The impact of agricultural innovation system interventions on rural livelihoods in Malawi

Abstract

This study, conducted in central Malawi, assessed the way a research intervention using an agricultural innovation system affected rural livelihoods. Propensity score matching was used to establish one village as a control, against which the impact of the intervention on two study villages [0]could be measured. Using the Enabling Rural Innovation intervention as a case study, it was established that rural livelihood outcomes pertaining to crop and livestock production, household income, asset ownership and fertiliser use were significantly improved by this intervention. In-depth analysis, however, demonstrated that although the participating households had more robust livelihoods during the intervention, when the research programme was phased out the effect was reduced. The authors recommend that local agricultural extension officers should receive more capacity building and budgetary support to ensure proper understanding of agricultural innovation systems concepts and correct application so as to sustain their positive effects.

Keywords:

• agricultural innovation systems
• rural livelihoods
• enabling rural innovation
• impact evaluation
• Malawi

1. Introduction

In many third world countries like Malawi, the problems of poverty and how to improve rural livelihoods are complex. There has therefore been a shift in global agricultural research to systems that enable greater individual and community innovation, proper use of knowledge and overall transformation (World Bank, 2007a; Sanginga et al., 2009; CTA, 2010). This shift was precipitated by the realisation that, despite stronger national research systems, agricultural productivity remains low for a number of reasons: lack of appropriate technologies and access to technologies; lack of inputs, credit and access to markets and rural infrastructure; and gaps in information and skills that prevent rural producers from adopting technologies and using them effectively. The new prevailing agricultural research paradigm strongly features agricultural innovation systems in national strategies for many countries working towards long-term agricultural development (Sanginga et al., 2009).

This has been the case in Malawi. Here, the government has created an enabling environment for agricultural research and development agencies who use agricultural innovation systems concepts to improve rural livelihoods, and thus reduce overall rural poverty, by helping smallholder producers to be innovative. Efforts like this have become critical, as nearly 40% of Malawi's population live in dire poverty (World Bank, 2007b), mainly because the prevalent smallholder agriculture relies completely on rain-fed farming and thus is prone to shocks caused by unfavourable weather conditions (World Bank, 2007b).

The literature contains only a few empirical studies conducted in Africa that specifically assess the way agricultural innovation systems help rural people to increase production by using natural resources more effectively (UN, 2008; Gildemacher et al., 2009), or to increase food security and nutrition (Morris et al., 2007), or to diversify their livelihoods and preserve the ecosystem (UN, 2008). The analytical methods used in the existing African studies are mainly qualitative and lack the rigor of the quantitative and qualitative methods used in developed countries' studies of the impact of innovation systems (Spielman et al., 2009). Furthermore, no empirical studies have applied rigorous quantitative methods to examine the impact of agricultural innovation systems interventions on the livelihood outcomes of rural producers in Malawi.

This paper presents the findings of an empirical study whose objective was therefore to assess the impact of agricultural innovation systems interventions on rural livelihoods in Malawi. The paper contributes to the ongoing debate on how agricultural innovation systems affect rural development and it aims to provide credible evidence, which can be used to inform policy, of the impact of such interventions on rural livelihoods.

2. The concept of agricultural innovation systems

An innovation system is a network of actors and organisations linked by a common theme with the aim of developing new technologies, methods and forms of organisation for end users to tackle identified problems (World Bank, 2007b). Such a system is governed by the prevailing institutions and policies that affect the performance of the actors involved and the regulation of the technologies developed (World Bank, 2007b). Hence an agricultural innovation system is a research tool for solving agricultural problems. The innovation systems concept embraces not only the researchers and scientists who are traditionally involved in agricultural research but also the end users of the technologies and the interactions that take place between all the actors in the research process (IAC, 2004:141).

In an agricultural innovation system, actors such as interdisciplinary teams of scientists, end users, extension agents and agribusinesses interact to identify problems for which innovative solutions are needed. This team is known as the Agricultural Innovation Platform. The interaction between the members of the platform is non-linear, with the various actors networking freely and institutional and human capacity development being key (Jones, 2008). Once problems are identified, the actors work together to develop technologies and adapt them to the local environment and this leads to improved production and ultimately better livelihoods. Throughout this process there is constant feedback using mechanisms that are put in place at the beginning of the process. Changes in the actors' attitudes, knowledge and skills are monitored as the actors interact and the results are used to develop strategies for improving their capacity to innovate (see Figure 1).

Figure 1: Representation of an agricultural innovation interface (author compilation)

3. The Enabling Rural Innovation intervention

The agricultural innovation system that this study investigated was the Enabling Rural Innovation (ERI) intervention developed by the International Centre for Tropical Agriculture (CIAT). In Malawi it was piloted in the three districts of Dezda, Lilongwe and Kasungu. ERI is an innovative research framework that helps resource-poor smallholder farmers to access market opportunities. Its main aim is to create an entrepreneurial culture in rural African communities.

The ERI ‘resource to consumption’ framework is designed on the following principles (Kaaria et al., 2008):

1.	Technology development and research agenda setting are based on the needs of beneficiaries, existing interests and available market opportunities.
2.	Technology development is guided by a comprehensive community assessment to identify the different intra-household allocation and control over resources and responsibilities and understand the constraints to and opportunities for technology adoption and reinvestment in natural resources.
3.	Gendered differences in roles and perceptions and differences in the roles of stakeholders are explicitly integrated into the technology development process to ensure equity in accessing technology and in distributing benefits.
4.	The ERI intervention helps communities to identify market opportunities and match them with existing community assets.

Kaaria et al. (2009) further add six key components for implementing the ERI intervention:

•	Agro-enterprise development and participatory market research
•	Farmer participatory research and natural resources management
•	Social and human capital development
•	Gender equity and empowerment of women
•	Community based participatory monitoring and evaluation
•	Effective development and management of partnerships.

Using the ERI guiding principles, the CIAT established one of three pilot innovation platforms in Malawi in the Ukwe Extension Planning Area in Lilongwe District. This research was carried out by a multi-disciplinary team of CIAT social scientists; extension agents from the Department of Agricultural Extension Services and local community-based extension staff; researchers from the Department of Agricultural Research Services; and other agricultural social scientists from the Ministry of Agriculture based in the Lilongwe Agricultural Development Division. The innovation platform worked together to select an appropriate community within the Extension Planning Area for piloting the ERI intervention. The criteria on which they based the selection were all-year-round road accessibility, availability of a motivated local level extension agent, willingness of other development partners working in the community to take an active role in the intervention, and community interest in further agricultural development and intensification (Sangole et al., 2003).

Once the community had been selected, the Ukwe innovation platform together with the community conducted a participatory diagnosis of the community challenges and opportunities and various development options. This initial engagement of the community was to sensitise them to the ERI intervention and to develop a shared vision for the future of the community (CIAT, 2007). From this was developed a collective plan of action for solving the identified problems using the available community resources and assets.

After the participatory diagnosis, the Ukwe innovation platform implemented the community action plan by forming a farmer research group and a farmer market group. The leaders of these groups would represent the community in the innovation platform. This was followed by a participatory market analysis of existing market opportunities, which culminated in the selection of a particular kind of agro-enterprise for intervention, based on the community's social and wealth differences and gender preferences.1 The farmer market group was trained in market research and this training guided their choice of pig keeping as the agro-enterprise to be developed under the ERI intervention.

Finally, through the farmer research group, the innovation platform planned and implemented simple research experiments using various other agro-enterprises. The aim was to build community capacity to conduct such experiments, to demystify the process of agricultural experimental research, to help the farmers understand their farming enterprises better, and to develop the farmers' ability to innovate.

4. Study methodology

4.1 Place of study and data collection

The study was conducted in Ukwe Extension Planning Area in Lilongwe District in Malawi's Central Region. Households were sampled from the villages of Katundulu, Mphamba and Kango. The first two were the intervention communities and Kango was the control. Katundulu was among the first villages where the ERI intervention was piloted, under the auspices of the agricultural innovation platform described above, while Mphamba was one of the villages where the local agricultural extension officers who had worked with CIAT in the Lilongwe pilot sites used the principles and concepts of ERI to implement similar interventions.

Purposive sampling was used to select the villages of Katundulu and Mphamba and then households in those villages were randomly selected. Households from the control village, which did not participate in the intervention, were also selected randomly. A total of 303 households were sampled in the study area, with the control sample size being double that of the intervention sample size to allow for better matching of households (Ravallion, 2003). Households from the control village whose socioeconomic characteristics and farming systems did not match those of the intervention villages were dropped from the analysis.

4.2 Data analysis

The key to a good impact evaluation is estimating what would have occurred in the absence of an intervention (Martinez, 2009). Since impact evaluations are carried out after the programme has started – or even after it has finished, as was the case in this study – ex post changes in outcome variables are used as a measure of impact. The problem with this is that there are many observable and non-observable time variant characteristics which may alter outcome variables for the participants. It therefore becomes difficult to attribute changes in the outcome variables to a specific intervention, since comparison of the before and after changes in the outcome variable can lead to either over- or under-estimation of the intervention's impacts.

To overcome the ‘attribution’ problem, it was therefore necessary to use data on outcome variables from a control group of non-intervention participants. To be valid, the control group had to have observed characteristics identical to those of the study participants, with the only difference being participation in the intervention programme. For this study, the observed characteristics were the households' socioeconomic characteristics and farming systems. The availability of data from non-participants is, however, in itself not sufficient for attributing differences in outcomes variables to an intervention, as changes in the outcome variables for participants may also arise from ‘selection bias’ in that participants may have been purposively selected (Ravallion, 2003, 2005). This entails that non-participants who are used for comparison purposes must, in addition to having identical characteristics, be those who would have had an equal chance of being selected for participation in the intervention – this overcomes the problem of selectivity bias. In the absence of randomisation, which equalises the probability of participating in an intervention and removes selection bias, propensity score matching becomes the solution to the problem of establishing a valid control group (Baker, 2000; Ravallion, 2003, 2005).

In the propensity score matching used in this study two groups were identified, one that took part in the intervention, denoted H_i = 1 and another that did not, denoted H_i = 0. Intervention households were matched to non-intervention households on the basis of the probability that the non-participants would have participated. This probability is called the propensity score. It is given by:

where X_i is a vector of pre-intervention control variables.

The pre-intervention control variables were farmers' knowledge of the programme under evaluation and the social, economic and institutional factors that might influence their participation in the intervention. The vector can also include the pre-intervention values of the outcome variables. Propensity score matching cannot reproduce the results of experimental randomisation designs if the variables that influence participation in the intervention are not properly defined.

In this study, propensity scores for each household in the sample were estimated using logit regression modelling. Using the estimated propensity scores, matched pairs were established on the basis of how closely the intervention and non-intervention samples' likelihood of participating in the ERI intervention matched. Unmatched non-intervention households were dropped from the analysis in order to remove bias and increase robustness (Rubin & Thomas, 2000, cited in Ravallion, 2003).

A logit regression model of participation in the ERI was estimated in order to determine the probability of a household participating in the intervention. Participation was modelled as a dichotomous dependent variable determined by a set of exogenous variables that determined participation in the ERI: frequency of contact with extension agents prior to the ERI intervention, the sex of the household head, an index of previous participation in other development interventions, and the size of the household.

5. Results and discussion

5.1 Validation of the logit model of participation in the ERI intervention

In general the logit model of ERI participation that was estimated was found to be a good predictor of participation, as demonstrated by the results of two alternative tests of goodness of model fit, the Hosmer and Lemeshow (H-L) static and the chi-square test (Table 1). The H-L goodness-of-fit test static was 10.310 and it was non-significant (p-value = 0.244), indicating that the model was a good fit, since the rule of thumb for accepting a logit model is that the H-L static must be greater than 0.05 and it should show non-significance (Hosmer & Lemeshow, 1989). The chi-square static of 23.747 showed that the model was statistically significant at the 1% confidence level, thus implying that all the predictors that had been included in the model were capable of jointly predicting participation in the ERI intervention.

Using propensity scores for participation generated by the logit regression model, households in the intervention villages were matched on the basis of the proximity of their propensity scores of participation to those of households in the control village. All other households whose propensity scores for participation were different from the range of scores for the intervention households were dropped from the analysis. Dropping all the control households whose probability of participation was very different from those of the intervention households made it possible to compare differences in livelihood outcomes between households that were similar and therefore comparable, and thus any differences in outcome variables between the participants and non-participants could be attributed to the intervention (Ravallion, 2003).

5.2 Impact of ERI on production outcomes

The ERI intervention affected many aspects of household production, with statistically significant differences being observed for livestock production, upland crop production, value of maize production and assets ownership. Differences in maize yields were, however, not found to be affected by participation in the intervention.

It was found that the intervention increased the value of all upland crops for participating households by US$812.34 and US$627.10 for the 2007/08 and 2008/09 cropping seasons respectively, and these differences were found to be statistically significant at the 1% and 5% confidence levels for the 2007/08 and 2008/09 seasons respectively (see Table 2).

Table 1: Results of the logit model of participation in the Enabling Rural Innovation intervention

	Coefficient (β)	Significance (ρ)	Odds ratio (Exp β)
Constant	–1.426	0.136	0.240
PRE_EXT_CONT	–0.157	0.019	0.855
SEX_HHEAD	0.571	0.118	1.771
HH_SIZE	0.092	0.087	1.092
PART_PREVIOUS		0.003
PART_PREVIOUS(1)	–0.559	0.310	0.572
PART_PREVIOUS(2)	1.486	0.068	4.419
Chi square	23.747***
H-L chi square	10.310 (p-value = 0.244)
Total households	303
*Significant at 10% level, **significant at 5% level, ***significant at 1% level.

Table 2: Impact of the Enabling Rural Innovation intervention on production outcomes

	Intervention		Control
Production outcomes (US$)	Mean	Standard deviation	Mean	Standard deviation	Effect
Total value of livestock	445.03	1 620.99	144.82	926.47	300.12*
Total value of upland crop production 2007/08	1 349.48	0.016	537.14	0.0114	812.34***
Total value of upland crop production 2008/09	992.24	0.0179	365.14	0.0084	627.10**
Value of maize harvest 2007/08	259.35	308.77	180.01	340.24	79.33
Value of maize harvest 2008/09	506.76	0.013	219.66	490.80	287.09*
Maize yield 2007/08 (Tons/hectare)	0.84	1.00	0.85	1.47	0.0055
Maize yield 2008/09 (Tons/hectare)	1.17	2.61	0.88	1.33	0.287
Total value of assets	550.74	3 008.51	159.65	581.58	391.00*
*Significant at 10% level, **significant at 5% level, ***significant at 1% level.

In the 2007/08 cropping season the difference in maize production between the intervention and the control households was not statistically significant, but in the 2008/09 season the value of the maize produced by intervention households was found to be US$287.09 higher than for control households. This difference cannot be attributed to the higher maize price for producers in the 2008/09 season, which was at US$295.89 per ton as compared to the price of US$125.99 per ton for the 2007/08 season, or to price differences between the intervention and the control households, as an analysis of the farm gate prices showed that all households in both the study area and the control area received the same farm gate prices. Nor can the differences in maize production in the 2008/09 season be attributed to yield differences, as the results indicate that there was no statistically significant difference between the yields of maize from households in the intervention and control areas.

The significant difference in maize production between the intervention villages and the control village can therefore mainly be attributed to intervention households planting more land than their counterparts in the control village. Households in the intervention villages had larger total land holdings and owned and planted more separate farm plots than households in the control village. Intervention households had on average 3.1 hectares of land while the control households had 2.2 hectares, and intervention households planted on average 1.72 separate farm plots while control households planted 1.23.These differences in amount of land owned and number of cultivated farm plots were statistically significant at the 1% and 5% confidence level, respectively.

Further analysis showed that the ERI intervention had been significant in increasing the value of households' total assets and livestock ownership by US$391.00 and US$300.12 respectively. Hence households in the intervention villages had higher valued assets than households in the control village. An analysis of the differences in livestock prices showed that there were small differences between the market prices of the three major types of livestock traded in the study area, with the average prices of chickens, pigs and goats for the 2008/09 season being US$5, US$54 and US$57 respectively for households in the intervention villages, while the corresponding prices for the control village were US$3, US$68 and US$62. Statistical analysis did not, however, reveal any statistically significant differences in the livestock prices between the two communities.

It can therefore be deduced that households in the intervention villages had larger numbers of livestock than households in the control village and this was confirmed by statistical analysis which showed that households in the intervention villages owned on average 3.7 more chickens and one more pig and goat each than households in the control village. These differences in ownership of all three classes of livestock were found to be highly statistically significant at the 1% confidence level. These larger livestock numbers explain the intervention households' higher values.

A major factor contributing to the larger livestock numbers in the intervention villages, especially for pig ownership, was that pig keeping was the agro-enterprise chosen as the intervention under the ERI. Participation in this intervention encouraged households to invest more in their pigs, in the form of improved housing, feeding and hygiene, and to improve their day-to-day management by keeping a record of all pig-keeping activities. Participating households were trained in constructing appropriate housing, formulating feed and controlling pests and disease. The farmer participatory research contributed by testing various feeding options and cultivating various types of feed (Njuki et al., 2007). Another improvement was better market access as a result of the contribution by the community marketing committee responsible for sourcing markets, which led to the establishment of a stable market, particularly for piglets. These changes in management and marketing improved household income. Participating households said in informal interviews that the increased income, along with changes in their decision-making processes, enabled them to invest more not only in their pigs but also in household assets and other livestock, especially poultry such as chickens.

5.3 Impact of ERI on household income

Further analysis showed that the ERI intervention positively influenced incomes in both the 2007/08 and 2008/09 cropping seasons for participating households. In rural areas of Malawi, including those in this study, household income is not synonymous with cash income but is a computed value which includes cash income earned from various temporary kinds of employment, the imputed value of non-cash income earned from the sale of own labour, cash income earned from selling agricultural crops and livestock, and the imputed value of all crops harvested, whether retained for home consumption or used as payment for farm workers. In this study 10 different sources of household income were identified and used to compute a household's total income.

Table 3 shows that households in the intervention villages had on average US$280.21 and US$340.54 more total income than their counterparts in the control village in the 2007/08 and 2008/09 cropping seasons respectively. These differences in household income were both statistically significant at the 10% confidence level. The increased cash incomes can be attributed to the ERI intervention's focus on helping farmers develop profitable agro-enterprises in order to meet existing market opportunities rather than simply selling any surplus they might have from their subsistence crops. The intervention communities analysed existing market opportunities at the start of the cropping year so they could determine which types of agro-enterprise would be most profitable. Their choices were pig keeping and dry bean cultivation (Njuki et al., 2007). Improved farmer-market linkages through the ERI process have also been found to increase rural households' incomes in other communities, not only in Malawi but also in Uganda (Kaaria et al., 2008).

Table 3: Impact of the Enabling Rural Innovation intervention on household incomes

	Intervention		Control
Household income (US$)	Mean	Standard deviation	Mean	Standard deviation	Effect
Total income for 2007/08 season	511.49	0.0072	231.28	465.61	280.21**
Total income for 2008/09 season	636.21	0.0088	299.56	655.75	340.54**
Total cash income from livestock sales 2008/09	51.34	138.48	23.60	186.08	27.78
*Significant at 10% level, **significant at 5% level, ***significant at 1% level.

5.4 Impact of ERI on fertiliser use patterns

The impact of the ERI intervention on fertiliser use patterns was assessed by analysing the differences in the number of 50 kg bags that farmers used per hectare of farm land. Inorganic fertilisers, in combination with hybrid seeds and good rainfall, play a crucial role in ensuring high maize production and food security in Malawi. Hence purchasing inorganic fertiliser demonstrates a household's decision-making patterns in terms of reinvestment in their farm enterprise. Table 4 shows differences between the amounts of inorganic fertiliser applied by intervention and control households in the 2004/05, 2005/06 and 2006/07 agricultural seasons that are significant at the 1%, 5% and 10% confidence levels respectively. Between the 2004/05 and 2006/07 cropping seasons, intervention households applied on average nearly one more 50 kg bag of inorganic fertiliser than control households. This difference can be attributed to the ERI intervention, since the better income from the markets acted as an incentive for households to reinvest in their farms in order to sustain their agro-enterprise. In general, the study found that all households in both the intervention and control areas were applying an amount of fertiliser that was below the recommended rates for Lilongwe Agricultural Development Division, where the communities are located. The recommended seasonal application for home consumption for Lilongwe is two bags of 23:21:0 + 4S and three bags of urea (with 46% nitrogen), while for production for the market it is one bag each of 23:21:0 + 4S and Urea (Benson, 1999). Table 4 shows that only the intervention households were close to reaching the recommended application for market production for the 2008/09 year: a mean of 1.92 bags of inorganic fertiliser per hectare.

Table 4: Impact of the Enabling Rural Innovation intervention on fertiliser use patterns

	Intervention		Control
Fertiliser use patterns (no. of 50 kg bags of inorganic fertiliser)	Mean	Standard deviation	Mean	Standard deviation	Effect
2004/05	1.24	0.62	0.567	0.44	0.679***
2005/06	1.38	0.63	0.624	0.11	0.761**
2006/07	1.50	0.63	0.858	0.32	0.644*
2007/08	1.68	0.68	1.38	1.22	0.297
2008/09	1.95	0.90	1.77	1.70	0.171
*Significant at 10% level, **significant at 5% level, ***significant at 1% level.

Further analysis showed that in more recent years the differences in fertiliser application between intervention and control areas were less pronounced and in the 2007/08 and 2008/09 cropping seasons they were insignificant. This can be attributed to the implementation of a fertiliser subsidy programme in Malawi, which increased the availability and accessibility of inorganic fertilisers throughout the rural areas of the country, hence increasing the opportunity for all farmers to access and use this fertiliser.

5.5 Impact of ERI on training and group membership

An assessment of the ERI intervention's impact on membership of farmer groups and the number of training sessions attended by a household was also carried out. Table 5 shows that five years ago, when the ERI was in implementation, intervention households attended on average 1.62 more training sessions than control households and this difference was statistically significant. These results indicate that the ERI intervention provided participating communities with significantly more training opportunities than are provided by the local agricultural extension officers.

Table 5: Impact of the Enabling Rural Innovation intervention on training and group membership

	Intervention		Control
Training and group membership	Mean	Standard deviation	Mean	Standard deviation	Effect
Total number of memberships of farmer groups per household	0.35	0.865	0.49	0.0743	–0.14
Average number of training sessions attended five years ago	2.64	5.67	1.02	3.86	1.62*
Average number of training sessions attended in 2007/08	1.92	5.29	1.26	3.66	0.66
Average number of training sessions attended in 2008/09	1.14	4.43	1.22	3.55	–0.08
*Significant at 10% level, **significant at 5% level, ***significant at 1% level.

Further analysis showed, however, that in the 2007/08 and 2008/09 cropping seasons, after the ERI intervention had been phased out in the 2006/07 season, there were no statistically significant differences between the number of training sessions attended by intervention and control households. It seems that the phasing out of the ERI intervention led the local agricultural extension officers to revert to pre-ERI training strategies in the intervention communities and these entailed less training. The results therefore indicate that during its implementation the ERI intervention had a positive impact in that it increased the number of training sessions that a household attended.

Another finding was that the ERI intervention did not have a statistically significant impact on households' membership of farmer groups, as the membership level was similar for the intervention and control communities. This finding is surprising, as the ERI intervention worked towards establishing and strengthening farmer organisations because it recognised that this was the most important success factor for increasing market access (Kaaria et al., 2008).

6. Conclusions and recommendations

This study concluded that agricultural research interventions that use an innovations systems approach have a strong positive impact on some but not all aspects of rural livelihoods, with stronger positive impacts being seen for incomes, upland crop production and fertiliser use, given the absence of government policies that provide subsidised fertilisers. In the presence of a subsidised fertiliser policy, innovative research interventions have a weaker positive impact on fertiliser use. In addition, weaker positive impacts are seen for maize production and training opportunities given similarities in the geographical location. Innovative agricultural research interventions can therefore have a positive effect on rural households' production, incomes and training opportunities.

The sustainability of intervention effects is, however, threatened by the phasing out of the interventions because local agricultural extension agents lack the human and financial capacity to maintain the higher level of contact and innovative strategies employed in implementing the intervention using agricultural innovation systems principles. Hence, to ensure that the communities continue to use these principles and the positive effects on rural livelihoods are sustained, agricultural research organisations need to invest more in building local public extension agents' capacity for understanding and applying agricultural innovation systems. Further, these systems should be mainstreamed in all public agricultural development interventions. This will, however, require deliberate and greater budgetary support by all public agricultural extension and research programmes. For this mainstreaming to be effective, it must be done concurrently with the capacity building efforts and budgetary support; without this, mainstreaming of innovation systems principles in public agricultural policies runs the risk of becoming rhetorical, with no real implementation.

Notes

¹In Malawi both men and women keep pigs. However, because of the high labour, managerial and financial requirements, women are often unable to keep large numbers of pigs. The ERI intervention therefore used ‘gender facilitation’ to fully understand women's preferences, which involved separating men and women into different groups during the initial phase so that the women could speak freely. At the meetings, women said they also wanted pig farming to be the agro-enterprise chosen for the ERI intervention but were constrained by the factors stated here.