to reflect beyond their own individual circumstances, which is particularly useful when discussing sensitive topics [25,26].
Attitudes. For questions reporting attitudes, a five-point Likert scale was used. Following a vignette (please see S1 File for examples), women were read a series of statements about the vignette and asked the extent to which they agreed or disagreed. Descriptive results for attitudes were recoded to a three-point scale (Agree, Neutral, Disagree-for presentation simplicity), and unadjusted and adjusted analyses used a five point scale to avoid loss of information (Table 4).

Analysis
Data cleaning and analysis were conducted in Stata 14. We restricted analysis to women who were surveyed both before and after the training (N = 347). Descriptive analyses and paired ttests were used to examine whether there were significant changes in awareness scores before and after the training. Unadjusted analyses for changes in binary outcomes were examined using McNemar's test, with Wilcoxon signed ranks tests used to examine changes in Likert scores before and after the training.
In adjusted analyses, we used mixed effects models to analyse the relationship between whether a participant had ever received information on worker's rights or working away from home, as a primary predictor of before and after training scores. We hypothesized that participants who had received prior information would report higher pre and post training scores,

PLOS ONE
Pre-post evaluation findings from pre-migration training for prospective female migrants in Odisha, India compared to participants who had not received prior information. Linear mixed effects models were used for continuous scores, mixed effects logistic regression models for binary outcomes, and mixed effects ordinal logistic regression models for Likert scores respectively [27]. Mixedeffects models account for the fact that multiple responses from the same participant are more similar (within subjects variation) than responses from other participants (between subjects variation) [27]. Models were adjusted for age and education levels. Coefficients and adjusted odds ratios for whether participants received prior information are reported in the final columns of Tables 3, 4. and 5.

Intervention context and WiF's target group: Female migrants in Ganjam, Odisha
Participants were selected from a community with predominantly male migration. Of 4,671 households surveyed in Ganjam District, Odisha, 7% reported female migration compared to 44% reporting male migration. Most of these migrant women were from scheduled castes (SC) (74.1%), and one-fifth (19.8%) were from communities that had been involved in conflicts. Christian women were disproportionally represented among female migrants, with Christian households 2.5 times more likely to have a history of female labour migration. The majority of women migrants interviewed (n = 112) made their migration decisions with the input of others (57.4%), whereas 12.2% decided for themselves. Only 4.3% did not have family consent to migrate. The most important reason for migration was chronic or temporary inability to meet basic needs (73.9%). Almost all migrant women (80.9%) migrated to work in urban destinations. Women were employed in many different sectors, with the most common including construction (25.3%), domestic work (24%) and agriculture (13.9%). Among the migrant women, 4.35% reported forced labour (95%CI 1.5, 10.3)-as assessed by the ILO indicators.
Women who had migrated perceived they developed new skills (63.4%) and gained increased confidence (17.0%). However, most women had mixed feelings about their migration-related gains and losses. For example, approximately one in five (18.9%) believed that migration had both enhanced their skills but reduced their freedom.

Recruitment and reach of pre-departure training: Participant characteristics
Women and adolescent girls who participated in training were between the ages of 16 and 75, with a mean age of 37.3 years (SD 11) ( Table 2). The majority of women (76.0%) were currently married with an average of 2.7 children (SD 2.5) at the time of the study. Half of the women (53.5%) were from castes designated by the Indian government as Other Backward Castes (OBC) while 23.8% were from castes identified as scheduled castes [28]. All participants (100.0%) were Hindu. The majority (67.9%) had no formal education and over half (55.4%) could not read or write any language. One-fifth (18.7%) of participants could read and write Odiya. Women and girls were mainly agricultural labourers (38.4%) or responsible for household chores (27.1%). Just 11 participants (3.2%) had prior experience of migration, while among married women, 42.3% had husbands who had previously migrated. The median household income among participants was 36,000 Indian Rupees/year (Median Average Duration (MAD) 16,000) (USD 474/year, (MAD) USD 211), mainly from work in casual or daylabour in non-agricultural sectors (45.6%). The majority of women (82.4%) reported that their household had no debt. Most households did not own any land (71.0%) or hire in labour

Awareness and attitudes towards migration risks and opportunities
While unadjusted analyses indicate that there were statistically significant differences in awareness scores before and after the training, very few items were endorsed for each construct (shown in Table 1) at each time point. Changes in scores at pre-and post-training were not practically significant. For example, participants could name 1.2 (SD 0.9) of the main risks of moving away for work before the training, and after the training they could name an average of 2.1 (SD 1.1) (T(346) = -11.64, p<0.001, unadjusted) of 13 possible dangers (Table 3). Similarly, before the training, participants could cite 2.1 (0.9) of 14 types of information that a migrant would need before moving away, and, after the session, they could describe an average of 3.0 (SD 1.2) (T(346) = -12.32, p<0.001, unadjusted) ( Table 3). Before the training, participants said that females should be aged 20.4 years old (SD 4.0) before moving away for work, a response they changed to 19.5 years old (SD 2.6) after the training (T(315) = 3.87, p<0.001, unadjusted). When participants were asked to comment on whether the character in the vignette should accept an advance on her wages, before the training, 58.2% of participants thought she should, compared to after the training when 69.3% thought she should (McNemar X 2 = 10.98, p<0.001, unadjusted) ( Table 3).
In adjusted analyses, having information on migration before the session was associated with increased awareness of the main benefits of moving away for work (β = 0.37, p<0.001) and awareness of the main dangers or risks (β = 0.49, p = 0.001). There was a marginal relationship between having prior information and accepting an advance on wages (AOR 1.5, CI: 0.71-3.14).
In contrast to their awareness scores, participants had greater percentage changes in attitudes towards migration risks and opportunities after the training (Table 4). For example, pretraining, 40.2% of participants agreed that "if someone known to me offers to help me move away to find work, I can be sure they will not abuse or exploit me" compared to 33.0% afterward, indicating a 7.2% reduction in agreement (z = -2.96, p = 0.003, unadjusted) ( Table 4). In adjusted analyses, there was a marginal positive relationship between having prior information and being sure that they would not be abused or exploited if someone offered to help them (AOR 1.42, CI: 0.79-2.54). Before training, 87.9% agreed that "Before departing, it is a good idea to check if there is anyone from this village or GP (doctor) at the destination, and to take their contact details", compared to 96.4% after the training, indicating a 8.5% increase in agreement (z = -5.62, p<0.001, unadjusted) ( Table 4).

Awareness and attitudes towards earnings and savings
There were significant changes in attitudes towards savings in four of seven statements. Pretraining, 64.4% of participants disagreed that "Madhuri [character in the vignette] cannot open a savings account without her employer's permission" compared to 78.0% afterward, indicating a 13.6% increase in disagreement (z = -3.16, p = 0.001, unadjusted) ( Table 4). There was little change in agreement with the statement "If Madhuri accepts advances on her wages from a middleman or employer, she risks being trapped or cheated", with 53.3% agreeing before and 55.2% agreeing afterward (z = -0.30, p = 0.757, unadjusted).
In adjusted analyses, there was a marginal positive relationship between having prior information and; sharing pass book and ATM codes with employers when opening a bank account (AOR 1.50, CI: 0.82-2.74); opening an account before migration to access savings (AOR 1.54, CI: 0.76-3.51), and; it being safer and less costly to send earnings home by bank transfer than

PLOS ONE
Pre-post evaluation findings from pre-migration training for prospective female migrants in Odisha, India

Attitudes towards safe and risky migration practices
If someone known to me offers to help me move away to find work, I can be sure they will not abuse or exploit me �

Awareness and attitudes towards domestic work, worker's rights and collective bargaining
There was little awareness of rights of domestic workers before and after the training, with an average of 2.1 (SD 0.9) rights endorsed before training and 2.8 (SD 1.3) rights endorsed afterward (T(346) = -8.89, p<0.001, unadjusted) (Table 3), of a potential 11 rights that could be endorsed (Table 1). Similarly, participants had very low awareness of reasons to join a domestic workers association, with just 0.1 (SD 0.5) reason endorsed before the training and 1 (SD 1.1) reason endorsed after the training (T(346) = -13.58, p<0.001, unadjusted) ( Table 3).
For attitudes towards women's work, there was little change in agreement with the statement "Woman's work is not as important as men's work", with 47.3% agreeing before and 44.9% agreeing after the training (z = -0.83, p = 0.405, unadjusted) ( Table 4). There was little change in attitudes towards paid domestic work, which were broadly negative. For example,  nearly half of participants agreed with the statement "Sita should feel ashamed to do paid domestic work in someone else's home" both before (43.6%) and after (46.7%) the training (z = 0.95, p = 0.341, unadjusted). Similarly, 44.4% agreed before and 45.3% after the training that "Paid domestic workers are servants not workers" (z = -0.35, p = 0.722, unadjusted). Changes in attitudes were not practically or statistically significant for these two statements.
In adjusted analyses of awareness and attitudes towards domestic work, worker's rights and collective bargaining, having received information prior to the training was not practically or statistically significant.

Intentions to apply to welfare schemes and to migrate
Awareness of livelihood and welfare schemes (or programs) was comparatively better than for other constructs, with participants being aware of an average of 4.3 schemes before (SD 1.3) and 5.2 schemes after (SD 1.7) the training (T(346) = -8.85, p<0.001, unadjusted) ( Table 3) of a possible 9 schemes (Table 2). Participants planned to apply for an average of 2.1 livelihood and welfare schemes (SD 1.5) before the training, and 3.1 schemes afterward (SD 1.5) (T(346) = -12.81, p<0.001, unadjusted) (Table 3), i.e. they planned to apply for one additional scheme on average.
Intention to migrate was very low and remained low (decreasing slightly) after the training. Just 34 participants (10.4%) had thoughts of moving away from the village to work before the training, compared to 25 participants afterward (7.7%) (McNemar X 2 = 1.88, p = 0.169,
Among participants who planned to move for work, half (47.0%) cited seeking better paid employment as a reason, compared to a third (32.0%) after the training. After the training, 8 participants (32.0%) also cited seeking better quality employment as a reason to move for work. Among participants who did not have thoughts of moving away for work, before the training, 38.0% cited that there was no one to care for children as the reason, compared to 43.5% afterward (Table 5). One-fifth (22.6% pre and 22.2% post) cited that spouses would not give permission to migrate for work, while one in six (16.1% pre and 15.0% post) reported that another family member would not give permission to migrate for work.

Impressions of training delivery and content
The majority of women reported positive impressions of the training session, for example, 98.0% enjoyed taking part in the training, while 94.1% felt that the information received was relevant to themselves and family members ( Table 6). Almost all women (98.8%) felt that they followed and understood the information received, while 94.1% would recommend the training to other women. Slightly higher proportions of women reported learning new information about risks and dangers of migration (93.2%) and safe migration (95.3%) compared to new information on worker's rights (84.8%), women's rights (85.6%) and good health (85.3%) that they didn't know before (Table 6).

Discussion
To place this intervention evaluation in context, it is important to acknowledge the relatively low prevalence of households with a history of female migration in Ganjam. The current Table 6. Impressions of training delivery and content among women participating in WiF pre-orientation training (N = 347).

Impressions of orientation delivery and content Agree N (%) Neutral N (%) Disagree N (%) Don't know N (%)
i scarcity of female migration was subsequently reflected in the small percentage of training participants who intended to migrate. These low migration intentions suggest that the intervention design would have benefited from drawing on the secondary data analysis or conducting formative research to refine the intervention's geographical focus and inform participant recruitment strategy and programme content, before the roll-out of the activities. For example, the training could have been designed more specifically to recognise the severe economic distress that affects female migrants in Ganjam and the ways they engage in short term circulatory migration and receive casual wages. Moreover, the training content to promote protection could have built on the evidence showing that women often migrate jointly with others. This pattern is in line with the migration literature which posits that migration expectations usually operate in the context of social norms and gender roles, and that migration decisions and behaviours are largely influenced by families and local patterns [18,29,30]. In this sense, the apparent low knowledge gain of the WiF short-duration migration training may be linked to the erroneous assumption that individual changes in knowledge will lead to shifts in social norms or can overcome current norms. Community-based interventions that incorporate local actors are probably better-placed to shift social norms than individuallevel training by organisations external to targeted communities.
The household data also indicate marked differences between the WiF training participants and the local population of female migrants. For example, while three in four migrants were from scheduled castes, only one-fourth of training participants were from these castes. At the same time, despite the fact that Christian households were more than twice as likely to have female migrants, all women in the training were from Hindu households. These findings suggest that the most likely prospective migrant women were either not recruited into the programme or they were not attracted by the invitations. Future initiatives should consider the target groups' characteristics and dispositions in the design of the intervention and its implementation strategies.
Findings from this study simultaneously raise some broader questions about pre-migration training itself and whether there is sufficient value in investing in information and awareness-raising sessions that prospective migrants may not be able to use in the context of structural drivers of labour exploitation. It is worth asking whether this set of new knowledge on rights and migration processes, while intrinsically valuable to the women, will, in reality, provide protection against exploitation in the face of the enormous power imbalance between migrant women workers and employers. The narrow focus on individual-level interventions may overestimate an individual's agency in face of very real constraints of poverty, debt and family obligations, the complexity and multi-locality of recruitment networks, official corruption and employers impunity [2]. Women may have minimal levels of control over household decisions or migration decision making with agents and when they reach employers, their welfare and wellbeing is often entirely dependent on them. Employers' rights will generally dominate the rights of domestic workers because of social norms and weak implementation of legislation, where it exists, which means that domestic workers will have few avenues to assert new knowledge in a power relationship that widens as women progress along the path of migration [7].
Will women be able to apply their new perspectives to protect themselves against labour exploitation-which was the primary aim of the intervention? If, ultimately, pre-migration training is deemed sufficiently valuable by other studies, then our findings insist that foundational research and adaptive programming evaluation are needed, before substantial investments are made in these types of training events. Pre-departure orientation alone does little to address cultural norms and expectations, structural problems of limited legislation for workers' rights and even weaker enforcement. Coupled with what little rigorous evidence we have about the effectiveness of pre-departure training, the utility of such programmes should be debated.
It is worrying that women's feelings of shame about undertaking paid domestic work and the value associated with domestic work remained relatively fixed even after hearing training messages to the contrary. Yet, to a certain extent, there were increases in women's attitudes about domestic workers being worthy of respect and that domestic workers should have the same rights as other workers. As noted in the recommendations from the qualitative evaluation of the WiF programme in Bangladesh, in the design of pre-departure training, care must be taken in the promotion of abstract "rights" that women are unlikely to be able to assert [15]. Even where women's awareness of rights has increased, their inability to enforce them can be psychologically painful and damaging [15].
This study was not intended to assess the training techniques, so we are unable to attribute the weak learning levels to a particular aspect of the programme. However, what is clear is that women did not absorb the key messages. For instance, analyses of pre-and post-learning showed that awareness scores (across all domains) were very low before and remained low even after the training. There were greater changes in attitudes towards women's work but still very modest changes in attitudes about migration practices, paid domestic work and earnings and savings. Ultimately, very few women increased their awareness of migration risks and opportunities, worker's rights and collective bargaining. And, these only improved very slightly after the training. Even women's awareness of their rights changed little.
Despite participants' enjoying the sessions and feeling that they understood the information they received, findings show that the key messages were not absorbed during the training. Clearly, evaluations should not be based solely on how participants "felt" about training content, as is so often the case in donor-funded evaluations. Objective measures of whether information received has been absorbed should be included. Moreover, future evaluations should assess if and how women utilise this information. While pre-departure training programmes are often well-received by participants [14] and liked by donors, this simple pre-and post-session learning assessment suggests that challenges remain for the uptake of training messages that must be addressed. Our findings indicate that even after women are provided information about migration-related risks and suggestions about safe practices, women were unlikely to recall or feel able to apply key messages.
For future investments to support prospective migrant women, intervention-focused research should be conducted to identify, for example, what information is prioritised by women, what teaching techniques will be most effective for which populations, and what skill sets are needed by trainers. Future intervention development research should test the effects of different techniques using, for instance, cognitive interviews with participants to assess their acceptance and uptake of the training format and content. And, inquiries should be made about who might be appropriate training staff. For example, findings from the WiF evaluation in Bangladesh highlighted the substantial social and economic gap between the trainers and participants, which created significant learning challenges. For instance, during the sessions, when returnee migrants in Bangladesh wanted to discuss the violence and abuse that they encountered abroad, these reports made trainers feel ill-at-ease and so these discussions were supressed [15].
After the training, fewer women reported intention to migrate-which some might believe is a positive outcome, although this was not the intention of the intervention. Feedback from the local research team suggests a strong ceiling on the migration of women by family members which was internalized as a gender norm "at the cost of starvation". It is also possible that women were dissuaded by descriptions of risks they had not previously considered. Informal feedback from partner organizations suggests that almost all trainees had not ventured beyond their block or village, with many unable to imagine migrating beyond their home base to a modern city. Unsurprisingly, women's perceptions of the constraints they faced in their own households did not change over the course of the training. Women reported that they would remain burdened with childcare and household duties, and spouses or family members still would not grant permission to migrate. Such types of change will be difficult to achieve in the short term, as they are often embedded in the wider socio-cultural norms. If future programmes aim to rebalance household roles and equity, greater consideration should be given to activities aimed at shifting social norms among male household members [31] or the wider community-however, these activities should not necessarily be conceived of as anti-trafficking or safer migration interventions. Interventions targeting wider socio-cultural norms should arguably be financed by better resourced, overarching development programs, rather than by limited donor funds for anti-trafficking or safer migration interventions, which ought to be more targeted at intervention points along the migration pathway where the structural challenges need to be confronted.

Limitations
Pre-post surveys cannot indicate how new knowledge will be applied, so findings are limited to what women learned and recalled immediately after the session. The pre-post single group study design is a non-experimental design without a control group, so changes in scores cannot be causally attributed to the intervention.
The very low prevalence of previous migration among women participants in the predeparture training (N = 11) meant that we could not compare differences in awareness and attitudes between previous migrants and women with no migration experience. In the Nepal WiF prospective migrants survey, returnee women were no better informed than first-time prospective migrants on key aspects of living and working conditions, or on their rights and responsibilities as migrant workers [32]. Returnees were better positioned to share advice on practical and emotional aspects of migration and working abroad, and that may be the case with returnees in Ganjam as well. No upper age limit meant that some women who were very unlikely to ever migrate (i.e. 27.2% of participants were aged 45 or older) were included in the training. A targeted focus on women who had a definitive plan for migration in the short term, or who already had a job offer, may have yielded better evaluation results. However, the likelihood that women would disclose this information given the potential stigma associated with women's labour migration is unclear. Alternative strategies to encourage such discussions are needed. The very low prevalence of previous migration in our findings reflect the general pattern of low female outmigration in Ganjam district indicated by the household survey, where just 9.3% of households had a female migrant [20]. Social stigma associated with female outmigration may have affected reporting in either the household study or the current evaluation findings.
This study was designed to measure the immediate effect of the training among participants and not the long term impact of the intervention given the limitation of a single group prepost design [33]. However, given that short-term effects of the training were virtually absent, it seems likely that even these small knowledge gains would disappear in the long-term. More importantly, WiF's rationale was: trainings will lead to increased knowledge and empowerment, which will change migration behaviour, which will protect women from human trafficking. However, the evidence from our study did not even support the first link in this assumed causal pathway. This suggests that learning and adaptation are needed before we can recommend the use of empirical design to evaluate the intervention.
Program evaluation data can be useful to improve intervention design and implementation, including course correction in the face of adverse consequences. Practical constraints of program evaluation data include limited data availability, and information on intervention implementation when there is no accompanying process evaluation, as was the case in the present study [34]. A process evaluation may have enabled us to understand whether the training failed because of poor implementation, program theory (individual focus) or a combination of both of these factors.

Conclusion
Marginal shifts in awareness in this study raise questions about the utility of pre-departure orientations. Future programming requires process evaluations to examine the strengths and weaknesses of implementation and associations between programme delivery and learning outcomes-including longer-term application of knowledge-which will also reflect on the intervention's underlying theory, i.e., pre-migration knowledge is protective against trafficking. While training activities are politically expedient because they are typically easier to implement and calculate outcomes among attendees (e.g., versus interventions to shift structural drivers of gender-inequality, enforcement of worker's rights, occupational health and safety regulations), making women aware of rights that they cannot enforce can be both misleading and, as findings from the WiF evaluation in Bangladesh indicate, possibly psychologically damaging [15]. Before more funds, efforts and hopes are invested in anti-trafficking interventions based on the belief that women will be able to assert their rights, this field will benefit from foundational research to inform interventions that focus on meso, institutional and macro level structures, i.e. employers, workers, frontline civil servants and policy decision makers, which do not place the burden of "keeping safe" on potential migrants themselves.