Kangmennaang 2017.

Study characteristics
Methods	Study design: PCS Study grouping: N/A How were missing data handled? NR Randomisation ratio: N/A Recruitment method: villages in the Mzimba and Dedza districts were selected in consultation with residents after the project was introduced at awareness meetings. Community representatives generated a list of indicators for different levels of food security with project staff, and these were used as part of selection criteria. Participants were interviewed at baseline and asked to be re‐contacted at follow‐up. Sample size justification and outcome used: NR Sampling method: NR for intervention. Control: random (control HHs with similar characteristics to intervention group were randomly selected from nearby villages) Study aim or objective: to examine the impact of a participatory agroecology development project on family farmers' food security and assets following 2 years of intervention. Study period: about 2 years. Baseline: 2012. Follow‐up survey (1000 HHs) June–September 2014. Unit of allocation or exposure: cluster: HHs
Participants	Baseline characteristics Intervention or exposure Age: means, Table 2: wife: 1.233; husband: 1.063 Place of residence: NR Sex: NR Ethnicity and language: NR Occupation: NR Education: means, Table 2: wife's educational level: 0.787; husband's educational level: 0.826 SES: means, Table 2: wealth index: 1.059; HH size: 1.043; farm size: 0.42; cash cropping: 0.008 Social capital: means, Table 2: general HH well‐being: 1.251; marital status: 1.787 Nutritional status: Table 3: food insecurity, mean 0.966 (SE 0.036); Table 2: food insecurity, mean 0.966 (SE 0.036); higher score = more food insecurity Morbidities: NR Concomitant or previous care: NR Control Age: means, Table 2: wife: 1.255; husband: 1.007 Place of residence: NR Sex: NR Ethnicity and language: NR Occupation: NR Education: means, Table 2: wife's educational level: 0.794; husband's educational level: 0.765 SES: means, Table 2: wealth index: 0.917; HH size: 1.172; farm size: 0.363; cash cropping: 0.017 Social capital: means, Table 2: general HH well‐being: 1.123; marital status: 1.877 Nutritional status: Table 3: food insecurity, mean 0.873 (SE 0.049); Table 2: food insecurity, mean 0.873 (SE 0.049) Morbidities: NR Concomitant or previous care: NR Overall Age: Table 1: wife, number (%): aged < 30 years: 375 (30.49); aged 30–44 years: 374 (30.41); aged 45–60 years: 240 (19.51); aged > 60 years: 241 (19.59). Husband, number (%): aged < 30 years: 508 (41.30); aged 30–44 years: 346 (28.13); aged 45–60 years; 220 (17.89); aged > 60 years: 156 (12.68) Place of residence: NR Sex: NR Ethnicity and language: NR Occupation: NR Education: Table 1: husband, n (%): none: 439 (35.69); primary: 615 (50.00); secondary and higher: 176 (14.31). Wife, n (%): none: 368 (29.92); primary: 776 (63.09); secondary and higher 86 (6.99) SES: Table 1: wealth quintile, number (%): poorer: 260 (21.14); poor: 240 (19.51); middle: 246 (20.00); rich 199 (19.90); richer 243 (19.76); richest: 241 (19.59) Social capital: NR Nutritional status: NR Morbidities: NR Concomitant or previous care: NR Inclusion criteria: interest in doing farm experiments, food insecurity (determined at baseline using HFIAS and ability to farm (self‐reported) with further probes as to whether the HH had access to land and labour, and were already cultivating crops. Exclusion criteria: NR Pretreatment: agroecological users and non‐users were similar for HH structure, wife's age, husband's age, educational level of both husband and wife, knowledge of agricultural practices, food security and farm size at the baseline level. Significant differences for wealth, HH size, number of crops grown per field, dry season farming and general HH well‐being (see Table 2): non‐adopters were less wealthy, had a larger HH size, grew a lower number of crops and had a lower general HH well‐being. Attrition per relevant group: total 191/1191 (16%) HHs at follow‐up. Per‐group attrition unclear. Description of subgroups measured and reported: NR Total number completed and analysed per relevant group: 429 control, 571 intervention (based on MAFFA member) Total number enrolled per relevant group: 408 control, 793 intervention (based on MAFFA member) Total number randomised per relevant group: N/A
Interventions	Intervention characteristics Intervention or exposure Food access intervention category: increase buying power Intervention type: income generation Description: agroecological development project: encouraged farmer experimentation, community involvement and F2F teaching on agroecology, nutrition and gender equity. Farmers did their own experimentation with agroecological methods. MAFFA encourages farmers to adopt a suit of innovations rather than just a single innovation and to encourage farmer‐led learning. In addition to crop diversification, many farmers increased or began to apply compost and manure to their rain‐fed fields. Some farmers also experimented with botanical pesticides. MAFFA goes beyond agroecological training to focus on knowledge sharing, leadership support, nutrition and attention to social inequalities through an iterative process that integrates reflection and action, including the development of different educational activities, campaigns and training. Farmers were assisted with quality seeds, and transportation to experimental farms and community events, although farmers also incurred opportunity costs from lost farm work due to participation in these activities. Duration of intervention period: about 2 years (2012 to September 2014) Frequency: NR Number of study contacts: 1 Providers: Soils, Food and Healthy Communities organisation of Ekwendeni Hospital, Chancellor College, University of Malawi as well as Malawian and Canadian scientists Delivery: training, educational activities, campaigns, provision of seeds. Farmers shared knowledge with other farmers. Co‐interventions: NR Resource requirements: NR Economic indicators: NR Control: no intervention
Outcomes	HFIAS score
Identification	Sponsorship source: Global Affairs Canada of the Government of Canada, the Canadian Food Grains Bank, and Presbyterian World Service and Development. Sponsors had no role in study design, data analysis and interpretation, writing report, and decision to submit report for publication. Country: Malawi Setting: smallholder farm HHs Author's name: Joseph Kangmennaang Email: jkangmen@uwaterloo.ca Declarations of interest: yes; no conflicts of interest. Study or programme name and acronym: the Malawi Farmer to Farmer Agroecology project (MAFFA). Type of record: journal article
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (Selection bias)	High risk	Prospective controlled study. No randomisation carried out.
Allocation concealment (Selection bias)	High risk	Prospective controlled study. No randomisation carried out.
Baseline characteristics similar (Selection bias)	Low risk	Due to differences in baseline characteristics between MAFFA and non‐MAFFA HHs, study authors applied kernel‐based PSM to reduce the effects of confounding and account for any systematic differences in baseline characteristics to enable them to obtain unbiased estimates of the mean treatment effects on the outcomes (Austin 2011). However, the balancing test after weighting revealed no significant differences between participants and non‐participants (see Table 8).
Baseline outcome measurements similar (Selection bias)	Low risk	No significant difference between adopters and non‐adopters for HH food security outcome. Difference for wealth. However, (quote) "Due to differences in baseline characteristics between MAFFA and non‐MAFFA HHs, we applied kernel based propensity score matching to reduce the effects of confounding and account for any systematic differences in the baseline characteristics to enable us to obtain unbiased estimates of the average treatment effects on the outcomes (Austin 2011)."
Blinding of participants and personnel (Performance bias)	Low risk	Given the prospective longitudinal design of the study, it was not possible to blind participants or personnel; however, it was unlikely that the outcomes of interest (food security and HH wealth) were prone to performance bias.
Blinding of outcome assessment (Detection bias)	High risk	NR whether outcome assessors were blinded. Given the design of the study, participants could not be blinded, and it appeared as though outcomes were self‐reported (participants were interviewed and the HFIAS used to assess food security).
Protection against contamination (Performance bias)	Unclear risk	Adopters and non‐adopters coexisted in the same villages, and the nature of the intervention was farmers teaching other farmers about agroecology, making contamination likely. However, given that non‐adopters had to actively opt out of intervention, it is unclear to what extent contamination might have occurred. Despite the total number of HHs was smaller at the time of follow‐up, the number of HHs in the control group (based on MAFFA membership) increased.
Incomplete outcome data (Attrition bias)	High risk	Attrition was high (16%) in total group and there was no strategy to account for missing values.
Selective outcome reporting (Reporting bias)	Unclear risk	No protocol available.
Other bias	Unclear risk	Potential of recruitment bias. Although inclusion criteria were mentioned, was not clear from these criteria who was included or excluded from the study: (quote) "interest in doing farm experiments, food insecurity (determined at the baseline using the Household Food Insecurity Access Scale (HFIAS), and ability to farm (self‐reported) with further probes as to whether the HH had access to land and labor, and were already cultivating crops."