Understanding the effects of nutrition‐sensitive agriculture interventions with participatory videos and women's group meetings on maternal and child nutrition in rural Odisha, India: A mixed‐methods process evaluation

Abstract A trial of three nutrition‐sensitive agriculture interventions with participatory videos and women's group meetings in rural Odisha, India, found improvements in maternal and child dietary diversity, limited effects on agricultural production, and no effects on women and children's nutritional status. Our process evaluation explored fidelity, reach, and mechanisms behind interventions' effects. We also examined how context affected implementation, mechanisms, and outcomes. We used data from intervention monitoring systems, review notes, trial surveys, 32 case studies with families (n = 91 family members), and 20 group discussions with women's group members and intervention workers (n = 181 and 32, respectively). We found that interventions were implemented with high fidelity. Groups reached around half of the mothers of children under 2 years. Videos and meetings increased women's knowledge, motivation and confidence to suggest or make changes to their diets and agricultural production. Families responded in diverse ways. Many adopted or improved rainfed homestead garden cultivation for consumption, which could explain gains in maternal and child dietary diversity seen in the impact evaluation. Cultivation for income was less common. This was often due to small landholdings, poor access to irrigation and decision‐making dominated by men. Interventions helped change norms about heavy work during pregnancy, but young women with little family support still did considerable work. Women's ability to shape cultivation, income and workload decisions was strongly influenced by support from male relatives. Future nutrition‐sensitive agriculture interventions could include additional flexibility to address families’ land, water, labour and time constraints, as well as actively engage with spouses and in‐laws.

gender-and nutrition-sensitive agriculture intervention would increase agricultural productivity, boost women's agency, and promote nutritious crops that can be consumed by women or sold for income without increasing workload or infection risk (Kadiyala et al., 2014). Existing nutrition-sensitive agriculture interventions have had mixed effects on agriculture production, income, women's empowerment, workload and nutritional status (Girard et al., 2012;Johnston et al., 2018;Ruel et al., 2018), suggesting the need for further research (Sharma et al., 2020).
Upscaling Participatory Action and Videos for Agriculture and Nutrition (UPAVAN) was a four-arm cluster-randomised controlled trial testing three nutrition-sensitive agriculture interventions with participatory videos and women's group meetings to improve maternal and child nutrition in rural villages of Odisha, India. The interventions aimed to improve two primary outcomes: dietary diversity among children aged 6-23 months and the body mass index (BMI) of mothers or female primary caregivers of children aged 0-23 months. Secondary outcomes were maternal dietary diversity and weight-for-height z-score of children aged 0-23 months. UPAVAN interventions sought to influence these outcomes by supporting seasonally appropriate, locally feasible production and diversity of nutritious or income-generating foods, improving women's decision-making power in agriculture activities and reducing women's workload in agriculture.
In this study, we report the trial's process evaluation, which aimed to understand the fidelity of the interventions' implementation, their reach, underlying mechanisms and contextual factors affecting implementation, mechanisms and outcomes (Moore et al., 2015).  (Kadiyala et al., 2021).
The majority (81%) of households owned less than 1 hectare of land. Fifty-eight percent were from indigenous communities (scheduled tribes), 30% from other backward class communities and 9% from scheduled castes (Kadiyala et al., 2021). Indigenous families often have lower incomes, smaller landholdings, and poorer health compared to other social groups (Debasree, 2015;Kumar et al., 2011;Malik, 2020;Rath, 2005 Families in the study areas had a mean of 5.2 members (SD: 1.7).
Most practised subsistence agriculture alongside wage labour. Around

Key messages
• In rural eastern India, participatory videos and women's group meetings on agriculture and maternal and child nutrition increased women's knowledge, motivation and confidence to improve their and their children's diets.
• Given strong constraints linked to small landholdings, poor access to water and gender norms which meant that in-laws' and husbands' assent or support were often required for cultivation decisions, many women responded to interventions by increasing rainfed homestead garden cultivation for consumption rather than cultivating for income.
• Women's and children's diets may have improved because of discussions about nutrition and an increase in homestead garden cultivation. These dietary changes alone were likely insufficient to improve women's and children's nutritional status.
• The interventions' ability to influence the adoption of nutrition-sensitive agriculture practices could be improved by being family-centric: understanding women's decision-making power in a family context, using tailored problem-solving to address households' individual constraints to cultivation, and including women's husbands and in-laws. half (48%) of crops came from rainfed kharif (monsoon) agriculture and 70% of vegetable cultivation required irrigation (District Level Implementation Committee Pradhan Mantri Krishi Sinchayee Yojana, 2016).
Between 60% and 80% of families had homestead gardens of 5-9 m 2 , mostly tended by women (Figure 1). The trial baseline suggested that the median annual paddy yield per household (700 kg) was insufficient to feed most families. Eighty per cent of households in all arms produced paddy, 52% of households caught fish, 46% reared chickens, 9% produced dairy, and less than 16% produced vegetables. The baseline also found that 67% of women were involved in some or all decisions about crops grown for household consumption and livestock, but only 18% took part in decisions about crops grown for income generation. Newly married women worked an average of 11 h per day in and on the farm. Mothers-in-law had a strong influence over their diets and workloads.

| Interventions
The UPAVAN trial tested three participatory nutrition-sensitive agriculture interventions to improve maternal and child nutrition. The interventions included fortnightly women's group meetings followed by home visits over 32 months, as described in Panel 1 and in the intervention video (Sharma, 2021). All interventions were offered through women's self-help groups of 15-20 members, but all women living in intervention clusters were invited to participate in UPAVAN meetings, whether they were self-help group members or not. UPAVAN meetings therefore included a mix of self-help group members who contributed to the group fund, and nonself-help group members who did not contribute financially. Although usual self-help group activities (credit, loans and savings) were generally kept separate from the UPAVAN meetings, on two occasions (at the start and 6 months later) voluntary association for rural reconstruction and appropriate technology (VARRAT) led self-help group-strengthening activities that aimed to increase membership and record-keeping skills.
In the first of the UPAVAN intervention arms (AGRI), local salaried facilitators called Community Service Providers disseminated videos on nutrition-sensitive agriculture and facilitated related discussions with women's groups. Community Resource Persons trained as videographers filmed farmers and government frontline workers discussing or demonstrating core practices. Videos aimed to promote uptake or improvements in the cultivation of seasonally appropriate crops and livestock (e.g., introducing carrot cultivation, improving spacing between spinach plants, sowing seeds for cowpea cultivation, vaccinating goats, rearing chickens), develop skills (e.g., household budgeting, crop planning) and encourage group-based activities (e.g., buying seeds in bulk).
Video content was selected using social and agronomic information gathered during formative research, and by responding to requests or barriers faced by group participants, including land or water scarcity (Aakesson et al., 2017;Harris-Fry et al., 2020). For example, some videos focused on cultivating Indian spinach with wastewater. Others described how to grow mushrooms, which require minimal labour and are of high economic value.
In the second intervention arm (AGRI-NUT), women's groups viewed and discussed nutrition-sensitive agriculture videos and videos promoting nutrition-specific maternal, infant and young child nutrition (AGRI-NUT) designed using formative research, again under the guidance of Community Service Providers. Videos also described ways to overcome social and economic barriers to adopting these changes. For example, one video described a mother-in-law becoming an advocate for the consumption of Indian spinach, challenging widely-held beliefs about spinach being harmful for children and pregnant women. Other videos gave recipes and demonstrations on the appropriate diversity, quantity, frequency and consistency of complementary foods for young children.
In the third intervention arm (AGRI-NUT-PLA), women's groups viewed and discussed nutrition-sensitive agriculture videos, but also took part in a Participatory Learning and Action cycle of meetings in which they identified, prioritised and addressed problems related to maternal and child nutrition in four phases . In phase one, groups identified and prioritised problems related to maternal and child nutrition and hygiene. In phase 2, they sought to understand the causes of prioritised problems and identify locally appropriate strategies to address them. In phase 3, groups implemented and reviewed chosen strategies. In the final phase, groups evaluated activities and discussed strategies which they thought were impactful or F I G U R E 1 Homestead garden in Keonjhar, Odisha PROST ET AL.
| 3 of 15 difficult. Participatory Learning and Action meetings were either discussions without videos using techniques relevant to each phase of the cycle, or participatory videos on nutrition-specific topics that were developed as part of the Participatory Learning and Action process. The nutrition-specific videos used in this arm were therefore different to those in the AGRI-NUT arm. In the AGRI-NUT-PLA arm, groups had on average one nutrition-sensitive agriculture video and one Participatory Learning and Action meeting per month.
While creating all intervention videos, we used the transtheoretical model of behaviour change's stages of change to reinforce uptake and maintenance of core practices (Prochaska & DiClemente, 1982).
We hypothesised that one or more of the three UPAVAN interventions would improve nutrition-sensitive agriculture and nutrition-specific practices through six pathways: (1) mothers and children would consume nutrient-rich foods produced; (2) households would invest income from agriculture to improve nutrition; (3) pregnant and breastfeeding women would reduce their energy expenditure and hours worked; (4) women would influence and make more decisions about agricultural activities, workload and use of income; (5) pregnant and breastfeeding women would have adequate diets; (6) children would have adequate diets.
These changes would in turn lead to improved nutritional status for women and children. Our theory of change ( Figure 2) described these pathways and the intervention inputs, activities and enablers, with broad arrows reflecting hypothesised causal mechanisms. We used the term 'mechanism' to refer to both enablers and pathways in Figure 2. We anticipated that improvements in nutrition outcomes would occur through changes to multiple rather than single enablers and practices. For example, women might require both knowledge PANEL 1 All UPAVAN interventions used a participatory video approach designed by Digital Green, an international nongovernmental organisation. Community members identified locally relevant agriculture-related topics and developed packages of practices-key actions to improve agricultural practices-to discuss in videos. Community resource persons who were local community members were trained in storyboarding, video production and editing. Other community members were then filmed demonstrating and discussing the key practices. Salaried Community Service Providers showed these videos to women's groups, pausing videos at specified timepoints to stimulate discussion of the practices. Groups were then encouraged to discuss the feasibility, interest, apprehension or experience and possible barriers in adopting the practices. Facilitators visited group members in their homes or farms and asked whether they adopted the practices. Community Service Providers monitored video viewership, knowledge recall and adoption of practices. These data were collated in a monitoring system, and qualitative feedback was gathered during review meetings. This information was used by local implementers to identify future video content.
We adapted this participatory video approach to make it nutrition-sensitive and enhance participation, as described in depth elsewhere . Interventions delivered in the three UPAVAN arms built on the participatory video approach and were as follows: AGRI arm: fortnightly women's groups viewing and discussing videos to promote nutrition-sensitive agricultural practices developed through participatory processes in the local area. Nutrition-sensitive agriculture videos promoted practices aimed at increasing the availability of nutritious foods, increasing household income, improving women's decision-making in agriculture and reducing pregnant and breastfeeding women's workloads. Group participants who were pregnant or had a child (<2 years) were meant to receive followup visits at home after each video to check whether they retained the messages, discuss if mothers or household members had adopted practices shown in the videos and encourage continued participation in the intervention.
AGRI-NUT: as in AGRI, but with the addition of nutrition-specific videos on maternal and child nutrition focused on age-appropriate child feeding practices, care during child illness, and maternal diets and rest.
AGRI-NUT-PLA: fortnightly women's groups viewing and discussing participatory nutrition-sensitive agriculture and nutrition-specific videos combined with a cycle of Participatory Learning and Action (PLA meetings, with follow-up home visits. The PLA meeting cycle with women's groups had four phases. In phase 1, group members identified and prioritised nutrition problems. In phase 2, they explored the causes and effects of prioritised problems, planned locally feasible strategies to address these, decided on roles and responsibilities for implementing the strategies, and shared their learning with the wider community. In phase 3, groups implemented their strategies. Finally, in phase 4, they evaluated the process. about nutrition-sensitive agricultural practices and an enabling household environment to adopt new practices. In addition, changes to multiple practices would be required to affect an outcome such as BMI. We also anticipated that change would not be linear. For example, some practice-outcome relationships could be mutually reinforcing: a woman gaining income from selling mushrooms after watching a related video might become more motivated to attend additional video dissemination and adopt other practices.

| Trial results
Detailed UPAVAN trial results have been reported elsewhere (Kadiyala et al., 2021). Briefly, the trial was powered to detect a 9%

| Process evaluation objectives
Our mixed-methods process evaluation had three objectives: (1) To assess the fidelity of the UPAVAN interventions' implementation and their reach (2) To clarify the mechanisms behind the interventions' effects (3) To identify contextual factors associated with variations in implementation, mechanisms and outcomes

| Data collection
We used six sources of quantitative and qualitative data, as described in Table 1. Quantitative data included the trial surveys and information from routine monitoring. Trial surveys captured data on group attendance and home visits 6 months before the survey among mothers of children under 2, as well as household agricultural production diversity, gender parity in decision-making for agriculture and health, household expenditure, women's work, maternal and child dietary diversity, maternal BMI and child wasting. Methods for these surveys are described fully in the trial article (Kadiyala et al., 2021).
Supporting Information: Figure 1 is a timeline for the interventions and data collection. Five D-COR team members fluent in Odia collected the qualitative data over two phases (March to April 2018 and March to April 2019) with support from VARRAT, Ekjut and the London School of Hygiene and Tropical Medicine. We aimed for the qualitative sample to include community members who could be influenced by the intervention (self-help group members, and members who were pregnant or had a child aged under 2 years, their husbands and in-laws), as well as intervention workers. We also aimed for representation from communities across the trial's three intervention arms. We therefore, selected five trial clusters per intervention arm, broadly following the trial's stratification: one with a low proportion of scheduled caste and/or indigenous families (<30%); two with a medium (30%-70%) proportion (one close, i.e., <10 km, and one further away, i.e., ≥10 km from the nearest town) and two with a high (>70%) proportion (one close and one further away from the nearest town). In each village, D-COR asked for permission to attend a video dissemination meeting, and then approached pregnant women and mothers of children under 2 years to take part in semistructured interviews. These women, along with their mother-in-law and husband when available, were interviewed to create a case study. The case studies focused on changes to the diets of mothers (including during pregnancy) and young children and the adoption of nutrition-sensitive agriculture practices. We sought to understand what enabled these changes, and the reasons behind nonadoption. Topic guides were developed to capture elements in the theory of change and are included in the supplementary file.
Supporting Information: Table 1 shows the case study participants' characteristics.
We invited all self-help group members from the five selected villages for focus group discussions. These discussions focused on understanding who participated in self-help group meetings and members' experiences of the UPAVAN interventions. Group discussions conducted with Community Service Providers and supervisors explored barriers and enablers to delivering the interventions as well as community responses to interventions. We aimed for the final qualitative sample size to be around 20 group discussions with 10-15 participants each, and 30 case studies with 3-4 participants each. We felt that this would enable us to explore a variety of family circumstances and responses to interventions.  (Denzin & Lincoln, 2017). We first coded data using a framework derived directly from elements of the theory of change and captured each element under inputs, activities, enablers, promoted practices/ key pathways and outcomes. We then added new nodes for emergent themes (e.g., lack of water as a barrier to taking up nutrition-sensitive agriculture). A. P. coded case studies using the theory of change framework in Nvivo, then summarised them narratively with inputs from S. M. We discussed results with the entire UPAVAN study team to check our interpretations.

| Analysis
The final analysis of process evaluation data was conducted after the trial results were known. We explored fidelity in relation to the 'inputs' described in the theory of change, using intervention monitoring and qualitative data as well as review meeting notes.
We then described the interventions' reach and contextual factors that might affect it using the trial survey's coverage data and qualitative data. Finally, we reviewed mechanisms related to both enablers and promoted practices along the key pathways described in Supporting Information: Table 3, using qualitative data. We analysed qualitative and quantitative data separately but triangulated these F I G U R E 2 Theory of change, with the strength of evidence for the activation of components.
while examining all components of the intervention's theory of change. If quantitative surveys provided evidence for the activation of a component in one or more intervention arms compared to the control arm, and if the qualitative data supported this, we interpreted this as strong evidence that the component had been activated. If only one type of data supported activation, or if they suggested heterogeneous effects, we interpreted this as moderate evidence of activation and explored reasons for heterogeneity in the qualitative data. If neither the qualitative nor quantitative data supported activation, or if the component could only be measured using quantitative data and these did not suggest an effect, we interpreted this as 'no evidence of activation'. We provide supporting data related to the activation of some components in the Supporting Information: Appendix (e.g., SD1, is 'Supporting Data 1' in the Supporting Information: Appendix, Table 3). Figure 2 is a colour-coded version of the intervention's theory of change, indicating components for which we found strong, moderate or no evidence of activation. We coded several components as having moderate evidence of activation because qualitative data showed that women had differing experiences, both between and within arms. We now review individual components of the theory of change. We highlight contextual factors that could explain variations in implementation, mechanisms and outcomes throughout each section. All supporting data are presented in the text or in indexed numerically in the Supporting Information: File's Table 3 (e.g., SD1 is the first row in the Supporting Information: Data Table).

| Fidelity
UPAVAN interventions were implemented with high fidelity. VAR-RAT held intervention launch events with community members, local magistrates, self-help group representatives, Panchayati Raj members, school teachers and government community-based frontline workers (Anganwadi workers and Accredited Social Health Activists).
VARRAT also organised 2-day trainings with these frontline workers in all study areas, including the control arm. This aimed to increase collaboration between frontline workers and Community Service Providers, and ensure they had similar knowledge. Monitoring data suggested that 38.3% (646/1687)

| Group meetings
UPAVAN aimed to reach pregnant women and mothers of children under 2 by strengthening self-help groups. There was some evidence that this worked. In the trial's baseline data, only 19% of mothers of children under 2 years were 'active' in self-help groups, which was defined as participating in meetings and discussions for the last three  (Kadiyala et al., 2021) 3 Thirty-two family case studies including semistructured interviews of pregnant women, mothers of children under two, their husbands and in-laws (8 in the AGRI arm, 12 in the AGRI-NUT arm, and 12 in the AGRI-NUT-PLA arm or a total of 91 interviews).

4
Seventeen focus group discussions with a total of 181 self-help group members (5-6 groups per arm).

5
Three focus group discussions (one per intervention arm) with a total of 32 VARRAT staff. These included Community Service Providers, VARRAT's frontline workers, and their supervisors.

6
Five annual reports and six reports of intervention team review meetings.
Abbreviation: VARRAT, voluntary association for rural reconstruction and appropriate technology.
PROST ET AL.
| 7 of 15 agricultural seasons (1 year). By contrast, at endline, around half of mothers of children under 2 years were active, including in the control arm. Between 50% and 57% of mothers had seen any video (AGRI and AGRI-NUT) or been to any Participatory Learning and Action meeting (55% in AGRI-NUT-PLA) in the 6 months before the endline survey (Supporting Information: Table 2). Mean (range 0-11) number of events (videos of PLA meetings) mothers attended in the last 6 months before the endline survey were 3.3 (AGRI); 4.4 (AGRI-NUT) and 4.6 (AGRI-NUT +PLA).
In group discussions, self-help group members explained that some women did not join groups because they did not have money to contribute, or conversely because they already had money and therefore had no interest in joining a self-help group (SD5)

| Enablers
In UPAVAN's theory of change, participatory videos and women's group meetings were conceptualised as giving group members relevant knowledge, enabling peer support and, especially in the AGRI-NUT-PLA arm, engaging in collective learning and problemsolving and action with support from the wider community. We found that these objectives were largely met. In Participatory Learning and Action meetings, group members also identified strategies to address their prioritised problems and took collective and individual responsibility for implementation. Each group implemented a set of prioritised strategies and monitored progress in every meeting. For example, group members in one village decided to practise handwashing with soap before feeding children, counsel mothers-in-law and other family members not to impose food restrictions on pregnant women and mothers of children under 2, and give diverse foods to children after 6 months and develop homestead food production. At every meeting, the Community Service Provider discussed the implementation of these strategies and tried to resolve challenges. Some groups also took collective action by organising rallies for the wider community.
Overall, enablers specific to Participatory Learning and Action included engaging active learning techniques, in-built mechanisms to enable problem-solving, and included more community members than self-help groups alone.

| Exploring intervention mechanisms along the six key pathways to impact
We found that the effects of UPAVAN interventions on agricultural practices were highly heterogenous both between and within-trial arms. Figure 3 summarises changes in practices seen in the 32 family case studies. In the AGRI arm, three out of eight families who took part in case studies described making changes to agricultural practices. None mentioned changes to dietary practices caused by interventions. Instead, family members commonly said they ate 'what F I G U R E 3 Self-reported changes in nutrition-sensitive agriculture and nutrition-specific practices, by arm, in case studies. PROST ET AL. | 9 of 15 was available'. In the AGRI-NUT arm, 9 out of 12 families described changes to dietary practices due to the interventions. Many said that nutrition-sensitive agriculture would help them save money, be more self-reliant and eat 'fresh', organic food that they would grow and not buy, as opposed to 'junk' or pesticide-laden foods. In the AGRI-NUT-PLA arm, 11 out of 12 families described adopting some nutritionspecific or -sensitive practices, or both. The scale of adoption varied greatly. Supporting Information: Figure 2 summarises case studies from six women (two per arm), describing their responses to interventions. We now use these case studies and other process evaluation data to explore how context shaped mechanisms and outcomes along the six 'pathways' to improved nutritional status described in the theory of change.
3.4.1 | Pathways 1 and 2-Producing food and earning income: It was easier to adopt or improve small-scale, rainfed homestead garden cultivation for consumption than to production for income generation A minority of women whose families had land and were supportive could influence cultivation in medium to large areas, as described in this case study from AGRI-NUT (SD12): Savi explained row planting for paddy to her husband. He adopted it, along with corn and tomato cultivation, and the use of pot manure. The family fed their children corn and sold some too. They cultivated tomatoes by planting seeds at the same time as paddy in June, but could not provide water for more than 20-30 tomato plants as they had to carry it. Savi's husband recalled: 'When I made the seedbed, she [Savi] came and said, "no, we'll make this kind of seedbed". We planted in rows. After seven days, we applied fertilizer, manure, Gender-and age-related social norms greatly shaped agricultural decision-making. In most cases, in-laws or husbands owned the agricultural land, so decisions about cultivation could seldom be made without their involvement. The fact that most families planned crop cultivation and sales 'by consensus' also effectively meant that inlaws' assent was required for all major decisions. Heterogeneity existed even in the context of strong social norms. Some women, like Muni (case study 24, Supporting Information: Figure 2), were able to participate in agricultural decisions and felt listened to by their families. Others were completely excluded from agricultural decisionmaking, even though in-laws expressed great care for them, as the Anganwadi] and they also eat fish, meat, and eggs.
They also eat green vegetables and spinach… (Selfhelp group discussion 6, AGRI-NUT) Similarly, many women said that videos had given them knowledge, motivation and confidence to improve child feeding practices (SD20). For example, several described preparing more diverse complementary foods (e.g. Shital, in AGRI-NUT-PLA, case study 31 in Supporting Information: Figure 2). Group members mentioned that handwashing before feeding children was also more common because of videos, meetings, frontline workers' initiatives, and ongoing government campaigns (SD21). The Community Service Providers' work was helped by the fact that UPAVAN had purposefully trained government frontline workers to offer similar information to the videos and meetings, and that frontline workers often attended video dissemination and PLA meetings (SD22): P: Earlier I used to make breakfast for my child, but I used to skip it. […]. In the evening I cooked snacks for the child but I was not eating that. The Accredited Social Health Activists told me to prepare more snacks in the morning and evening so that both I and my child can eat properly. So I followed that. The same thing was shown in the video. (CS10, AGRI-NUT, pregnant woman)

| DISCUSSION
Our process evaluation found that UPAVAN interventions provided a 'menu' of seasonally appropriate crops and strategies to women, which enabled many families to take up nutrition-sensitive agriculture practices that suited their circumstances. However, uptake of nutrition-sensitive agriculture was heterogeneous and most often led to rainfed homestead garden cultivation for consumption (pathway 1) rather than cultivation for income and on larger land plots.
Small landholdings, lack of irrigation and lack of family support remained key structural and social barriers to cultivation for income (pathway 2). Although norms around heavy work in pregnancy were changing with the convergence of UPAVAN interventions and frontline workers' efforts, younger and less supported women still did considerable housework (pathway 3). Their ability to influence or take decisions about cultivation, income and workload was strongly shaped by the nature of their relationships with in-laws and spouses, and by the extent to which families were already invested in agriculture versus wage or salaried labour (pathway 4). UPAVAN's strongest effects were in accelerating improvements to women's and children's diets against a backdrop of supportive government initiatives (pathways 5 and 6), primarily by providing women with information, motivation and confidence.
This study adds to a growing body of research aiming to understand factors that influence the delivery of nutrition-specific and -sensitive interventions (Menon et al., 2013). Our process evaluation found that UPAVAN interventions were well-implemented in the context of an efficacy trial. In the UPAVAN impact evaluation, we hypothesised that the lack of detectable effects along four of the six pathways to improved nutritional status may have been partly due to the strong secular improvements in health, diet and nutritional status seen across Odisha during the trial period (Avula et al., 2020;Kadiyala et al., 2021). involving husbands and in-laws more actively. We elaborate on these suggestions below.
The trial found no effects on gender parity in empowerment in agriculture, but some modest effects on women's decision-making in agriculture and health-related activities in the AGRI arm. Previous research has found that relationships between women and their mothers-in-law are central to the household economy but also highly heterogeneous (Agarwal, 1997;Allendorf, 2007;Kandiyoti, 1988). This was echoed in our study: some daughters-in-law were given little support with housework and were largely unable to take decisions about their diets or workload, while others were given substantial help and shaped some or all decisions. A recent evaluation of a participatory nutrition-sensitive agroecological intervention in Tanzania also found no effect on women's agricultural decision-making (Santoso et al., 2019(Santoso et al., , 2021. Its authors argued for focused research to understand which barriers to increasing women's decision-making power matter most, and whether addressing these might increase the effects of nutritionsensitive agriculture interventions. We aim to address this first evidence gap in forthcoming analyses of UPAVAN data. The need for supportive legal frameworks and working conditions for women in agriculture also remain largely unaddressed in nutrition-sensitive agriculture interventions and policy across India and South Asia; changing these could greatly increase women's decision-making power and potentially the effects of community-based nutrition-sensitive agriculture interventions (Agarwal, 1994;De et al., 2021).
UPAVAN interventions were developed to respond to constraints that women faced, but production constraints linked to land, water and family support remained a challenge. As a result, nutritionsensitive agriculture practices were not taken up universally and were often season-and scale-limited. A recent systematic review of implementation and scale up of nutrition-sensitive agriculture interventions found that failure to achieve effects on nutritional status and pathways to improving it often arises from a lack of flexibility in intervention design (Di Prima et al., 2022). Some programmes found and overcame similar challenges (Nielsen et al., 2018). Future interventions therefore might emphasise the importance of flexibility in tailoring nutrition-sensitive agriculture advice to families' specific assets and constraints. Our study had several strengths. It used diverse data sources including a large body of qualitative data. Methodologically, our approach had similarities with the Programme Impact Pathways approach used in the handful of existing process evaluations of agriculture, nutrition, water, sanitation and hygiene interventions (Mbuya et al., 2015;Olney et al., 2013). Our theory of change served as a map to impact pathways, and we prespecified criteria for reporting components as 'activated' or not. Our process evaluation also had several limitations. We did not investigate sources of heterogeneity in agricultural production, household consumption, and women's empowerment quantitatively to test some of the hypotheses generated by our qualitative analysis. Finally, repeated probing for changed practices during interviews and group discussions may have induced socially desirable responses, though we sought to mitigate the effects of this by triangulating data within families and across sources.

| CONCLUSION
UPAVAN's interventions were delivered with high fidelity. They improved women's and children's diets by providing women with information, motivation and confidence. UPAVAN also enabled many PROST ET AL. | 13 of 15 families to take up nutrition-sensitive agriculture practices, but adoption was heterogeneous. Some families started rainfed homestead garden cultivation for consumption, fewer cultivated for income and many were unable to take up practices due to small landholdings, lack of irrigation or family support. UPAVAN interventions and frontline workers' efforts shifted norms towards reducing work in pregnancy, but younger, less supported women still did considerable work. Support from in-laws and husbands strongly influenced women's abilities to shape decisions about cultivation, income and workload.
We recommend furthe research to understand opportunities for increasing women's decision-making power within families, additional intervention tailoring to address families' specific land, water, labour and time constraints, and involving in-laws and husbands in nutritionsensitive agriculture interventions.

DATA AVAILABILITY STATEMENT
Data are available in the article Supporting Information Material.

ETHICS STATEMENT
The trial and its process evaluation were led by the London School of