Adoption of fertiliser and hybrid seeds by smallholder maize farmers in Southern Malawi

Abstract

Despite decades of agricultural policies that promoted the adoption of fertiliser and hybrid seed technologies as ways of improving productivity in maize farming, smallholder farmers in Malawi have been relatively slow to adopt the new technology. Using bivariate probit analysis and controlling for technology acquisition through grants, we found that fertiliser adoption was positively associated with higher levels of education, larger plot sizes and higher non-farm incomes, but negatively associated with households headed by women and distance from input markets. The adoption of hybrid seeds is positively associated with market-based land tenure systems and fertile soils, but negatively associated with age of the farmer and distance from input markets.

1. Introduction

Agriculture in Malawi remains the most important sector for achieving economic growth which will benefit the poor over the medium term, and increasing agricultural incomes will be the key source of poverty reduction, as more than 90 per cent of the population derive their livelihoods from agriculture (Government of Malawi, 2002). One of the strategies for improving agricultural incomes set out in the Malawi Poverty Reduction Strategy Paper (Government of Malawi, 2002) is to expand and strengthen access to agricultural inputs such as fertilisers, manure and improved seeds, against a background of low rates of fertiliser and hybrid seed adoption among smallholder farmers. In the past several decades the government has invested substantially in the promotion of modern technologies in the agricultural sector, including the provision of subsidies for agricultural inputs. There was considerable public sector investment in a series of integrated rural development projects which introduced a range of services, including extension services and rural credit facilities.

More particularly, there is evidence that technological developments such as the introduction of new seed varieties, adoption of fertilisers and integrated farming systems have been central to these efforts in Malawi (Government of Malawi, 1971, 1987; Lele, 1989; Smale, 1995; Smale et al., 1995). The marketing and pricing of fertilisers in the smallholder sector was placed under the control of government through the Agricultural Development and Marketing Corporation (ADMARC) and the Smallholder Farmer Fertiliser Revolving Fund (SFFRF). ADMARC monopolised the distribution of fertilisers to the smallholder sector at subsidised prices in order to encourage the use of fertilisers, so as to achieve its overall objective of raising agricultural output particularly for maize, the dominant smallholder crop (Sahn & Arulpragasam, 1991). However, within the programme of economic reforms implemented more recently by the Malawi government the subsidy on fertilisers and agricultural inputs has been removed and agricultural input markets have been liberalised.

Maize is the main staple crop in Malawi, and is mainly grown by smallholder farmers on land holdings of less than two hectares. Alwang & Siegel (1999) note that about 70 per cent of Malawian smallholder farmers cultivate less than one hectare and that the median area under cultivation is about 0,6 hectares. This category of smallholder farmers devotes about 70 per cent of the land to maize. As the population of Malawi has grown, land holdings have been declining over time through family subdivisions. It is apparent that the success of the agricultural sector in Malawi hinges on the use of high-yielding technologies and farming systems.

Our understanding of the factors that affect the adoption of fertiliser and hybrid seed technology will be likely to help redress the policy failures experienced to date with technology promotion in Malawi. Although there have been studies in Malawi on the factors affecting technology adoption (Green & Ng'ong'ola, 1993; Zeller et al., 1998), the focus was always on fertiliser technology adoption by itself, and ignored the joint decision of adopting both fertiliser and hybrid seed technologies together. Hence, previous studies assumed that there was no interdependence between the decisions to adopt fertiliser and hybrid seed technologies. This study contributes to the literature on the adoption of agricultural technology in Malawi, by exploring the factors that are important for the adoption of fertiliser and hybrid seed technologies among smallholder farmers.

The rest of the paper is organised as follows. Section 2 reviews the existing theoretical and empirical literature on factors influencing agricultural technology adoption. Section 3 outlines the methodology and data used in the study. Section 4 reports and discusses empirical results from the bivariate probit regression. Section 5 provides concluding remarks.

2. Review of the literature

The literature suggests several theoretical or conceptual models on farmers' decisions to adopt new technology (Feder & Slade, 1984; Abadi Ghadim & Pannell, 1999; Negatu & Parikh, 1999; Isham, 2002). Feder & Slade (1984) develop a model of technology diffusion based on human capital and land constraints. Their model postulates that farmers with more education and larger land will have more knowledge of improved farming systems and are likely to adopt technology more rapidly. Isham (2002) extends the model of Feder & Slade (1984) by incorporating social capital as a fixed input into the decision to adopt technologies. This extended model predicts that farmers with neighbours who adopt the technology, and those with higher levels of social capital, accumulate more information and adopt technology more rapidly. Abadi Ghadim & Pannell (1999) place emphasis on the role of learning by doing and the impact of the learning on personal perceptions of the innovation.

Negatu & Parikh (1999) review three groups of models on the adoption of agricultural technologies by smallholder farmers:

•	the innovation–diffusion or transfer of technology model, in which technology is transferred from its source to the smallholder farmer through an intermediary such as an extension system, and the diffusion of the technology depends on the characteristics of the farmer
•	the economic constraint model takes the view that farmers have different factor endowments and that the distribution of endowments determines the adoption of technology
•	the technology characteristics–user's context model assumes that the characteristics of the technology and the underlying agro-ecological, socio-economic and institutional circumstances of farmers play a central role in the adoption of technology.

These technology adoption models have been tested empirically using data from developing countries, investigating particularly the factors that affect the adoption of fertiliser and improved seed varieties (Green & Ng'ong'ola, 1993; Smale et al., 1995; Croppenstedt & Demeke, 1996; Zeller et al., 1998; Negatu & Parikh, 1999; Weir & Knight, 2000; Kosarek et al., 2001; Doss & Morris, 2001; Isham, 2002).2 The factors influencing technology adoption decisions include farm size, risk exposure and capacity to bear risks, human capital, labour availability, land tenure, access to financial and produce markets, access to information, participation in off-farm activities, social capital, household characteristics and ecological and environmental factors. The Government of Malawi (2002) attributes the low rate of technology adoption in the smallholder agricultural sector to the problem of incomplete financial markets, an argument supported empirically in studies by Green & Ng'ong'ola (1993) and Zeller et al. (1998). Green & Ng'ong'ola (1993) found that tobacco farming, improved varieties, access to credit, participation in off-farm activities and regular employment were the main factors influencing fertiliser adoption in Malawi. Zeller et al. (1998) found that the adoption of hybrid maize and tobacco was affected by factor endowments, exposure to agro-ecological risks, and access to financial and commodity markets.

3. Methodology and data

Since independence in 1964, the government has promoted two types of technologies in Malawi, particularly for maize farmers. First, government has been investing in maize seed research, and has since released several high-yielding varieties. Smale (1995) notes that most of the maize hybrid varieties that were developed in the national research programme were dent varieties, based on the belief that dents had higher yield potential than flint varieties.3 Secondly, the government has been promoting the use of inorganic fertilisers in improving the fertility of soils, as a way of increasing maize productivity. The prices of fertilisers were heavily subsidised, and the marketing and distribution of fertilisers were controlled by the Ministry of Agriculture and ADMARC (Sahn & Arulpragasam, 1991; Zeller et al., 1998).

The adoption of hybrid seeds and inorganic fertiliser technologies were promoted concurrently. The smallholder farmer in Malawi had to decide whether to adopt hybrid seeds or inorganic fertilisers, or both. Although yields per hectare for hybrid maize without fertiliser application were higher than yields of local maize in both normal and drought situations, hybrid maize was more responsive to fertilisers (Smale, 1995). However, adoption rates as between hybrid seeds and chemical fertiliser technologies are bound to differ among smallholder farmers in Malawi. Doss & Morris (2001) note that hybrid seeds are relatively simple technologies, considering that they require relatively few changes to farmers' usual practices, while chemical fertiliser technologies are complex and require farmers to know different varieties and optimal application rates.

Although alternative econometric techniques have been used in the empirical literature to analyse the factors associated with the decision by smallholder farmers to adopt new technologies, the most common method has been modelling the decision as a latent variable, using either probit or logit regression analysis. Others have used the decision to use new technology as a selection variable and then estimate the demand for fertiliser (Croppenstedt & Demeke, 1996). In most of the studies on the adoption of new technology, only a single technology decision is considered, and the possibility of joint or sequential decisions over complementary technologies is ignored. Doss & Morris (2001) recognise the possibility that a link exists between the decisions to adopt chemical fertiliser and improved maize varieties, and use a two-stage approach in which predicted fertiliser adoption and predicted maize variety adoption are included as explanatory variables in the maize variety and fertiliser adoption models respectively. Following the linkage recognised by Doss & Morris (2001), we estimate the models using bivariate probit or seemingly unrelated bivariate probit regressions represented in the following system:

where fertiliser and hybrid are dummy variables representing whether farmer k adopted fertiliser and hybrid seeds on plot i, respectively; X _i is a vector of plot-level and smallholder farmer characteristics; Y _j is a vector of household, institutional and infrastructure characteristics; ϵ₁ and ϵ₂ are error terms whose covariance is non-zero.

Several characteristics of the farmer and plot are included as explanatory variables in the models. The farmer characteristics include gender, age and education of the smallholder farmer cultivating the plot. The gender differences in technology adoption are captured by a dummy variable, female, that takes a value of 1 if the farmer is female. Age and education are expressed respectively as years of age and number of years of formal schooling completed by the smallholder farmer. We also include plot-level characteristics including plot size, land tenure, perceived soil fertility, perceived steepness of the land and use of hired labour in crop production on the plot. The plot size variable, plotsize, measured in hectares, captures the effects of land constraints on technology adoption. We expect technology adoption to be positively associated with land size. Doss & Morris (2001) note that although fertiliser and maize variety technologies are scale-neutral, those with more land are able to afford fertilisers because they produce a marketable surplus. Some of the farmers control more than one plot of maize and a dummy variable, mplots, takes a value of 1 if the plot belongs to a farmer who has more than one plot. This variable also captures the effect of land fragmentation on technology adoption. Land tenure systems in Malawi are non-market oriented and based on traditional marriage systems (Place & Otsuka, 2001). A dummy variable, landmkt, takes the value of 1 if the land tenure on the plot is market-based such as lease and freehold and captures the incentive effects in investing in high productivity technologies. The use of hired labour on the plot, hiredlabor, captures the effect of augmenting family labour supply to implement adopted technologies successfully. Dummy variables that control for ecological differences are included in the model captured by perceptions about soil fertility (good soil fertility – soilgod and average soil fertility – soilavg) and perceptions about the terrain, terrain, which is equal to 1 if the terrain of the plot was perceived by the farmer as being generally flat.

Two household characteristics, the headship of a household to which the smallholder farmer belongs, and non-farm incomes, are included in the model. The poverty profile in Malawi reveals that female-headed households tend to be poorer and more constrained by resource availability (Government of Malawi, 2002), which may affect the adoption of technology, especially fertiliser technologies that are more expensive than maize variety technologies.4 The gender of the household headship is captured by a dummy variable, hhfemale, which takes a value of 1 for female-headed households. It is also important to control for the role of non-farm incomes in technology adoption decisions. Non-farm incomes provide farmers with additional resources which may be used to purchase new technologies. Green & Ng'ong'ola (1993), in an earlier study on fertiliser adoption in Malawi, found that the probability to adopt was an increasing function of non-farm incomes and regular labour. The variations in adoption of technologies due to differences in household non-farm incomes is captured by the variable, nonfarm, measured in thousands of Malawi kwacha.

We also include three institutional variables and one infrastructure variable in the model, to account for differential access to basic social services and agricultural extension services. The variable, club, takes the value of 1 if the any member of the household belongs to a farmers' club or association. The club or association represents some form of social capital and can serve as a forum for disseminating important agricultural messages. Prior to the collapse of the smallholder agricultural credit scheme in 1992, the farmers' clubs were also used as a vehicle for farmers' access to agricultural credit for inputs. These clubs are less common in the agricultural sector today, but it is important to investigate the importance of associations in technology decisions. Isham (2002) argues that farmers in villages with higher levels of social capital are likely to have higher levels of cumulative information and adopt new technology more rapidly. Access to agricultural extension services may also be important in influencing the decision to adopt technology. We include the number of extension visits, extension, the farmer had in the agricultural season to capture the effect of expert advice. Since the 1990s, the Malawi Government and non-governmental organisations have been implementing various safety net programmes in the agricultural sector through the distribution of free inputs (fertiliser and hybrid seeds) to food-insecure households (Sibale et al., 2001; Dzimadzi et al., 2001). In a model of technology adoption, it is important to control for use of technology acquired through gifts or grants. We therefore include a dummy variable, grants, taking the value of 1 for farmers who received agricultural grants and used these grants on their plots. The distance to input markets, distance, is captured by the distance from the plot to ADMARC markets. Although ADMARC has recently been experiencing financial difficulties that led to problems in supplying inputs to smallholder farmers, it remains the most accessible source of inputs (Mvula et al., 2003).

The data used in this study were collected through a questionnaire administered to 156 households in the Machinga district in southern Malawi. We selected randomly two Traditional Authorities in the selected district and two enumeration areas in each Traditional Authority.5 In each selected enumeration area at least 37 households were selected randomly, based on a simple household listing. In each of the selected households, we interviewed the household head or a person with information about the farming activities of other household members, and also individual members where necessary. The 156 households interviewed had a total of 444 plots used for the production of various crops. Of these 202 plots farmed by 139 farmers generated usable data to estimate the technology adoption model.

4. Empirical results

Table 1 presents the definition of variables used in the econometric model and their descriptive statistics.6 Despite a long history of promotion of fertiliser and hybrid seed technologies in Malawi, only 54,5 per cent and 40,6 per cent of the plots used the technologies, respectively. The gender distribution of control of plots (persons making most farming decisions) shows that women control 42,8 per cent of the plots in the sample. The level of education among maize farmers is low; the mean number of years of schooling was 3. Most of the plots on which maize is grown are small, with a mean plot size of 0,57 hectares. This implies that most of the farmers interviewed were net food buyers. Thus, the maize on these plots is grown purely for subsistence. Only on 5 of the 202 plots was maize sales reported during the 2001/2002 season. The preceding season, 2000/2001, was a famine year and many households ate most of the maize while green, as they did not have sufficient food to eat before the harvest. About 45,1 per cent of the plots are under the control of farmers with multiple plots.

Table 1: Definition of variables and descriptive statistics

Variable	Description	Mean	SD
fertiliser	Dummy: 1 if the farmer used fertiliser on the plot	0.5446	0.4992
hybrid	Dummy: 1 if the main type of maize on the plot is hybrid	0.4059	0.4923
female	Dummy: 1 if the farmer in control of the plot is female	42.802	15.201
age	Age of the farmer in number of years	0.4356	0.4971
education	Number of years of schooling for the farmer	3.4406	3.4883
plotsize	Size of the plot under maize cultivation (hectares)	0.5641	0.5041
mplots	Dummy: 1 if the farmer controls more than one maize plot	0.4505	0.4988
landmkt	Dummy: 1 if the land tenure system is market based (lease/freehold/rented)	0.0545	0.2275
hhfemale	Dummy: 1 if the farmer belongs to a female-headed household	0.2723	0.4462
nonfarm	Household non-farm income in thousands of Kwacha per year	7.2709	8.8289
soilgod	Dummy: 1 if the fertility of the soil is good	0.2178	0.4138
soilavg	Dummy: 1 if the fertility of the soil is average	0.3614	0.4816
terrain	Dummy: 1 if the terrain of the plot is flat	6.7525	4.5712
distance	Distance of the plot from ADMARC market (kilometres)	0.6535	0.4770
hiredlabor	Dummy: 1 if hired labour was used in crop production	0.1980	0.3995
club	Dummy: 1 if any of the members of the household belongs to a club	0.1089	0.3123
extension	Number of extension visit the farmer in the household had in the agricultural season	1.4951	2.3862
grants	Dummy: 1 if the farmer received agricultural input grants and used them on the plot	0.2228	0.4171

The average distance from the plot to ADMARC markets is 6,8 km. Only 5,5 per cent of the plots have a market-based land tenure system. Family members supply most of the labour used in the cultivation of maize; hired labour was used in nearly 20 per cent of the plots. Only 10,9 per cent of farmers come from households in which at least one member has membership in a club or association. The average number of extension visits to any household member per year is 1,5, with substantial variations in the sample. In recent years, government agencies and non-governmental organisations have been implementing safety nets for resource poor farmers through the provision of agricultural input grants (for hybrid seeds and fertilisers). Our sample shows that 22,3 per cent of the plots utilised agricultural input grants.

Table 2 reports results of the standard bivariate probit estimation and marginal effects for the seemingly joint decision of fertiliser and hybrid seeds technologies adoption by smallholder farmers. The Wald (χ²) test for overall performance of the model shows that we cannot accept the null hypothesis that all the coefficients are equal to zero. Similarly, the null hypothesis of zero covariance between the error terms in the fertiliser and hybrid seeds adoption equations is rejected by the Wald (χ²) test at the 1 per cent significance level.

Table 2: Bivariate probit model of adoption of technologies

	Fertiliser adoption		Hybrid seeds adoption
Variables	Coefficient	t‐ratio	Coefficient	t-ratio	Marginal effects
age	0.0028	0.42	−0.0143^b	−2.00	−0.0031
female	−0.1461	−0.58	0.0963	0.37	0.0034
education	0.0930^a	2.7	0.0432	1.30	0.0231
plotsize	1.0489^a	3.49	−0.1853	−0.79	0.0958
mplots	−0.3565^c	−1.75	−0.1846	−0.93	−0.0926
landmkt	0.0311	0.07	1.1609^a	2.70	0.2334
hhfemale	−0.7341^a	−2.57	−0.2330	−0.85	−0.1525
nonfarm	0.0186^c	1.65	0.0064	0.54	0.0041
soilgod	−0.3114	−1.16	0.5793^b	2.22	0.0727
soilavg	−0.3065	−1.32	0.4581^b	2.03	0.0604
terrain	0.1594	0.73	0.2854	1.41	0.0894
distance	−0.0574^b	−2.21	−0.0530^b	−2.21	−0.0207
hirelabor	0.4486^c	1.73	−0.1130	−0.44	0.0213
extension	0.0071	0.15	0.0667	1.40	0.0173
club	−0.0769	−0.30	−0.6065^c	−1.86	−0.1493
grants	1.3723^a	4.89	0.6987^a	2.84	0.3550
constant	−0.5101	−1.05	0.0070	0.01	–
Model statistics
Log likelihood	−208.26
N	202
ρ	0.521
Wald χ(16) ^2	111.22	[0.00]
Wald test: ρ = 0	13.169	[0.00]
The standard errors of the model are heteroscedastic-consistent. Figures in square brackets are the probabilities. The superscripts a, b and c represents statistically significant at the 1 per cent, 5 per cent and 10 per cent levels, respectively.

Of the farmer and plot level characteristics included in the model, age of the farmer, education of the farmer, size of the plot, land tenure and soil fertility are statistically significant at the conventional levels. The gender of the farmer, terrain of the plot and hired labour are not significant determinants of technology adoption both with respect to inorganic fertilisers and improved maize varieties. The age of the farmer is statistically significant at the 10 per cent level only in the adoption of hybrid seeds. The negative relationship shows that older farmers are less likely to adopt hybrid seeds. Local maize, which has a flint grain texture, has always been favoured by smallholder subsistence farmers due to the high proportion of hard starch granules, the ease in storage management and high flour to grain extraction to make the favoured refined flour for cooking nsima, the traditional food (Smale, 1995). Older people, who are used to the traditional flour, are more resistant to change to hybrid seeds which have either dent or semi-flinty grain textures.

Although the coefficient of education has expected signs in both technology adoption functions, the level of schooling is only statistically significant at the 1 per cent level in farmer's decision to adopt inorganic fertilisers. The results with respect to inorganic fertiliser adoption support the hypothesis that farmers that are more endowed in human capital are receptive to new ideas and are therefore more likely to adopt productivity-enhancing technologies, but contrary to earlier studies in Malawi agriculture.7

The size of the plot is only a significant factor in influencing the decision to adopt fertiliser technology and the coefficient is statistically significant at the 1 per cent level. Thus, it may not be economically efficient for smallholder farmers with small plot holdings to apply fertilisers due partly to the packaging of fertilisers. Control of multiple maize plots by the farmer is negatively related to adoption of technology, but the coefficient is statistically significant at the 10 per cent level only in the fertiliser adoption model. This implies that land fragmentation reduces the probability of fertiliser adoption. The market-based land tenure system is positively associated with technology adoption decisions, but the coefficient of landmkt is only statistically significant at the 1 per cent level with respect to the adoption of hybrid maize seeds. Although Place & Otsuka (2001) find that some traditional land tenure systems provide security, are therefore not a binding constraint to in land improving investments in Malawi, the evidence in this study is that farmers are more likely to use hybrid seeds on rented or leased or freehold land. This behaviour may be motivated by the relative profitability of hybrid maize over local maize and the fact that maize on market-based land tenure systems is cultivated to generate commercial returns rather than meet household subsistence needs. The results also suggest that perceived fertility of soils is important in the farmers' decision to adopt hybrid maize seeds and the probability of adoption increases with the quality of soils, lower for average soils (soilavg) and higher for good soils (soilgod). However, soil fertility is not a significant factor in the adoption of inorganic fertiliser technology.

The two household characteristics in the model are significant determinants of adoption of technology only in the inorganic fertiliser decision. We find evidence that farmers that belong to female-headed households are unlikely to adopt farming technologies; however, the coefficient of hhfemale is only statistically significant at the 1 per cent level with respect to the adoption of fertiliser technology. Green & Ng'ong'ola (1993) observe that fertiliser adoption among female-headed households may be low due to their limited contact with extension services. However, as we note below, contact with extension services is not important and this may suggest that the household resource envelope and distribution may be central to understanding adoption rates among farmers from female-headed households. Although the coefficient of non-farm income, nonfarm, has the expected signs in both adoption decision functions, it is only statistically significant at the 10 per cent level with respect to fertiliser technology adoption. Because the fertiliser technology is more expensive than hybrid seeds technology, income augmentation from non-farm sources increases affordability for such technologies.

Among the institutional and infrastructure variables in the model, distance, club membership and grants are important factors in explaining the probability of technology adoption. Contacts with extension services (extension) and membership of clubs or association (social capital) have inconsistent signs, with the former being statistically insignificant in both cases. The insignificance of extension visits reflects the inefficiency of the agricultural extension system in recent times. According to the Government of Malawi (2002), the agricultural extension system has, in recent years, come under pressure from financial and human resource constraints. Club membership, club, is negatively associated with hybrid maize seeds adoption and the coefficient is statistically significant at the 10 per cent level. The distance to ADMARC markets from the plot, distance, is negatively associated with farmers' decisions to adopt fertiliser or hybrid maize technologies. In addition, most of those farmers that received grants used the technologies in the ‘grant packs’ – fertilisers and seeds.

The last column of Table 2 reports the marginal effects of joint probability of adopting inorganic fertiliser and hybrid maize seeds technologies, indicating the change in the probability resulting from a unit change in continuous explanatory variables and change from zero to one for dichotomous explanatory variables. The predicted probability (computed at the means) that a smallholder maize farmer will jointly adopt fertiliser and hybrid seeds technology is 30 per cent. The probability of adopting both inorganic fertiliser and hybrid maize seeds technologies falls by 16,1 percentage points for female-headed households, a unit increase in years of education increases the probability of adopting both technologies by 2,5 percentage points while a unit increase in the distance to ADMARC markets reduces the probability of adoption by 21 percentage points.

5. Conclusions and policy implications

After years of government investment in the agricultural sector, particularly investments in research on modern maize seed varieties and promotion of the use of fertilisers among smallholder farmers as productivity-enhancing technologies, the adoption rate remains low and Malawi has not yet achieved self-sufficiency in food production. This study sought to understand the factors that determine the adoption of agricultural technologies by smallholder farmers cultivating maize, the main staple crop in Malawi. Using the bivariate probit model, after controlling for technologies provided as grants, we find the probability of adopting both inorganic fertiliser and hybrid maize seeds technologies to be 30 per cent. The probability of fertiliser adoption increased with the level of education, size of the cultivated plot and level of non-farm incomes, but is a decreasing function of female headship of the households and distance of the plot from ADMARC markets. The probability of hybrid maize seeds adoption is an increasing function of market-based land tenure systems and soil fertility, but a decreasing function of age of the farmer, distance of the plot from ADMARC markets and membership of a club or association.

We can derive several policy implications from the results of this study. First, similar to other studies of technology adoption among smallholder farmers in developing countries, education plays a central role in the adoption of more complex technologies such as use of fertilisers compared to relatively simple technologies such as maize seeds, in which historical inertia and not education is critical in the adoption decision. Thus, improving the level of education of smallholder farmers in the medium and long term is likely to lead to productivity gains in food production in Malawi. Secondly, addressing the land holding size among smallholder farmers has the potential to increase fertiliser adoption rates and hence increase productivity in maize production. The implementation of the land reallocation programme articulated in the Malawi Poverty Reduction Strategy Paper (Government of Malawi, 2002) is likely to lead to increased productivity through the application of fertilisers.

Thirdly, the significance of female-headed households and non-farm incomes reflects the importance of household resource endowments. Improving fertiliser adoption among smallholder maize farmers will require addressing the resource constraints of female-headed households and promotion of livelihoods diversification. Fourthly, the negative relationship between adoption of technology and distance to ADMARC markets casts doubts on the wisdom of closing some of the markets insofar as access to inputs are concerned. Due to the operational inefficiency of ADMARC, government is under pressure from the World Bank, International Monetary Fund and the donor community to privatise ADMARC – a policy that would consequently lead to closure of non-commercial markets. As Mvula et al. (2003) note, the private sector has not responded favourably to the liberalisation of agricultural input markets – only a few large-scale enterprises sell agricultural inputs and these are usually located in urban or peri-urban centres far away from many smallholder farmers.

Additional information

Notes on contributors

Ephraim W Chirwa

Department of Economics, Chancellor College, University of Malawi. The data used in this paper form part of a research project on ‘Sources of Technical Efficiency among Smallholder Maize Farmers in Malawi’, for which I acknowledge the financial support provided by the African Economic Research Consortium (AERC) through grant no. RT01555. I also thank anonymous referees for their comments and suggestions. The usual disclaimer applies.

Notes

²See Feder et al. (1985) for a review of earlier studies.

³Flinty maize types have higher proportion of hard starch granules in the kernel than dents and they have the higher flour-to-grain extraction rate because the germ separates more easily from the bran when pounded in a mortar. Flinty varieties also have tip cover and harder grains that protect them from the weevils (Smale, 1995).

⁴Similarly, Croppenstedt & Demeke (1996) argue that female-headed households in Ethiopia are less likely to use and apply fertiliser because they tend to be poorer and more subsistence and labour-constrained.

⁵An enumeration area is the smallest stratification that is used by the National Statistical Office in national surveys and has an average of 250 households.

⁶The descriptive statistics are presented at plot level, which implicitly assumes that farmers treat each plot equally. However, plot level analysis in the econometric modelling accounts for the likelihood that farmers with multiple plots may treat them differently, aspects that are masked when the analysis is done at farmer or household levels like in most studies.

⁷Similar evidence of a positive role of education is reported from studies in Ethiopia (Weir & Knight, 2000; Croppenstedt & Demeke, 1996) and Ghana (Doss & Morris, 2001). However, others such as Isham (2002) for rural Tanzania, Green & Ng'ong'ola (1993) and Zeller et al. (1998) in Malawi do not find significance evidence on the role of education in technology adoption.