Plant different, eat different? Insights from participatory agricultural research.

We examine the association between on-farm production diversity on household dietary diversity in Malawi using microdata collected as part of an environmentally sustainable agricultural intensification program. The program primarily focuses on the integration of legumes into the cropping system through maize-legume intercropping and legume-legume intercropping. Relative to staple cereals such as maize, legumes are rich in micronutrients, contain better-quality protein, and lead to nitrogen fixation. Given the systematic difference we document between program beneficiaries and randomly sampled non-beneficiary (control) households, we employ causal instrumental variables mediation analysis to account for non-random selection and possible simultaneity between production and consumption decisions. We find a significant positive treatment effect on dietary diversity, led by an increase in production diversity. Analysis of potential pathways show that effects on dietary diversity stem mostly from consumption of diverse food items purchased from the market made possible through higher agricultural income. These findings highlight that, while increasing production for markets can enhance dietary diversity through higher income that would make affordable an expanded set of food items, the production of more nutritious crops such as pulses may not necessarily translate into greater own consumption. This may be due to the persistence of dietary habits, tastes, or other local factors that favor consumption of staples such as maize and encourage sales of more profitable and nutritious food items such as pulses. Pulses are a more affordable and environmentally sustainable source of protein than animal source food, and efforts should be made to enhance their nutritional awareness and contribution to sustainable food systems and healthier diets.

Introduction

Agricultural interventions in most of Africa historically had narrow focus on increasing yield per unit area of a few staple crops through labor-saving (e.g., mechanization) and land-saving (e.g., fertilizer and pesticide) technologies without adequate consideration for the whole ecosystem and implications for nutrition and health [1,2]. In light of persistent food insecurity and climatic risks the region is facing, interest has grown on how best to leverage agriculture to tackle undernutrition while simultaneously improving the natural resource base and resilience. A recent review paper from Africa and other developing regions shows that most of the evidence supports a positive association between adoption of agroecological practices (e.g., crop diversification, cereal-legume intercropping, agroforestry, crop-livestock integration, and integrated soil and water management practices) and food security [3].

Agriculture contributes to the food security of poor rural households both directly, by boosting food availability for subsistence-oriented farmers, and indirectly, by enhancing income for commercially oriented farmers. Diversification of agriculture production into nutrient-rich crops and animal-sourced foods (ASF) is often considered as one of the options for improving diets and nutrition among smallholders considering their reliance on own-produced foods [4]. Especially when access to markets for buying food and selling agricultural production is limited, diverse agricultural production can play a vital role in ensuring diversified food consumption [5]. For example, evidence from Ethiopia shows that households that live far away from market centers not only consumed less diverse foods but also had smaller food consumption expenditure relative to households who live close to markets [6].

Most of the empirical studies on the linkages between on-farm diversity and diets as well as the role of mediating factors such as access to food markets rely on cross-sectional data to generate evidence about associations versus causality [7-16]. The evidence generated from these studies is mixed not only across studies but also within a study based on geography, diversity indicators, and commodities [4,16-18].

Dependence on cross-sectional data poses several challenges while establishing linkages between production diversity and dietary diversity due to a host of confounding factors that jointly and simultaneously affect production and consumption decisions. This study uses cross-sectional data from Malawi to examine linkages between on-farm production diversity and household dietary diversity. We contribute to the literature by analyzing household level dietary patterns among households who participated in an agricultural intensification program that aims to integrate pulses into Malawi’s maize-dominated farming systems.

Taking advantage of quasi-experimental agricultural and dietary data collected from program participant and non-participant (control) households, we use the recently developed mediation analysis approach applied to instrumental variables (IV) frameworks [19,20] in which multiple endogeneity and transmission mechanisms are simultaneously at play and are controlled for using a single instrument. The approach has been successfully employed in several recent publications [21-23].

We find that participating into the program leads to a significant increase in production diversity that, in turn, translates into more diverse diets. The main underlying mechanism, however, appears to be working through higher purchase of more diverse foods, rather than through an increase in own consumption of pulses, the main crops targeted by the program. Finally, these effects are primarily due to an increase in the production of a secondary targeted crop, maize, and not to higher pulse production. These findings highlight that, while increasing production for markets can lead to more diverse diets through increased income that would make affordable an expanded set of food items, increasing production of more nutritious crops such as pulses may not necessarily translate into greater own consumption due to the persistence of dietary habits, tastes, or other local factors that favor consumption of staples such as maize and encourage sales of more profitable and nutritious food items such as pulses.

Study setting

Malawi’s crop production is highly dominated by continuous maize production. For example, a recent study based on a panel of field-level data from Central and Southern Malawi showed maize was continuously grown for four years on more than 55% of the plots in Central region and more than 82% of the plots in Southern region [24]. This practice has implications not only for depletion of essential soil nutrients but also for the diversity of household diets. Through Malawi’s Farm Input Subsidy Program (FISP), the government has been providing subsidies to smallholders for inorganic fertilizers and improved seeds for maize since 2004/05. The provision of improved seeds as part of FISP was expanded to legumes in 2008/09 in recognition of their contribution for improving soil quality and nutrition [25]. Nonetheless, maize is still Malawi’s main staple food accounting for about half of the total plant-based caloric intake with pulses and groundnut accounting for just 7% [26].

More than 90% of Malawi’s population lives under $1.9 a day poverty line (in purchasing power parity), 19% of the population is undernourished, and 40% of children below five years old are stunted [27]. Amid evidence of FISP failing to meaningfully reduce food insecurity and enhance dietary diversity, interest has grown on how best to leverage various agricultural strategies to effectively tackle undernutrition including through adoption of agroecological principles and diversification [5,28-30]. While crop diversification is among the goals of Malawi’s agricultural sector development strategy, ensuring maize self-sufficiency remains the main focus of public agricultural investments [31]. Adoption of agroecological practices in Malawi has been linked with improved food security [30] and stronger association between on-farm and dietary diversity [28]. A positive association has also been documented between crop diversification and dietary diversity with varying magnitude [5,9,29].

This study is conducted as part of an agriculture research for development program called Africa RISING being implemented in Dedza and Ntcheu districts of Malawi since 2012. The program aims to promote sustainable intensification primarily through integration of legumes (groundnut, pigeon pea, cowpea, soybean, and beans), organic and inorganic fertilizers and livestock innovations. Through a participatory ‘mother-and-baby’ trial design [32], the program has been validating integrated packages of agricultural technologies for subsequent scaling. Interactive and replicable demonstration (“mother trial”) plots were established around farmer action groups, whose members subsequently set up adaptive (“baby trial”) plots to test a subset of technologies from the mother trial (hereafter program beneficiaries). Fig 1 shows the mix of agricultural innovations tested by program beneficiaries at the time of the baseline where pulses -with or without maize and with or without inorganic fertilizers- were quite common.

Fig 1

Mix of innovations tested by program beneficiaries.

Note: The five pulses include groundnut, pigeon pea, cowpea, soybean, and common bean; NPK: Nitrogen, phosphorous, and potassium; comp: Compost.

One of the criteria for setting up baby trials was that farmers shall select no more than four (integrated) technologies and be able to devote at least 10 square meters of accessible land for each treatment chosen. As shown later, beneficiaries appear to be systematically different from randomly selected control households including cultivating bigger land area. Significant attention was given to the practice of intercropping, namely maize-legume intercropping or intercropping of two legumes with different growing periods-known as doubled-up legume technology [33]. Compared to cereals, pulses are rich in crucial micronutrients, contain better and higher quality protein, can be a more affordable source of protein compared to ASF. Legume intercropping helps improve soil fertility, yield, and nutrition while reducing fertilizer requirements due to nitrogen fixation [34-36]. Various efforts are underway to promote pulse production and their multifunctional roles within the smallholder farming systems [37-39].

Materials and methods

Data and key variables

Microdata analyzed here were collected as part of a baseline survey approved by the Institutional Review Board of the International Food Policy Research Institute (IRB # 00003487). Written informed consent was obtained from all survey participants. Data were collected between August and October 2013 after program beneficiaries obtained their first harvest since joining the program [40]. As reported in a previous study [41], detailed agricultural and socio-economic data were collected from two groups of households: 1) all program beneficiary ‘baby’ farmers -and their households- that were engaged in testing program technologies as of June 2013 (N = 397) and 2) randomly sampled (control) households drawn from non-program target villages with comparable biophysical and agro-ecological conditions as program target villages (N = 538). Agricultural production data refer to 2012/13 long rainy season while food consumption data are based on 7-day recall period.

Production diversity was largely captured at the household level based on crop and/or livestock species and alternative indicators (e.g., count of unique food items, count of food groups with similar nutritional content, indicators of crop species richness and/or evenness) while dietary diversity was measured either the household or individual level based on metrices that include dietary diversity scores, food variety scores, and weighted food consumption scores [18]. In this study, we adopt a good group approach to measure diversity.

First, we classify all food items into 12 groups [42]: cereals, roots and tubers, pulses and nuts, vegetables, fruits, meat, eggs, fish, milk and dairy products, oils and fats, sugar and sweets, and miscellaneous items including spices, condiments, and beverages. Next, we compute total food consumption expenditure as the sum of expenditures on purchased food and imputed values of food consumed from own production and gifts. Imputation is based on food item-specific unit price values [43] computed by dividing total expenditure on purchased food by total quantity of purchased foods. We performed an outlier check and corrected for outliers in food value by replacing monetary values higher than +3 Standard Deviations (SD) or lower than -3 SD from the median by the distribution median. Finally, we construct Simpson’s diversity index [44], defined as , where j is food group, e is annualized household food expenditure on j in local currency -Malawian kwacha (MWK)-, and .

Simpson’s diversity index measures both richness -i.e., the number of food groups- and relative abundances -i.e., the extent to which food expenditure is uniformly spread across food groups- with values ranging between zero and one. It represents the probability that two randomly selected food items belong to two different food groups [45], with its value increasing with the number of food groups consumed, the evenness of household budget share distributions, or both. To assess the diversity of purchased foods, we construct an additional Simpson’s index based only on food items that were purchased.

Similarly, we construct Simpson’s on-farm production diversity index based on self-reported data on production of crops, livestock, and animal by-products. Production diversity is based on the following nine food groups: cereals, roots and tubers, pulses and nuts, vegetables, meat, eggs, fish, milk and dairy products, and cash crops such as sugarcane, cotton, and tobacco. Similar to the approach used for food consumption, we compute unit values of agricultural commodities by dividing total sales revenue by total quantity sold and subsequently use them to monetize the total value of agricultural production. Outliers in the value of agricultural production are corrected by replacing values higher than +3 SD or lower than -3 SD from the median by the median. As a sensitivity analysis, we estimate models of production and dietary diversity using Shannon’s diversity index. While Simpson’s index is a dominance index that assigns more weight to dominant food groups, Shannon index emphasizes the richness component of diversity and is given by where ln is natural logarithm and e is as defined before.

In addition to diversity indicators, we construct several socioeconomic variables that may be correlated with program participation and agricultural production, and mediate the interactions between production and dietary diversity. These include household demographic characteristics, use of agricultural inputs and practices, and four standardized (with mean zero and standard deviation one) indices based on the number of durable agricultural assets (excluding land), the number of durable non-agricultural assets, the quality of the household’s dwelling condition, and market access based on self-reported travel time to various services including the nearest markets (daily and weekly), nearest roads (motorized, all season, and asphalt), and schools. The indices are constructed using factor analysis based on principal-component factor method [46].

Identification

Our primary identification strategy relies on IV-mediation analysis [19,20]. Mediation analysis is common in the social sciences outside economics, and is gaining prominence in applied economic research in recent years [47-49]. The goal of mediation analysis is to unpack the transmission mechanisms in which a treatment T and a mediator M jointly cause an outcome of interest Y, by disentangling the total effect of T (TE) on Y into two components: the indirect or mediated effect, that is the effect of T on Y that operates exclusively through its effect on M; and the direct effect, that is the residual effect of T on Y that is not mediated by M, that is holding the distribution of M constant.

The traditional mediation analysis assumes T to be randomly assigned, an assumption that does not hold in our case. Non-random treatment assignment makes T endogenous with respect to both M and Y. Additionally, in the model M is potentially endogenous to Y. While the standard IV framework can be used to address endogeneity bias in the form of non-random selection and reverse causality, it is not suitable to identify the causal effect of the mediator on the outcome of interest -the focus of this study. Furthermore, finding an instrument that is relevant and meets the exclusion restriction is empirically challenging, especially when two or more valid, exogenous, and strong instruments are necessary to isolate the causal effects as would be needed here.

Our empirical approach blends the potential of mediation analysis in disentangling the causal chain between different outcomes with the ability of the IV framework to tackle endogeneity, called IV mediation analysis [19,20]. The approach allows one to use the same instrumental variable to identify the complex chain among the outcomes within a standard IV strategy that controls for treatment endogeneity with respect to the intermediate and final outcomes. In particular, mediation effects can be identified when the IV model is partially confounded, that is when the unobserved confounding variables expected to affect the treatment and the intermediate outcome are independent of the confounders that affect the intermediate and final outcomes [50]. This is the case when T is endogenous in a regression of M on T due to confounders that jointly affect M and T, and T is endogenous in a regression of Y on T due to the same confounders that affect Y primarily through M.

The two stage least squares (2SLS) estimation procedure to identify the causal effect of T on M () can be formalized by the two-equation system shown in Eqs 1 and 2 where Z is the instrument and is the predicted value of T from Eq 1. Under the identifying assumption for IV mediation analysis[20], causal effects of T on Y can be estimated via 2SLS estimation of Eqs 3 and 4 and a single instrument Z.

In Eq 4, the indirect or mediated effect of T on the outcome is provided by , the direct or residual effect of T on the outcome is given by , and the total effect is given by . In our case, this approach allows us to investigate the production-consumption transmission chain while accounting for non-random selection into the program, endogeneity, and potential simultaneity between production and consumption decisions. In this complex transmission chain, we assess the effect of participation in the Africa RISING program (T) on the Simpson’s household dietary diversity index (Y) as mediated by the effect of T on another mediator outcome (M), the Simpson’s on-farm production diversity index. We posit the partial identifying assumption to hold in our setting, as it is plausible to assume that program participation is endogenous in a regression of dietary diversity (DD) on T only due to confounders that jointly affect T and production diversity (PD).

As the explicit goal of the program was to diversify agricultural production, there are only two main channels through which the treatment (participation) could affect the dietary diversity of beneficiary households: either by increasing their production diversity through the cultivation of new cash crops, thus raising their agricultural income and, in turn, their consumption of more diverse foods; or by diversifying the variety of subsistence crops and, in turn, directly boost their dietary diversity. Both these channels are mediated by agricultural production diversity. For these reasons, which are strictly related to the nature and design of the project, we can safely assume that T is endogenous with respect to DD due to confounders that affect DD primarily through PD.

Adapting the formalization of the IV mediation analysis framework to our study, we estimate Eqs 5 and 6 using 2SLS to examine the associations between program participation, production diversity, and dietary diversity. where PD and DD represent Simpson’s production and dietary diversity indices, respectively; T is indicator for participation in Africa RISING program; Z is the instrument; X contains a set of control variables including household size, age and gender of the household head, average years of education in the household, dependency ratio, indices for non-agricultural assets and distance from basic services, temperature, slope, and indicators for self-reported shocks (drought and crop diseases); and εPD and εDD are model error terms. The estimate for the PD-mediated indirect effect of T on DD is given by , the direct or residual effect of T on DD is given by , and the total effect T on DD is .

Our choice of instrument was guided by the fact that beneficiaries were more likely to operate plots closer to the homestead relative to the control group. A summary of plot location based on self-reported travel time data shows that 48% and 52% of beneficiaries’ plots (N = 1,079) and 38% and 63% of control group plots (N = 1,008) were located, respectively, within 15 minutes of travel (nearby) and more than 15 minutes of travel (faraway) from the homestead. That is, the average plot owned by beneficiaries is more likely to be nearby while the opposite is true for the average plot owned by the control group. Average plot size was also statistically different by plot location‒0.76 (0.9) hectares (ha) for nearby (faraway) plots for the whole sample and 0.73 (.92) ha for beneficiaries. Smaller nearby plots may be due to shortage of agricultural land in and near residential areas.

While we do no not find statistically significant differences in several agronomic indicators including yield by parcel location for the whole sample, we observe statistically significant differences in the number of crops grown per parcel and the use of cereal-legume intercropping by parcel location and treatment status. Specifically, and relative to nearby plots, beneficiaries grow higher number of crops per parcel (2.5 versus 2.3) and are more likely to practice cereal-legume intercropping (55% versus 45%) on faraway plots while the opposite is true for control households‒where they grow 2.2 crops on faraway plots (versus 2 crops on nearby plots) where they also practice cereal-legume intercropping on 55% of the plots (versus 47% for nearby plots). The fact that beneficiaries needed devote at least 10 square meters to each program promoted technologies to participate in the program [41] and that faraway plots were, on average, bigger might explain the higher incidence of use of intercropping practice and number of crops on faraway plots.

We use total area of nearby plots to instrument for program participation and conduct two tests to provide indirect empirical evidence in favor of the exclusion restriction of our instrument. The first placebo test involves running a reduced form regression of the intermediate outcome‒production diversity‒ controlling for the instrument, other covariates discussed above, and EPA fixed effects separately on overall, treated, and control group samples. We expect statistically significant coefficients of the instrument for the treated group but not for the control group. In the second placebo test, we test whether the exclusion restriction holds by estimating a reduced form model of the final outcome ‒ dietary diversity‒ separately for overall, treated, and control groups. Once again, the coefficient of the instrument for the subsample of control households should not be statistically significant since the only way through which landholdings close to the homestead would affect dietary diversity is through production diversity to which the control group has not been exposed.

Since an IV estimator yields inconsistencies and finite-sample biases when the instrument(s) are weakly correlated with the endogenous variable(s), we conduct diagnostic tests of instrument relevance based on the significance of the excluded instrument in the first-stage reduced form regression [51]. With one instrument, the general rule of thumb is to reject the null of weak instrument if the F statistic is at least 10. As sensitivity analysis and to better understand impact pathway from program participation to household diet, we conducted three additional tests. First, we employ three different specifications of the system of Eqs (3) and (4). The first uses Simpson’s dietary diversity index constructed based on purchased food consumed inside the household, instead of Simpson’s diversity index based on all food consumed used in the main specification. The second and third specifications replace the production diversity indicator with the average value of maize, and legumes and nuts harvested per hectare, respectively. These alternative specifications allow us to assess the extent to which the effect of program participation on dietary diversity is driven by increase in the diversity of purchased food (as opposed to own consumption) and the contribution of program target crops.

Second, we estimate Eqs (3) and (4) using two household-level production and dietary diversity indices based on Shannon’s diversity index constructed for agricultural production diversity and household dietary diversity. While the association between production and dietary diversity may depend on indicators used to measure diversity [17,18], several food-group based indicators of dietary diversity are found to be positively correlated with each other and with food and nutrition security [52].

Third, we use inverse probability weighting with regression adjustment (IPWRA) [53] to estimate average treatment effect on the treated (ATT) and average treatment effect (ATE). IPWRA addresses the endogeneity associated with self-selection into treatment by modelling both treatment selection and outcome variables rendering it a “doubly robust” estimator, meaning that either the treatment or outcome (but not both) must be correctly specified to consistently estimate treatment effects [54]. IPWRA is consistent if either the selection or outcome models are correctly specified and is more efficient, especially relative to weighting adjustment, if the outcome model is correctly specified [53,55]. The selection model controls for area of nearby plots (the instrument) and household characteristics discussed above while the linear model for dietary diversity controls for household characteristics, climatic variables, self-reported shock experience, as well as production diversity. Controlling for production diversity minimizes omitted variables bias in the and mimics the IV mediation analysis where the mediator is production diversity.

While the conditional mean independence assumption for matching is inherently untestable, we assess matching quality based on Rubin’s bias (B) and ratio of variance (R) [56] and propensity score distributions before and after matching using box plots and density function. Rubin’s B refers to the absolute standardized difference in the means of the propensity score between beneficiary and control groups while Rubin’s R is measured as the ratio of variances of the propensity scores between the beneficiary and control groups. Rubin’s R should be below 2 to avoid over-correction of bias and above 0.5 to prevent under-correction, while Rubin’s B should be below 25. Robust (for IV mediation and reduced form models) and bootstrapped (for IPWRA) standard errors are reported.

Results and discussion

Before presenting regression results, Table 1 compares selected socioeconomic and biophysical variables by treatment status mimicking balance t-tests of baseline variables. To recall, survey data were collected right after beneficiaries obtained their first harvest as a program beneficiary. But we maintain that variables reported in Panel A of Table 1 are unlikely to have been affected by program participation given the short time lapse. Beneficiaries are more likely to have larger family size, be male headed, better educated, and less likely to be poor where poverty is measured based on durable agricultural and non-agricultural assets, and quality of dwelling conditions (Table 1, Panel A). Total land size operated as well as land size within 15 minutes of travel from the homestead -our selected instrument- are also greater for beneficiaries. We also observe some differences in biophysical conditions that may affect agricultural potential (Table 1, Panel B).

Table 1

Descriptive summary of socioeconomic variables by beneficiary status.

	(1)	(2)	(3)
	Beneficiary	Control	Stat. sign.
Panel A. Sample characteristics
Household size	4.97	4.59	***
Household head age (years)	45.8	45.3
Female households (%)	27.0	33.8	**
Average adult years of education (years)	5.20	4.72	***
Number of adults (age> = 15)	2.66	2.45	***
Number of children (age<15)	2.30	2.13	*
Land size (ha)	1.20	0.86	***
Total land area within 15 minutes of travel	1.18	0.51	***
Poor households based on durable agr. assets excluding land (%)	29.5	40.7	***
Poor households based on durable non-agr. assets (%)	26.4	40.7	***
Poor households based on dwelling condition (%)	35.8	62.8	***
Livestock (Tropical Livestock Units)	0.45	0.21	***
Distance to basic services index	0.036	-0.044
Remote households (%)	34.5	32.3
Panel B. GIS variables
Elevation (meters)	864.6	945.6	***
Slope (degrees)	1.25	0.98	***
Total annual rainfall (millimeters)	931.5	919.2	***
Average monthly temperature (degree Celsius)	21.4	21.0	***
Observations	397	538	935

Note. Households are defined as poor if they fall in the lowest tercile of wealth index constructed based on durable agricultural (agri.) assets, durable non-agri. assets, or quality of dwelling condition. Households are defined as remote if they fall in the highest tercile of the index constructed based on travel time to various services. Columns 1 and 2 report means and column 3 reports statistical significance (stat. sign.) from mean comparison tests.

* p<0.1

** p<0.05

*** p<0.01.

Table 2 presents descriptive summary of variables that are likely to have been affected by the program. Beneficiaries are more likely to use manure and hired labor; operate larger number of intercropped plots, and are more likely to have received agricultural extension information in the preceding year (Table 1, Panel A). Program beneficiaries also have better agricultural performance in terms of maize yield, total value of crop harvested, net agricultural income, on-farm diversity, and marketed surplus (Table 2, Panel B). Inter-group differences are also observed in term of the share of households producing different food groups as well as the monetary value of food groups produced (see supplemental S1 Table). These gains in the value and diversity of agricultural production appear to have been translated into higher food consumption from own production as well as from purchases (Table 2, Panel C). Household dietary diversity based only on purchased food is also higher among beneficiaries. Food group-level summaries reported in supplemental S2 Table also show inter-group differences with beneficiaries having higher per capita consumption expenditure on ASF such as eggs, milk, and dairy products but lower consumption of fruits.

Table 2

Descriptive summary of agricultural and dietary outcomes by beneficiary status.

	(1)	(2)	(3)
	Beneficiary	Control	Stat. sign.
Panel A. Agricultural inputs and practices
Agricultural labor used (person-days/ha)	321.3	317.9
Household uses hired labor (%)	49.9	39.0	***
Household uses communal labor (%)	35.3	31.6
Inorganic fertilizers applied (kg/ha)	114.1	103.3
Number of intercropped plots	1.88	1.16	***
Received extension services (last year) (%)	91.9	41.4	***
Uses manure (%)	68.3	44.6	***
Panel B. Agricultural production
Maize yield (kg/ha)	2352.3	1813.6	***
Legume yield (kg/ha)	798.1	755.1
Value of all crops harvested (’000 MWK)	213.2	124.2	***
Value of maize harvested (’000 MWK)	98.2	78.3	***
Value of legumes and nuts harvested (’000 MWK)	41.0	29.9	***
Net agricultural income (’000 MWK)	172.8	103.2	***
Value of harvest sold (’000 MWK)	52.4	22.8	***
Percent of harvest sold (%)	23.2	18.0	***
Simpson production diversity index	0.41	0.31	***
Panel C. Household food consumption
Per capita annual food expenditure (’000 MWK)	67.2	55.3	***
Value of purchased food (’000 MWK)	96.8	83.6	***
Value of food from own production (’000 MWK)	148.6	91.1	***
Simpson household dietary diversity index	0.62	0.64
Simpson household dietary diversity index for purchased foods	0.67	0.63	***
Observations	397	538	935

Note: Columns 1 and 2 report means and column 3 reports statistical significance (stat. sign.) from mean comparison tests.

* p<0.1, ** p<0.05

*** p<0.01. MWK: Malawian Kwacha.

Beyond establishing linkages between on-farm production diversity and household dietary diversity, understanding the causal pathways through which more diverse production can lead to better diets is key for policy. The first -and direct- channel is through higher consumption of own produced food, while the second -and indirect- channel is through higher agricultural income and food purchasing power [57]. The strength of the latter channel depends on the availability of nutritious foods locally and their affordability.

In our case, the extent to which pre-existing differences between treated and control households shown in Table 1 mediate the interaction between production and dietary diversity plays an important role in shaping the causal relationship. Intermediate regression results reported in supplemental S3 Table show the effect of program participation, instrumented by area within 15 minutes of travel from the homestead, on the mediator (Simpson’s production diversity index) to be 0.41 controlling for other variables (S3 Table, Column 1a). The estimate of the effect of the mediator on the outcome, controlling for treatment and other factors, is also significant in both cases where Simpson’s household dietary diversity is measured using all foods consumed inside the household (S3 Table, Column 1b) and purchased foods (S3 Table, Column 1c).

Placebo test results from reduced form regressions implemented to assess the validity of our instrument are presented in S5 Table. As expected, the coefficient of the instrument is significant both in the production and dietary diversity models for the treated group (as well as the whole sample) but not for the control group. As our approach assumes that the treatment is endogenous with respect to the final outcome (dietary diversity) due to confounders that also jointly affect agricultural production, these test results provide indirect support for the validity of the exclusion restriction. That is, the effect of the program (that has not benefited the control group) on dietary diversity operates exclusively through the program’s effect on production diversity.

Results from the IV mediation analysis are reported in Table 3. Program participation increases the Simpson’s household dietary diversity by about 0.29 with the indirect -or mediated- effect estimated at 0.38 (Table 3, Column 1). This mediated effect accounts for 132% of the total effect of program participation on household dietary diversity, hence it is partly offset by the negative direct effect of program participation on dietary diversity. Albeit the fact that the mediated effect is larger than the total effect may appear counterintuitive, it has been noted that a positive total effect stemming from a positive (larger) mediated effect partly offset by a negative direct effect would be perfectly conceivable [19]. In our setting, for example, regardless of the production channel the direct effect could be negative likely owing to a substitution mechanism at play, for which an increase in dietary diversity driven by program participation is partially reduced via a decrease in consumption of other food groups. For instance, substantial time and labor investments in the production of crops targeted by the program may results in lower investments in the production, and consumption, of other food sources.

Table 3

Impact of program participation on production diversity and dietary diversity.

	(1)	(2)
	Simpson’s dietary diversity (all food)	Simpson’s dietary diversity (purchased food)
Total effect	0.290***	0.192***
	(0.077)	(0.074)
Direct effect	-0.094***	-0.033
	(0.033)	(0.031)
Mediated (or indirect) effect	0.384***	0.225**
	(0.130)	(0.113)
Observations	935	935
Kleibergen-Paap F-statistic for the excluded instruments in first stage one (T on Z)	31.17	31.17
Kleibergen-Paap F-statistic for the excluded instruments in first stage two (M on Z|T)	20.13	20.13
Mediation effect as a percentage of the total effect (%)	132.3	117.1

Note: Results from the IV mediation analysis are reported. Dependent variables in columns 1 and 2 are Simpson’s dietary diversity indices based on all food consumed and purchased food consumed by the household, respectively. T is indicator for program participation. M is Simpson’s production diversity index. Excluded instrument (Z) is area of household plots within 15 minutes of travel. Control variables include household size, age and gender of the household head, average years of adult education, number of adults and children in the household, indices for dwelling condition and durable agricultural assets, temperature, slope, precipitation, and indicators for self-reported experience of droughts and crop diseases. Parameter estimates of exogenous controls not shown as they are partialled out using the Frisch-Waugh-Lovell theorem for ease of estimation. Eicker-Huber-White standard errors reported in parentheses.

*** p<0.01

** p<0.05, * p<0.1

On the other hand, beneficiaries seem to purchase more diverse foods from the market, relative to the control group, highlighting the role of the indirect -or income- effect of program participation on household diets (Table 3, Column 2). Both total and mediated effect sizes when diversity is measured based on purchased foods only account for approximately two-thirds of the increase in the Simpson’s index based on all foods consumed inside the household (Table 3, Column 1). This result suggests that the market channel is more important than higher food own-consumed generated by greater production diversity among beneficiary households in line with previous findings [4]. Moreover, the direct effect turns smaller and not statistically significant when the Simpson’s index of purchased food items is used as dependent variable, suggesting that the substitution effect is associated to a reduction in consumption of other food crops produced by the household. These findings persist when we measure production and dietary diversity using Shannon’s diversity index as shown in supplemental S4 Table.

To further examine pathways from program participation to dietary diversity, we re-estimate Eqs 5 and 6 using two alternative mediators (M) based on program target crops -the value of maize harvest per hectare and the value of pulse harvest per hectare. The mediated -or indirect- effect of program participation on Simpson’s dietary diversity index based on all food consumed through maize harvest is significantly higher than that through pulse harvest (Table 4, Columns 1 and 2). When the Simpson’s dietary diversity index based on purchased food is used as dependent variable, the mediation effect through maize harvest is still statistically significant, unlike the same effect through pulse harvest (Table 4, Columns 3 and 4). Perhaps more importantly, the direct effect is positive and significant when pulse harvest value is used as mediator variable, suggesting other important channels affecting overall positive total effect that are not captured by the mediator, such as maize income. Overall, these findings suggest that the positive impacts of the program -both through enhanced market-related purchasing power as well as through production-led increases in own consumption–on dietary diversity are primarily generated by a propulsive effect of program participation on maize profitability.

Table 4

Impact of program participation on dietary diversity, as mediated by the value of maize and pulse harvests.

	(1)	(2)	(3)	(4)
	Simpson’s dietary diversity (all food)		Simpson’s dietary diversity (purchased food)
Total effect	0.299***	0.290***	0.201***	0.192***
	(0.080)	(0.077)	(0.076)	(0.074)
Direct effect	-0.036*	0.091***	0.001	0.075***
	(0.020)	(0.031)	(0.019)	(0.027)
Mediation (or indirect) effect	0.335***	0.199*	0.200**	0.116
	(0.121)	(0.103)	(0.100)	(0.072)
Observations	935	935	935	935
Kleibergen-Paap F-statistic for the excluded instruments in first stage one (T on Z)	31.16	31.17	31.16	31.17
Kleibergen-Paap F-statistic for the excluded instruments in first stage two (M on Z|T)	27.71	17.31	27.71	17.31
Mediation effect as a percentage of the total effect (%)	111.9	68.56	99.31	60.68

Note: Results from the IV mediation analysis are reported. Dependent variable in columns 1 and 2 are Simpson’s dietary diversity index based on all food consumed. Dependent variable in column 3 and 4 is Simpson’s dietary diversity index based purchased food consumed by the household. M in columns 1 and 3 is per capita value of maize harvest in thousands of Malawi Kwacha (MWK). M in columns 2 and 4 is per capita value pulse harvest in thousands of MWK. T is indicator for program participation. PD is Simpson’s production diversity. Excluded instrument (Z) is area of household plots within 15 minutes of travel. Control variables include household size, age and gender of the household head, average years of adult education, number of adults and children in the household, indices for dwelling condition and durable agricultural assets, temperature, slope, precipitation, and indicators for self-reported experience of droughts and crop diseases. Parameter estimates of exogenous controls not shown as they are partialled out using the Frisch-Waugh-Lovell theorem for ease of estimation. Eicker-Huber-White standard errors reported in parentheses.

*** p<0.01

** p<0.05

* p<0.1.

S1 and S2 Figs compare distribution of estimated probabilities of treatment (propensity scores) used to estimate ATT and ATE based on IPWRA. The distributions are more comparable after matching as depicted by both box and kernel density plots. Matching reduced Rubin’s B from 56 to 12 and Rubin’s R from 1.8 to 1, both in the recommended range. ATT and ATE estimates are reported in S6 Table. Only ATT is marginally significant (at 10% level) when dietary diversity is measured based on all foods consumed inside the household (columns 1 and 2), while both ATT and ATE are significant when diversity is measured based only on purchased foods (columns 3 and 4), although ATE is only marginally significant (column 4).

Stronger effect on the diversity of purchased foods is consistent with results from the IV mediation analyses in Table 3 where approximately two-thirds of the increase in the Simpson’s dietary diversity index ‒computed based on all foods consumed inside the household‒ was due to increase in the diversity of food purchased from the market. On the other hand, the magnitude of impact estimates based on IPWRA is smaller than that from IV mediation analysis. While both IV and risk adjustment (RA) approaches such as IPWRA are designed to mitigate confounding bias in non-experimental methods for impact estimation, the two approaches are not directly comparable and IV estimates cannot be truly interpreted as ATT or ATE [58]. When treatment effects are homogeneous across the target population or, if heterogeneous, are unrelated to treatment assignment, RA and IV estimates produce comparable results when corresponding identifying assumptions hold. On the other hand, when treatment effects are heterogeneous and potentially related to treatment assignment, RA and IV approaches may produce asymptotically different estimates as has previously been noted [58-60].

Our findings on the limited increase in own consumption of pulses appear to be in line with the literature. Indeed previous evidence from Malawi shows that pulse consumption is relatively inelastic to both income and pulse production [61,62]. Pulses are highly income inelastic among urban Malawian households, and even inferior goods among better-off households, showing an expenditure elasticity close to unity among rural households which may be related to the high value of international pulses trade [62]. Contrary to expectations, a study has found [62] a decline in per capita pulses consumption in rural Malawi despite the country’s large-scale FISP, where legume seeds are either subsidized or granted for free. In the program under analysis, beneficiaries are exposed to pulses-based technologies through demonstration field days, and therefore they might not have gained enough insights on their nutritional benefits.

Boosting agricultural production diversity is often considered a promising approach to improve dietary diversity for poor and vulnerable farmers, either by increasing availability of more nutritious food for subsistence-oriented smallholders, by enhancing their purchasing power, or both. Nonetheless, existing empirical literature on the linkages between production and dietary diversity is ambiguous. Earlier evidence points out that, while improvements in agricultural production diversity -and productivity- are necessary to enhance access to food and rural household income, they may not be sufficient to ensure dietary diversity, because agricultural innovations that increase the production of high-value and nutrient-dense crops could yield limited effects on their consumption due to the persistence of dietary habits and limited nutritional awareness. Several factors mediate the interaction between production diversity and dietary diversity, including market access, awareness about the nutritional content of targeted agricultural commodities, and intra-household decision making that a sound policy can effectively contribute to shape. Evidence is also mixed on the role of the direct and indirect impact pathways. For example, while the association between production and dietary diversity was found to be due to the direct pathway in a study from Ghana [12], the income pathway was found to be relevant in a cross-country study that includes Malawi [4].

Each of these factors require different policy course. Our findings of higher maize yield, higher value of maize and pulse harvest, higher net agricultural income, and higher crop sale (in levels and as a share of total harvest) point towards the importance of increasing the productivity and profitability to enhance market purchasing power. Efforts aiming at reducing barriers to better integration into output markets (e.g., limited information about prices and high transportation costs) could enhance participation in profitable markets and boost household dietary diversity. On the other hand, these market-oriented actions should be complemented with an active soil fertility monitoring, aimed at increasing nutritional content of crops that are otherwise sourced from the market with a consequent welfare-decreasing effect. Hence, the initial specialization strategy in the main staple crop should be accompanied by a diversification strategy to the extent that the new legume crops attain desirable physical properties to be able to substitute market-sources commodities.

During this process, attention should be given to intra-household decision making regarding the production, marketing, and consumption of different commodities. For higher production diversity to translate into more diverse diets, it is crucial that both men and women are aware of the nutritional values of different commodities. Specific to Malawi, for example, one study finds that in households with both adult men and women, informational campaigns including about nutrition that jointly target men and women have a stronger effect on household food security relative to campaigns that target only one gender [63]. Embedding adequate gender considerations will be especially important given the increasing role Malawian men are playing in decisions about food purchases as well as food preparation [64].

Our findings highlight once more that regardless of the strength of the linkages between production and dietary diversity, sequential actions in food policy would be necessary to address a host of complementary, and not contrasting, factors to maximize household-specific comparative advantages in crop cultivation, with the additional benefits of adoption of new crops and varieties that further increase the productivity and profitability of crops already grown by the household. The mediation role of pulses’ adoption driven by the program under analysis is indeed a case in point for boosting crop diversification policy.

Conclusion

We investigate the statistical associations between production diversity and dietary diversity in Malawi using cross-sectional household survey data collected as part of an environmentally sustainable agricultural intensification research program from program beneficiaries and a random sample of non-beneficiary, and pure control households. Program beneficiaries test various innovations and practices, including sole maize with different fertilizer application rates, manure application, multiple legumes, maize–legume intercropping, and intercropping between two legumes.

Descriptive evidence shows that program beneficiaries were systematically different from control households along several dimensions considered. They were also able to attain higher value of agricultural production and net agricultural income as well as more diverse production, compared to the control group. Considering the systematic differences documented between treated and control groups, we employ an instrumental variables (IV) mediation analysis framework to estimate the impact of program participation on household dietary diversity. While traditional IV framework is used to identify unbiased impact estimates based on observational data, it does not allow us to unpack causal impact when both treatment -program participation in our case- and an intermediate outcome -on-farm production diversity-, jointly cause a secondary outcome -household dietary diversity.

Results point to a positive and significant impact of the program on dietary diversity, which is mainly driven by the increase in production diversity of beneficiary households. In sum, participation led to a more diverse agricultural production, and, through this mediating channel, to an increase in dietary diversity. However, this increase does not seem to be primarily related to cultivation of pulses, the main crop group targeted by the program, which are rich in crucial micronutrients and contain better and higher quality protein than other grains. Rather, positive effects are mostly associated to maize production, a secondary program targeted crop. Enhanced maize production boosts both own-consumption and, perhaps more importantly, agricultural income, allowing households to purchase and consume more diverse food items.

We document a weak association between production and own consumption of pulses, underscoring the importance of complementing production-oriented programs with demand-side interventions to promote nutritional awareness. The positive and significant correlations between production diversity and diversity of purchased food highlights the importance of access to food markets for increasing and reinforcing nutritional gains associated to enhanced on-farm production diversity. While our study provides useful insights on the linkages between agriculture and nutrition, it does not address potential intrahousehold reallocation and inequalities in food consumption. Also, diets may shift over time, especially due to nutritional education, albeit only in the medium to long-term. Also, the effects we document may be limited to the short-run due to limited time elapsed between beginning of the program and data collection -just one completed cropping season-.

Lastly, some caveats are necessary regarding the interpretation of our findings. While the novel empirical approach adopted allows us to overcome econometric challenges related to selection bias, endogeneity, simultaneity, and the cross-sectional nature of the data, the results provided here should nonetheless be interpreted with caution: the significant associations and correlations found are suggestive of causal relationships, but causal claims will have to be more thoroughly supported by new empirical evidence when follow-up data finally become available. Therefore, a key area for future research is a longitudinal analysis of longer-term effects of the program to examine whether and to what extent gains in knowledge, adoption of improved innovations, and environmental services can bring about longer-term effects on production and consumption patterns of beneficiary households.

Box plots of propensity scores before and after matching.

(TIF)

Click here for additional data file.

Kernel density plots of propensity score before and after matching.

(TIF)

Click here for additional data file.

Value of agricultural production by food group.

(XLSX)

Click here for additional data file.

Values of consumption expenditure by food group.

(XLSX)

Click here for additional data file.

Intermediate (first stage) results from IV mediation analysis.

(XLSX)

Click here for additional data file.

Impact on production diversity and dietary diversity (by diversity index).

(XLSX)

Click here for additional data file.

Reduced form regressions for production and dietary diversity.

(XLSX)

Click here for additional data file.

ATT and ATE effects on household dietary diversity (IPWRA).

(XLSX)

Click here for additional data file.

26 Nov 2021

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Gideon Kruseman, Ph.D.

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide

3. Please update your submission to use the PLOS LaTeX template. The template and more information on our requirements for LaTeX submissions can be found at http://journals.plos.org/plosone/s/latex

4. We note that you have referenced (ie. Bewick et al. [5]) which has currently not yet been accepted for publication. Please remove this from your References and amend this to state in the body of your manuscript: (ie “Bewick et al. [Unpublished]”) as detailed online in our guide for authors

http://journals.plos.org/plosone/s/submission-guidelines#loc-reference-style

5. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: I Don't Know

Reviewer #3: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Reviewer’s Comments

The manuscript, is an interesting piece of research. This paper analyses the causal effects of on-farm production diversity on household dietary diversity in Malawi based using microdata collected as part of a participatory agricultural intensification program. This is a relevant topic in the body of literature particularly due to the fact that nutrition plays such a significant role human capital development and consequent productivity. Poor nutrition especially for children has long-lasting consequences to the society. After reviewing this paper, I have the following observations concerning the suitability of the article for publishing in the PLOS ONE journal.

In what follows I detail my suggestions for further improvement of the paper.

Abstract

Needs minor improvements.

1. The word commodities in the sentence “Among other innovations, the program aims to promote the production of pulses, commodities that are rich in micronutrients, have better-quality protein, and have nitrogen fixation benefits compared to other grains” can be removed with no alteration in meaning.

2. In the sentence “We find a significant positive treatment effect on dietary diversity, led by an increase in production diversity.” The treatment should be stated directly instead of letting the reader figure it out.

Introduction

3. “An important lesson from these studies is that while a diversification strategy has the potential to improve diets and nutrition, the strength of the association is quantitatively controversial.” In this statement authors have not stated what these quantitative controversies and makes the consequent paragraph on filling in this gap invalid. Authors should clear state what these controversies are and how their study solves those controversies if not justify why and how their study cannot be counted as another adding more controversy.

4. The statement “while 19% and 40% of the country’s population and children below the age of five years are estimated to be undernourished and stunted, respectively” is not clear whether the figures refer respectively to country population and children or undernourished and students.

5. The sentence “Analysis of the linkage between production diversity and dietary diversity in a crosssectional setting is challenging due to a host of confounding factors …….”

Should be moved from its current paragraph and be used as an opening sentence in the next paragraph

6. The finding that “While production diversification can lead to more diverse diets, increasing production of more nutritious crops such as pulses may not necessarily translate into their greater own consumption, due to the persistence of dietary habits or other location specific factors” should be more elaborately explained; why then should increase in income change dietary diversity despite “the persistence of dietary habits or other location specific factors”

Materials and methods

7. Authors need to explain why they have excluded group 2 in the analysis. Obviously if groups 1 and 2 were targeted and only those in group one engaged in the project then you have self-selection. But then the random (as stated) targeting with non-compliance only on the treatment side could provide the Intention To Treat (ITT) which is of interest. In case of non-compliance on both the treatment and the control side the random (as stated) targeting could be used as a clean instrument for participation in the program.

8. Given the nature of the instrument variable (Z) that is used then it makes sense if group 2 is included in the analysis. Because if only groups 1 (program participant) and group 3 (control) are used then there is no way the instrument will be explaining participation (by design) since the control group did not get exposure to the program.

9. Second paragraph of page 6, on-farm production should be removed “Diversity in on-farm production and household diets is measured based on the Simpson’s diversity”

10. There are different measures of production and dietary diversity, authors should argue why they chose to use the Simpson’s index over others.

11. The use of values (instead of quantities) to construct the Simpson’s Index is prone to being affected by food items with high values; authors should also explain why we should not be worried about this. In addition they should also explain how these values were obtained (market price? How about those which were not purchased/sold?)

Results and Discussion

12. A number of variables presented in Table 1 could have been affected by the program despite the claim by authors that these variables are unlikely to be affected by the program. For example wealth (measured by assets), dwelling quality, all agricultural inputs and practices variables. Since the program targeted agricultural production, obviously agricultural practices will be affected, incomes and wealth accumulation.

13. Page 14, monitory �  monetary

14. That the claim is the effect ion dietary diversity is mainly through the indirect- channel is through higher agricultural income and food purchasing power. It would be crucial to show empirically whether the program had an effect on household income.

Reviewer #2: The paper touches a timely and relevant topic. Although the connection of production and consumption diversity has attracted some attention in the rural development literature, findings on the relationship are mixed. The paper therefore provides a valuable contribution to the existing body of literature. Despite its thematic appeal and well-presented structure - in view of this review - the manuscript should undergo some major revisions to be considered for publication. Main concerns include the claim of causality when interpreting regression results throughout the paper, the suitability of the selected instrument in the regression framework, the lack of background information on the Africa RISING program, as well as the absence of a comprehensive review of the literature on the linkages of production and consumption diversity.

Please find more detailed comments below. Major comments are listed first, minor comments are listed further below. (Line numbering of the manuscript would have been appreciated).

Major comments:

1. Instrumental variable is not convincing. Authors use area of agricultural land within 15 minutes of homestead to instrument household participation in the program. Authors do not argue convincingly why they expect agricultural land within 15 minutes of the homestead to only influence production diversity through program participation. Cropping patterns may differ with increasing distance from a plot to the homestead, e.g., households may cultivate more diverse crops on their plots closer to the homestead - irrespective of program participation. The authors do not present data on production diversity in close to homestead plots for beneficiaries and control households. Authors also do not present information on how the program aimed at increasing production diversity among its beneficiary households. More background information would be needed to support the argumentation for the instrument of choice. In its current narrative, the selection of the IV is not convincing (see also further comments on the IV selection below).

2. Claims of causality. The authors employ IV mediation analysis, which is an appealing tool, especially in the context of analysis of complex interactions - here participation in a program, agricultural production diversity, and household dietary diversity. Through its mediation framework, the tool allows the analysis of different pathways from the treatment to the outcome. Authors also provide different mediators and outcomes to complement their analysis. However, in several parts of the paper, authors claim causality. In context of the complex relationship and the previous comment on IV selection, authors should consider to be more cautious in interpretation of results (e.g., by speaking of associations/correlations instead of causation).

3. Framing of the story. The authors frame the story around PP – PD – DD (program participation- production diversity-dietary diversity). Yet, authors results (and interpretation thereof) indicate that positive associations between PP and DD are not necessarily associated with PD per se, but with increases in maize output and income, which in turn is used to increase HH DD. Authors claim in some parts of the manuscript that PD leads to higher DD, which should be rephrased to avoid misinterpretation.

4. Lack of background information on Africa RISING. Although the program is introduced briefly in the paper, the authors may want to provide more information on Africa RISING. In its current form, reader does not learn what the program does in terms of promotion of legume and/or maize production, when interventions started, from when the data presented stems, how beneficiaries were selected (targeted?) into the program, intervention logic, and modes of implementation. The presentation of program objectives, mode of beneficiary selection, location, timing, and concrete interventions are essential to contextualize the findings of regression results.

5. Lack of comprehensive review of literature examining linkages between production and consumption diversity. Although authors briefly refer to existing literature on the topic (e.g., last sentence of last paragraph of introduction p.3 and last sentence of following paragraph, p.3), the manuscript does not provide a comprehensive review of their findings. Authors should include such a review – also highlighting that findings are inconclusive and inconsistent, also pointing to the fact that the relationship (production diversity and consumption diversity) is likely to be highly complex and context dependent.

Minor comments (in order of appearance in the manuscript- line numbering would have been appreciated):

6. P.4 “We tackled these challenges by using household survey microdata […]”. Authors correctly point to the challenges associated with the examination of linkages between production and consumption diversity in the previous sentence. The cited sentence claims to present how these problems are tackled, yet it only presents the data source.

7. P.5 “In this study we use recently developed mediation analysis […]”. Authors may want to include a reference in relation to the method, as well as references in which the method has been applied in similar contexts

8. P.5 “The underlying mechanism, however, appears to be through higher purchase of more diverse food […]” and “[…] these effects are primarily due to an increase in the production of a secondary target crop, maize, and not to higher pulse production.” Considering this finding, authors may want to rephrase statements in the manuscript which may be misinterpreted by readers in the sense that diversification of production leads to diversification in production. E.g., abstract: “These findings highlight that, while diversifying production can lead to more diverse diets […]”.

9. P.5 Before Materials and Methods section, authors may consider providing background information on Africa RISING (also see comment #4).

10. P.6 “This study analyzes data […]”. Please explain why group 2 households were not considered in the analysis.

11. P.6 “Agricultural production data refer to […]”. Please briefly discuss advantages and disadvantage of 7-day recall approach. Please also reflect on how the harvest cycle/period of cultivated crops coincides with the concrete dates of the recall period. That is can households consume own produce in the 7-day period prior to interview.

12. P.7 Identification. Please define T, M and Y as used in the manuscript, when mentioned for the first time.

13. P.8 “[…] validity, exogeneity, and strength [of the IV]”, replace by relevance, exclusion restriction, and independence assumption?

14. P.9 “We posit the partial identifying assumption to hold in our setting, […]”. This assumption merits further discussion. e.g., what variables do authors expect to jointly affect T and PD (to confound DD through PD)? Why do authors expect other confounders of DD (not through PD) to be of less importance?

15. P.9-10 “[…] we estimate Equations 5 and 6 using 2SLS […]”. Considering the complex relationship between PD and DD, and the IV at hand, estimation of the causal chain seems quite ambitious. Authors may consider referring to associations instead of causal chains.

16. P.10 “Our choice of an exogenous instrument for program participation is guided by program design, […]”. As noted in an earlier comment, the manuscript does not contain much information on program design. Please include, also to support your argument for the IV.

17. P.10 “Specifically, we use total land area operated by the household […]”. It would be interesting to see whether there are differences in cropping patterns on these plots between participants and non-participants.

18. P.10 “[…] land area within 15 minutes travel would not affect DD unless through program participation.” This is a strong statement and goes along with major comment #1. There may be channels other than program participation through which land area within 15 minutes of homestead may affect DD. E.g., if households tend to cultivate certain crops more often on plots closer to their home, and if these crops tend to be marketed/ consumed more often, DD could also be affected by (pre-existing) cropping patterns on these plots, irrespective of program participation.

19. Page 11 The reference to Fig. 1 appears twice.

20. P.12 “Since survey data were collected […]. Please indicate when data was collected and when program activities started.

21. P. 13 Table 1. Number of observations (n=935) does not match with the total number of observations presented in the data section (N=1149). Please explain.

22. P. 14 Table 2. Table 2 shows that beneficiaries tend to have higher maize yields, higher value of maize harvested, and higher net agricultural income with relative importance of income from maize >> income from legumes. How does maize production fit into the assessed relationship between production diversity and dietary diversity? From the manuscript, it seems like the program mainly tries to introduce additional legumes into household production. What is the role of maize production?

23. P.15 “As noted in previous studies, […]”. The authors may want to contextualize their findings regarding the sources included in the cited literature (e.g., sources 1-7). E.g., literature analyzing linkages between on -farm production diversity and household consumption diversity has produced mixed findings: Ecker (2018) finds that production diversity is associated with HH consumption diversity mainly through direct pathways (HH consuming own produce), while Sibhatu et al. (2015) find that increase in HH consumption diversity is associated with increased HH income (and not by consumption of own produce).

24. P.16 “This result suggests that the market channel is more important […]”. Please also put into context of literature on PD and DD.

25. P.22 “Results point to an overall positive impact of the program on dietary diversity […]”. In the context of results presented earlier (direct effect being outsized by indirect effect), this formulation may be misleading. Consider rephrasing.

26. P.22 “Enhanced maize production boosts both own-consumption and, perhaps more importantly, agricultural income, allowing households to purchase and consume more diverse food items.” Would this finding not imply that PD does not play a large role in DD. After all, Maize is the most grown cash crop in Malawi, and as such not associated with PD (or is it?). How would T (program participation)->M (production diversity)->Y (dietary diversification) hold in this case? What does the program promote in terms of maize production? Does PD even matter?

27. Supplementary Material. Results from models S1(1b) and Table 3 (1) are not the same, which they should (same T, M, and Y). Please explain, also why number of observations is not the same for both models (S1 n=931, table 1 n=935).

Reviewer #3: Overall:

The study design (cross-sectional survey) does not seem to adequately measure program participation, and it seems unlikely that there would be significant impacts on production after just 1 year of the study, so the overall results are not very compelling or convincing. The authors need to provide more information about the intervention and study design to be able to justify their conclusions. Further, there are several recent studies in Malawi that are not included in the paper and the authors reference to this as an ‘agricultural intensification’ study seems at odds with the diversification emphasis. Finally, gender dynamics in agriculture need to be adequately addressed in the paper.

Introduction:

- The authors conclude from their brief review of the linkages between diversified farming systems and nutritional outcomes is that ‘An important lesson from these studies is that while a diversification strategy has the potential to improve diets and nutrition, the strength of the association is quantitatively controversial.’ [italics added]. My understanding of the literature is somewhat different from this conclusion. Recent reviews of a considerable literature on this subject have all shown a positive, significant association between production diversity and dietary diversity for farming households, including 2 reviews included in the literature review (Jones 2017; Sibhatu et al. 2015). The question is perhaps to what extent other factors matter more or mediate that relationship, and several studies have pointed to the significance of factors such as gender relations in mediating that relationship, including studies done in Malawi (e.g. Bezner Kerr et al. 2019).

- The authors refer to their intervention as ‘agricultural intensification’ but from the description the intervention appears to be crop diversification. Crop/farm diversification is often contrasted to intensification in the literature, because intensification usually entails monocropping and increased input use rather than diversification. Diversification, in contrast, or diversified farming systems, is often linked to an agroecological approach and has been a subject of policy debate for farming and food systems because of many environmental services as well as food security and nutritional outcomes. This broader debate within which diversification of food systems sits might be briefly noted in the literature review. Some possible references:

o Bezner Kerr, R. et al. 2021. Can agroecology improve food security and nutrition? A review. Global Food Security 29. https://doi.org/10.1016/j.gfs.2021.100540

o Rasmussen, L.V., et al. 2018. Social-ecological outcomes of agricultural intensification. Nat. Sustain. 1, 275–282. https://doi.org/10.1038/s41893-018-0070-8

- The authors could integrate more recent studies that examine these relationships in Malawi or similar contexts that were not included in this review:

Madsen, S. et al. 2021. Explaining the impact of agroecology on farm-level transitions to food security in Malawi. Food Security 13: 933–954.https://doi.org/10.1007/s12571-021-01165-9

Kansanga, M.M. et al. 2021. Agroecology and household production diversity and dietary diversity: Evidence from a five-year agroecological intervention in rural Malawi. Social Science and Medicine 288, 113550. https://doi.org/10.1016/J.SOCSCIMED.2020.113550

Santoso, M.V. et al. 2021. A nutrition-sensitive agroecology intervention in rural Tanzania increases children’s dietary diversity and household food security but does not change child anthropometry: results from a cluster-randomized trial. Journal of Nutrition.

https://doi.org/10.1093/jn/nxab052

Bezner Kerr, R. et al. 2019. Participatory agroecological research on climate change adaptation improves smallholder farmer household food security and dietary diversity in Malawi. Agriculture, Ecosystems and Environment 279: 109-121. https://doi.org/10.1016/j.agee.2019.04.004

Snapp, S. S.,&Fisher, M. (2015). "filling themaize basket" supports crop diversity and quality of household diet in Malawi. Food Security, 7(1), 83–96.

Methods:

There is very limited information about the intervention. Who was involved, and what support was provided to the household? How was it participatory?

How was participation measured in the study – simply that they were enrolled? In which case how was lack of participation taken into account? Other mediation studies have much more detailed information about the intervention in order to determine the overall relationships, eg. Cetrone et al. 2021. Food security mediates the decrease in women's depressive symptoms in a participatory nutrition-sensitive agroecology intervention in rural Tanzania. Public Health Nutr. 24(14):4682-4692. doi:10.1017/S1368980021001014. Epub 2021 Mar 12. PMID: 33706829.

Results

Given that this study took place only 1 growing season after the intervention began, it seems somewhat premature to anticipate impacts on dietary diversity from production diversity – it may take 2 or more growing seasons for farmers to realize some of the longer term impacts from diversified production systems, including production as well as improved soil fertility, reduced pests and diseases etc. In addition they may not be able to produce significant enough yield of the new crop in the first season to generate enough for sale, seed and consumption and so may prioritize one of those options. These aspects need to be discussed in relation to your results.

The authors state that ‘This result suggests that the market channel is more important than the increase in food own-consumed generated by enhanced production diversity..’. This is a cross-sectional study, so the survey results do not demonstrate that there has been ‘enhanced production diversity’ only that participants grow more food, they might have been doing so prior to the intervention.

The authors also seem to somewhat misrepresent the findings of a review on agricultural diversification and dietary diversity. They indicate that “on-farm diversity has a small and nonlinear association with dietary diversity” and cite Jones (2017). The nonlinear association is true, pointing to several different pathways discussed in the review by Jones (2017), but note that the author found that: “agricultural biodiversity has a small but consistent association with more diverse household- and individual-level diets, although the magnitude of this association varies with the extent of existing diversification of farms. Greater on-farm crop species richness is also associated with small, positive increments in young child linear stature.”

Discussion

The authors seem to mention gender issues as an afterthought at the end of the section, and only in relation to involving both men and women. Previous research in Malawi and in other places has demonstrated that unequal gender dynamics within households can also influence dietary outcomes, such that increased production diversity may not translate into consumption of the food products (see Ruel and Alderman 2013 for example). Addressing gender dynamics as part of a nutrition-sensitive intervention has also been shown to increase the likelihood of improved dietary diversity (e.g. Bezner Kerr et al. 2019; Santoso et al. 2021). this aspect of gender inequality is not discussed in the paper and does not seem to have been taken into account adequately in the study design or analytical methods.

Ruel MT, Alderman H. Nutrition-sensitive interventions and

programmes: How can they help to accelerate progress in improving

maternal and child nutrition? Lancet 2013;382(9891):536–51.

The authors state that “Our results point towards the importance of increasing the productivity and profitability ‒ including of Malawi’s main staple crop maize ‒ to enhance market purchasing power.” Please explain how your results point to the importance of increasing maize production – given the dominance of maize in production and consumption, it is not clear how your results point to such a finding.

They also state that “Efforts aiming at reducing barriers to better market integration (e.g., limited information about prices and high transportation costs) could enhance participation in profitable markets and boost household dietary diversity.” The results presented do not appear to say anything about market integration, so it is not clear how these conclusions are drawn. Could not better market integration increase consumption of unhealthy imported purchased foods, as observed in many other parts of the Global South?

The authors go on to state: “On the other hand, these market-oriented actions should be complemented with an active soil fertility monitoring, aimed at increasing nutritional content of crops that are otherwise sourced from the market with a consequent welfare-decreasing effect. Hence, the initial specialization strategy in the main staple crop should be accompanied by a diversification strategy to the extent that the new legume crops attain desirable physical properties to be able to substitute market-sources

commodities.”

The results presented do not appear to provide any information about soil fertility or nutritional content of crops, so again it is not clear how such conclusions are drawn.

It is not clear how an intensification strategy would be accompanied by a diversification strategy since these often work at cross purposes, as a number of studies of the FISP program demonstrates:

Chibwana, C., Fisher, M., Shively, G., 2012. Cropland Allocation Effects of Agricultural Input Subsidies in Malawi. World Dev. 40, 124–133. https://doi.org/10.1016/j.worlddev.2011.04.022

The authors note that this is surprising, but based on my own knowledge and understanding of FISP from long-term research in Malawi, while legumes are officially a part of the program, they constitute a much smaller proportion of the total amount of seeds distributed, and the emphasis is on maize production.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Submitted filename: PONE-D-21-30466_Review.docx

Click here for additional data file.

13 Jan 2022

Please see the attached response to the three reviewers.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

9 Feb 2022

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Gideon Kruseman, Ph.D.

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: (No Response)

Reviewer #3: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors have addressed all the comments well and provided their responses clearly.

I congratulate them for that.

Reviewer #2: Thanks to the authors for their detailed responses to the comments. Most of the remarks made during the first review have been fully addressed. In particular, the manuscript now contains a section on program details, authors have altered wordings in relation to claims of causal relation and have linked their findings on the relation of production and dietary diversity to relevant literature. The manuscript continues to be well-structured, well-written and comprehensive in terms of data presentation and analysis.

Despite the reviews undertaken, the central comment during the first review – the choice of the instrumental variable - has only been addressed partially. Please find the main arguments against the IV in its current way of presentation below. Being fully aware of the complexity of discussions about IV selection, some suggestions to support the current IV are provided for the authors consideration.

1. Choice of IV.

It is acknowledged that authors have put great effort into the selection of an appropriate instrument for their analysis. The new manuscript contains additional tests (placebo test) in the supplementary material, and a comprehensive discussion of its results and implications for the analysis in the main body of the manuscript. Authors have also revised the Identification chapter and provided a theoretical background on IV choice in IV mediation analysis, based on which authors have motivated their IV choice. The arguments presented by the authors can be followed, and the instrument is certainly a suitable predictor for program participation (T) – as it was a precondition to be selected as beneficiary.

Yet the presented discussion and results are not (yet) conclusive regarding the overall suitability of the IV. This concerns mainly the exclusion restriction. Authors use land within 15 minutes of homestead to instrument program participation. In the context of smallholder agriculture, land holdings and locations of land holdings may be correlated with several (unobserved) factors that may simultaneously also influence crop and livestock production patterns, management practices, yield levels, output marketing, all of which ultimately may influence HH consumption decisions. In its current form, the manuscript does not show convincingly that treatment and control households do not differ systematically with respect to some unobserved features that influence both size of land holdings within 15 minutes of the homestead (the precondition to receive treatment) and the outcome (dietary diversity). This is not trivial, as descriptive comparison of treatment and control groups presented in tables 1 and 2, shows significant differences across several variables, e.g., treatment households tend to be better educated, less female led, have larger total land holdings, are less likely to be poor, and own more livestock. While it can be argued that certain differences are associated with the treatment itself, especially Simpson’s production diversity index and marketed value of legumes, other differences such as households’ land holdings, location of landholdings, and poverty incidence (based on both durable non-agricultural assets, and on dwelling condition) are likely independent of the treatment and therefore pre-existing. Thus, it cannot be ruled out that households with relatively more land holdings within 15 minutes of the homestead (essentially the treatment households) systematically share unobserved characteristics that affect their wealth status, consumption expenditures and dietary diversity. In this case, estimates of treatment effect on DD would be biased, most likely upward biased.

Authors have responded, with inter alia, a placebo test. As stated by the authors, this test provides indicative evidence that the size of land holdings within 15 minutes of the homestead seems to affect production diversity only through program participation (treatment) as the reduced form model yields an insignificant coefficient for the regression of PD on landholdings within 15 minutes of the homestead among control households. Thus, land holdings as such are not likely to directly affect production diversity of a given household. Yet, the results do not show whether there are any links between landholdings within 15 minutes of the homestead and dietary diversity other than through production diversity. In other words, while the test provides strong evidence for the instrument’s relevance, it does not provide support for the exclusion restriction to hold (the main critique of this comment).

To provide additional support for their choice of IV, authors may consider running the placebo test directly on the outcome (not on the mediator) to test whether the exclusion restriction holds (IV only affects DD through PD). For the treatment group the coefficient of such a test would indicate the “[..] effect of T on Y that operates exclusively through its effect on M” (citation from manuscript p.14). For the control group, the coefficient should not be statistically different from zero, as the only way through which landholdings close to the homestead would affect DD should be through PD (T). Results could support the statement made on p. 16 “we can safely assume that T is endogenous with respect to DD

due to confounders that affect DD primarily through PD”.

In addition, authors may consider running a (supplementary) propensity score matching, using available control variables, especially land holdings, wealth status, as matching parameters to estimate the ATE and ATT of the treatment on dietary diversity. Although bias due to unobserved pre-treatment differences may still not be ruled out, PSM would ensure that treatment effects are assessed based on similar HH in control and treatment groups, and magnitude of results could be compared with those of IV mediation analysis. In this context, authors mays consider using sensivity analysis to test for unobserved confounders.

Additional minor comments

2. Introduction and study setting.

Authors may consider sharpening the introduction and study setting chapters, by focusing on the relationship of diversification of agricultural production and household nutrition. Especially the opening sentences of the study setting chapter feel out of context (although content is certainly relevant for the study). Authors may consider opening the study setting chapter with an overview of the status of farms’ production diversity (or lack thereof) and continue with a description of how the objectives of the program under study aim at enhancing production diversity and what are expected outcomes (e.g., diversifying maize production).

3. Page 6 first paragraph.

The preceding paragraph states that findings are mixed across and within studies. It is not evident how the cited examples for South America support this contradiction of findings (tuber are promoted and consumed more often; livestock ownership leads to greater animal sourced food consumption).

Reviewer #3: The authors have addressed my concerns and comments. The paper provides a useful and relevant paper in this topical area.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

22 Feb 2022

Please see the attached response.

Submitted filename: Response to Reviewers-Feb192022.docx

Click here for additional data file.

11 Mar 2022

Plant different, eat different? Insights from participatory agricultural research

PONE-D-21-30466R2

Dear Dr. Haile,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Gideon Kruseman, Ph.D.

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: I thank the authors for their additional work undertaken in relation to the choice of instrumental variable. The discussion and supplementary analysis now more comprehensively support the IV used in the analysis. There are no further comments.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

16 Mar 2022

PONE-D-21-30466R2

Plant different, eat different?  Insights from participatory agricultural research

Dear Dr. Haile:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Gideon Kruseman

Academic Editor

PLOS ONE