We present the results of the expert consultations and discussions with the genebank scientist and breeders. Based on the genebank contribution according to any ancestral parent from the genebank, we found that four of the improved groundnut varieties could be traced to the genebank (Table 2). Two of these varieties (Kakoma and Manipintar) have no identifiable relationship with the genebank. Kakoma is the product of a national breeding program in India. No information about the origins of Manipintar or links to the genebank were reported after consultations. Moving to the actual genebank contribution based on the calculated contribution from each parent, we again apportioned zeros to Kakoma and Manipintar. For CG7 and Chitala, we apportioned full provenance values of 100%, since we identified that both varieties had parents that can be traced to the genebank. A similar apportioning value was also given to Baka, which happens to be an Indian landrace with roots to the ICRISAT genebank. Nsinjiro was given a 50% value, since only one of its parents could be traced to the genebank.
The summary statistics of the key model variables are presented in Table 3. Market participation was low, with just about 22% of households participating in markets to sell groundnuts. Coming to the explanatory variable of interest, we reported an RCP average of 38.26%. Similarly, about 44% of improved groundnut varieties had some direct ancestral links with genebanks. Households allotted about 0.38 hectares of land for groundnut cultivation on average. Most of the household heads were males (80%) with about 8 years of experience in the cultivation of groundnuts. Households had about 7 years of formal education and were generally located around extension agents.
We also performed a mean difference test to examine the differences between adopters and non-adopters (Table 4). Mean differences are observed for some of the socio-economic and contextual characteristics between these two groups of households. Adopters were generally more educated than non-adopters. Looking at our variables of interest, we observed significant differences in market participation between adopters and non-adopters. Adopters of improved groundnut varieties are more likely to participate in output markets as sellers and sell more quantities of groundnut than their non-adopting counterparts. While the results are intuitive and offer some insights into adoption, it may be inconclusive to rely on this without controlling for other confounders in a regression framework, as shown below.
Genebank contribution and market participation
The results of the empirical analyses are presented in this section, which shows the estimates of the relationship between genebank contribution and market participation. Table 5 presents a positive and significant relationship between genebank ancestry and market participation. Households that cultivated improved groundnut varieties whose ancestry could be traced to the ICRISAT genebank were more likely to participate in groundnut markets. Conditional on participating in groundnut markets, we also obtained a positive and significant effect on the quantity of groundnuts sold.
Moving to the RCP measure of genebank contribution, the findings remained unchanged for market participation, which maintained high statistical significance (Table 6). For the conditional outcome on the quantity of groundnuts sold, the magnitudes were positive, although not statistically significant. Given the low statistical significance in the binary case and no significance in this finer algorithm of genebank contribution, the analysis suggests that the cultivation of improved varieties with breeding materials from genebanks pushed smallholder households to markets, with no significant effect on quantity sold in these markets.
These results are plausible given the food insecurity status faced by many households in Malawi (Ragasa et al. 2019; Gelli et al. 2020). Based on this premise, and with insights from the non-separable household model, our findings should be correct. Faced with consumption demands, households will only participate in markets to the extent that their household food demands are met. They will only participate in markets as distress sales (to buy back later) or to sell the surplus of their production. Groundnuts constitute an essential part of the diet of most households in Malawi (Gelli et al. 2020). As highlighted by Carletto et al. (2017), the Living Standard Measurement surveys in Malawi, Tanzania, and Uganda, show that households participate in markets but sell very minimal quantities. This is even more the case in Malawi, also for food crops. Conditional on participation, an average of about 233.39 kg of groundnuts were sold by households.
The results of the (LPM) model for the two measures of genebank contribution were very similar to the original probit models, as shown in Table 7. This confirms the robustness of our estimates and further bolsters the positive association between genebank contribution and market participation of smallholder farmers in groundnut markets.
Once we established a positive association between genebank contribution and market participation, we sought to examine the mechanism through which improved varieties produced with materials from ICRISAT genebank leads to market participation. Previous literature (Bezu et al. 2014; Carletto et al. 2017; Verkaart et al. 2017) have highlighted the role of the extent of adoption in driving various welfare outcomes. We therefore verified if this explains the market-orientation of households. We refer to the extent of adoption as the area under improved groundnut. We performed two different sets of regressions to establish whether the extent of adoption is the underlying mechanism explaining the relationship between genebank ancestry and the market outcomes. First, we regressed the area under improved groundnut on our market outcomes. We hypothesized a positive association between adoption extent and market participation. Second, we took a step to further confirm whether our genebank proxies have any relationship with the extent of adoption.
For the first regression, we employed both panel and pseudo-panel, fixed estimators to control for time-invariant unobserved heterogeneity, given that our outcomes have different within-variation properties. We also controlled for time-variant unobserved characteristics using a control function approach with the specification of three instruments. We estimated a Tobit model of adoption extent, including soil characteristics, access to irrigation, and distance to extension agents, which served as a source of exogenous variation. From this, we calculated the generalized residual, which we used in the main outcome equations. Including this residual in the outcome equation served as both a test of endogeneity and a way of controlling it (Wooldridge 2015).
The results of Table 8 show the relationship between adoption extent and market participation. As hypothesized, adoption extent increases the participation of smallholder farmers in groundnut markets. In both outcomes, the results are highly statistically significant, suggesting that the extent of adoption may well be the mechanism explaining the positive association of genebank contribution on market participation.Footnote 2 Looking at the relationship between our two measures of genebank contribution and the extent of adoption, we further reported a positive association, which was again highly significant at the 1% level of probability (Table 9). Our results here are in line with earlier findings that commercialization increases with the area of cultivation (Carletto et al. 2017).