Scholarly article on topic 'Does storage technology affect adoption of improved maize varieties in Africa? Insights from Malawi’s input subsidy program'

Does storage technology affect adoption of improved maize varieties in Africa? Insights from Malawi’s input subsidy program Academic research paper on "Agriculture, forestry, and fisheries"

CC BY-NC-ND
0
0
Share paper
Academic journal
Food Policy
OECD Field of science
Keywords
{"Food security" / Storage / "Improved maize seed adoption" / "Input subsidies" / Malawi / "Sub-Saharan Africa"}

Abstract of research paper on Agriculture, forestry, and fisheries, author of scientific article — Jacob Ricker-Gilbert, Michael Jones

Abstract To date there is limited knowledge of how having access to post-harvest storage technology affects a smallholder African farmer’s decision to adopt higher-yielding improved maize varieties. This is a key issue because higher yielding varieties are known to be more susceptible to storage pests than lower-yielding traditional varieties. We address this question using panel data from Malawi, and incorporating panel estimation techniques to deal with unobserved heterogeneity. Our results indicate that acquiring chemical storage protectants after the previous harvest is associated with a statistically significant and modest positive impact on the probability of adopting improved maize, total area planted to improved maize varieties, and share of area planted to improved maize varieties in the next planting season. We also find that the storage chemical subsidy is associated with significant crowding out of commercial storage chemical purchases, as farmers who acquire subsidized chemicals are more than 50 percentage points less likely to purchase commercial chemicals on average. These findings have implications for maize adoption and input subsidy policies, and they indicate that researchers, extension staff, and policy makers should consider post-harvest issue when promoting adoption of improved varieties.

Academic research paper on topic "Does storage technology affect adoption of improved maize varieties in Africa? Insights from Malawi’s input subsidy program"

Contents lists available at ScienceDirect

Food Policy

journal homepage: www.elsevier.com/locate/foodpol

Does storage technology affect adoption of improved maize varieties in Africa? Insights from Malawi's input subsidy program

Jacob Ricker-Gilbert *, Michael Jones

Dept. of Agricultural Economics, Purdue University, 403 W. State Street, West Lafayette, IN 47907-2056, United States

ARTICLE INFO ABSTRACT

To date there is limited knowledge of how having access to post-harvest storage technology affects a smallholder African farmer's decision to adopt higher-yielding improved maize varieties. This is a key issue because higher yielding varieties are known to be more susceptible to storage pests than lower-yielding traditional varieties. We address this question using panel data from Malawi, and incorporating panel estimation techniques to deal with unobserved heterogeneity. Our results indicate that acquiring chemical storage protectants after the previous harvest is associated with a statistically significant and modest positive impact on the probability of adopting improved maize, total area planted to improved maize varieties, and share of area planted to improved maize varieties in the next planting season. We also find that the storage chemical subsidy is associated with significant crowding out of commercial storage chemical purchases, as farmers who acquire subsidized chemicals are more than 50 percentage points less likely to purchase commercial chemicals on average. These findings have implications for maize adoption and input subsidy policies, and they indicate that researchers, extension staff, and policy makers should consider post-harvest issue when promoting adoption of improved varieties.

© 2014 Elsevier Ltd. All rights reserved.

CrossMark

Article history:

Received 2 December 2013

Received in revised form 29 September 2014

Accepted 29 October 2014

Keywords: Food security Storage

Improved maize seed adoption

Input subsidies

Malawi

Sub-Saharan Africa

Introduction

Increasing adoption of modern inputs such as improved seeds and chemical fertilizer is essential for boosting staple crop production and increasing smallholder food security in sub-Saharan Africa (SSA). Numerous studies in SSA find that adoption of improved maize varieties contributes to raising productivity which increases household income and food security (Smale, 1995; Katengeza et al., 2012; Mason and Smale, 2013; Bezu et al., 2014). However in addition to increasing productivity, it is essential to recognize that food security does not simply end at harvest because susceptibility to pests during storage can cause tremendous post-harvest dry weight (quantity) losses of up to 30% in six months of storage for grains (Boxall, 2002). In addition, previous work confirms common rural knowledge that higher yielding but softer dent hybrids, the most commonly promoted improved maize varieties in SSA, offer less natural protection against storage insects such as maize weevil and larger grain borer due to their softer husks, than do lower yielding but harder traditional flint varieties (Smale et al., 1995;

* Corresponding author at: 403 W. State Street, Room 623, W. Lafayette, IN 47907, USA. Tel.: +1 (765)494 4260; fax: +1 (765)494 9176. E-mail address: jrickerg@purdue.edu (J. Ricker-Gilbert).

Adda et al., 2002).1 Therefore farmers face a rational trade-off at planting time between choosing an improved variety that may boost production but where the harvested maize is more susceptible to pests when stored vs. choosing a traditional variety that is lower yielding but less vulnerable to pests in storage. Nevertheless, issues related to post harvest loss are often overlooked in studies that model smallholder improved seed adoption behavior.

With these considerations in mind, the first objective of this article is to determine how use of storage technology in the form of chemical protectants affects a smallholder's decision to adopt improved varieties of maize seed in Malawi.2,3 In doing so this study makes an empirical contribution to both the technology adoption lit-

1 Smallholder perceptions of greater storage pest damage in improved vs. local maize varieties has also been recently verified in Malawi (Lunduka et al., 2012; Jones, 2012).

2 In this study improved maize seeds are defined as hybrid varieties and open pollinated varieties (OPV). Although smallholder farm households in Malawi report that more than 95% of the improved maize seed they acquire is hybrid, anecdotal evidence from Malawi indicates that most farmers refer to any improved seed as hybrid.

3 Storage chemicals are currently the most widely used form of "modern" protection against post-harvest pests in Malawi, and Table 1 indicates that more than half of all farmers use them in either liquid or powder form. Chemicals are commonly applied even when bags of maize are stored in the kitchen, sleeping, or living area of the home as external storage facilities may not exist.

http://dx.doi.Org/10.1016/j.foodpol.2014.10.015 0306-9192/® 2014 Elsevier Ltd. All rights reserved.

erature and the input subsidy literature in SSA. Malawi has received wide-spread recognition for scaling up a large inorganic fertilizer subsidy program in 2005 and a subsidy for improved maize seeds in 2006 (Dugger, 2007). With the expansion of the seed subsidy program, by the 2008-2009 agricultural year almost 40% of smallholder households had received subsidized improved seed (Mason and Ricker-Gilbert, 2013).4 However less attention has been paid to the fact that Malawi implemented a subsidy for maize storage chemicals beginning after the 2009 harvest and running through 2012 harvest as a compliment to the fertilizer and seed subsidy. The storage chemical component was added to the subsidy program based on a recognition that post-harvest pests may undermine increases in maize production that are achieved by farmers who adopt improved varieties through the subsidy program.

Therefore, the second objective of this study is to test whether or not, and to what extent the storage chemical subsidy may crowd out or crowd in the commercial market for storage chemicals. This is an important issue because for the storage chemical component of the subsidy program to be successful it must increase the amount of storage chemicals that households use. If acquiring subsidized storage chemicals makes people more likely to buy commercial storage chemicals then the subsidy program crowds in commercial storage chemical use, and adds to the total quantity of storage chemicals applied to farmers' maize. Conversely, if those who acquire subsidized storage chemicals use some or all of it in place of commercial purchases, then the effect of the subsidy on total chemical use will be reduced, causing crowding out of commercial chemicals, and undermining the effectiveness of the program.

The first wave of data from our study provide evidence on storage chemical use after the 2007/08 growing season, the year before the storage protectant subsidy was scaled up, but when the fertilizer and seed subsidy was in full swing. In the first wave all purchases of storage chemicals are from the commercial market. The second wave of data provide information on storage pro-tectant use after the 2010 season when the storage chemical subsidy, the fertilizer subsidy, and the seed subsidy were all in full effect. During that season households could potentially purchase storage chemicals from either commercial or subsidized sources. As a result, this article should provide useful insights about acquisition to storage technology and how it potentially serves as a complimentary input to fertilizer and seed.

There is a growing literature measuring the impact of input subsidy programs on smallholder behavior and well-being in SSA. One related study in Malawi finds that households who acquire subsidized seed and fertilizer plant a significantly larger share of their land to maize and tobacco, the crops targeted by the country's input subsidy program, than do other households (Chibwana et al., 2012). Another study uses household-level panel data from Malawi and Zambia and finds that in both countries households who acquire subsidized improved maize seed varieties purchase significantly less improved seed varieties on the commercial market (Mason and Ricker-Gilbert, 2013). The present study adds to the literature on input subsidies by estimating the impact of storage chemicals on a farmer's improved seed adoption decision in the context of a large-scale input subsidy program.

To our knowledge, there is little research investigating the relationship between investment in storage technology and adoption of improved maize varieties. One previous study in Ghana (Gyasi et al., 2003) and one study in Zambia (Langyintuo and Mungoma, 2008) consider how a farmer's perception of hybrid maize storabil-ity affects his or her decision to adopt it. Both studies estimate

4 Smallholders receiving (100%) subsidized improved seed acquired on average

5.7 kg and purchased an average of 0.9 kg. The 61% of smallholder farmers not receiving subsidized seed purchased an average of 5.5 kg of commercial seed (Mason and Ricker-Gilbert, 2013).

hybrid maize adoption and include "storability" as a dummy variable equal to one when a farmer perceives that hybrid maize stores better than local varieties and 0 otherwise. However, these studies do not consider a farmer's ability to protect maize stocks in their model. One limitation of the previous approach is that there is likely limited variation in the storability dummy, as evidence from Malawi suggests that most farmers believe local varieties to store better than hybrid (Smale, 1995; Lunduka et al., 2012). Therefore, the present article builds upon past work by considering how accessing storage protectants affects a farmer's decision to adopt improved varieties of maize.

In this article we first set up a model of smallholder maize adoption decision making, where the farmer chooses whether or not to adopt improved maize varieties as a binary decision. Second we model the farmer's decision of how much absolute area to plant to improved maize varieties. Third we estimate the farmer's decision on the share of his or her area to plant to improved maize varieties. The key right hand side (RHS) variable of interest is whether or not the household used storage chemicals on their maize crop after the previous harvest. In doing so, we empirically test whether or not households who access storage chemicals are significantly more likely to adopt improved maize seed and also plant larger areas of land to improved maize varieties in the next growing season. Since the key RHS variable is whether or not the household uses storage chemicals after the previous harvest it is pre-determined when the household makes planting decisions the following season. This structure avoids possible concerns about reverse causality. In addition, we use several panel estimation techniques including first-differencing and the Mundlak-Chamberlain device to deal with potential correlation between covariates and unobservable factors that could potentially bias our coefficient estimates, particularly those variables that represent participation in the input subsidy program.

The rest of this article is organized as follows. In the next section we present a background of Malawian post-harvest challenges, improved maize adoption, and the input subsidy program. Then introduce the conceptual model, the empirical model, and the identification strategy. Subsequently, data, results, and conclusions are presented.

Background

Post-harvest losses in Malawi

Post-harvest storage losses in Southern Africa are predominately caused by molds, rodents, and insect pests (World Bank, 2011). The main harvest in Malawi is followed by a long dry season so mold damage to grain is not a significant storage problem for smallholders. Nevertheless, post-harvest grain damage due to insect pests is a major issue. While producers have always dealt with the maize weevil as a dominate pest, improving smallholder maize storage practices in Africa has become increasingly more important over the past thirty-five years since the larger grain borer (LGB) was accidentally introduced in Africa from Central America in the 1970s and 1980s (Golob, 2002). Lacking natural predators, LGB's nearly simultaneous initial infestation in Tanzania and Togo have since expanded throughout both Eastern and Western Africa. As a result farmers have had to abruptly and fundamentally shift storage practices in this time to avoid inevitable stock destruction as the threat from LGB has increased (Addo et al., 2002). LGB supposedly entered Malawi in 1991/92 through trade shipments from Tanzania through the northern district of Chitipa. LGB is now prevalent in almost every district of Malawi and poses an enormous constraint on smallholder maize storage (Singano et al., 2008).

In the past many farmers throughout the continent preferred to store husked maize on cob, but the husk provides LGB with a more

stable brace to penetrate grains. Shelled maize creates a less stable environment to somewhat mitigate losses, though admixing insecticides is universally recommended for medium to long term storage in LGB-infested zones (Golob, 2009). Previously, insecticides such as Actellic contained only a pirimiphos-methyl compound which effectively controls the maize weevil. Blends were found to best control LGB in long term storage, however, and heavy research investments led to the release of new products blended with permethrins or deltamethrins (Golob, 2002). The Actellic Super or Shumba Super labels are two widely available brands which combine the lethal chemicals for both pests, used in Malawi and elsewhere on the continent.

There is limited information about on-farm storage practices in Malawi. However, Jones (2012) uses data from the nationally representative Agricultural Input Support Survey conducted after the 2008/09 season in Malawi and finds that nationally 45% of households use storage chemicals. In addition, 54% of households store local varieties of maize in woven or plastic bags, while 78% of households store improved varieties of maize in woven or plastic bags. Jones also notes that farmers report losing 8.5% of their improved maize in storage, and 7.4% of their local maize in storage. This information is descriptive in nature, and does not account for the possibility that farmers storing improved maize may be more likely to treat it with chemicals than if they are storing local varieties. It does, however provide some useful prima facia evidence about on-farm storage practices among Malawian smallholders.

Use of improved maize varieties in Malawi

The spectrum of improved varieties available for Malawian farmers has changed greatly over the last several decades. Smale (1995) documents a structural shift in the 1990s as national research institutions began to push away from traditional improved dent varieties to improved semi-flint varieties. The flinty texture allowed farmers to increase yields while better maintaining desirable post-harvest qualities such as high flour-to-milling ratios, and better natural resistance to maize weevils. However this has evolved into a present-day reversion back to largely dent varieties, including selections from multi-national corporations like Pioneer and Monsanto. While the reasons driving this reversion to more storage susceptible varieties is not the subject of this study, the farmer is ultimately left with little choice outside of dent varieties when sourcing improved seed. Grain damage in storage is thus a large concern for all dent-growing producers who must later cope with pests like LGB and maize weevil. In fact a recent study, Lunduka et al. (2012) use data from the Mulanje district of Malawi and find that many farmers prefer local varieties of maize to improved varieties because of their storability, taste, ease of pounding, and high flour-to grain ratios, despite the fact that they know improved maize varieties have higher yields.

Storage chemical subsidies in Malawi

The Malawian government introduced subsidized storage chemicals in 2008/09 in acknowledgement of the growing constraint posed by storage pests. The maize storage chemical subsidy ran through the 2011/12 season. In the 2011/12 season, the price of subsidized storage chemicals was 100 Kwacha per 200 g bottle of actellic, as compared to prices of 250-350 MK per bottle in retail outlets (author's observation). Following recommended application doses of 25 g/50 kg maize grain, each should protect 400 kg of maize.5 Unlike the improved seed and fertilizer subsidy program, no vouchers

5 It is reported that application rates vary greatly by farmer. Some may overdose for longer protection, while others apply less due to financial constraints.

are required for the storage chemical subsidy. Any farmer is permitted to purchase as many subsidized bottles as he or she needs or can afford from the Extension Planning Area (EPA) offices while stocks remain, although extension agents have authority to regulate this quantity as they deem appropriate. Stock shortages are common and anecdotal evidence suggests that they vary by region since allocation is determined by district maize production.6

Fertilizer and seed subsidies in Malawi

Fertilizer subsidy programs have existed in almost every year for decades in Malawi. However, after a drought-affected poor harvest in the 2004/05 growing season, the government decided to greatly expand its subsidized fertilizer program and continue subsidizing improved maize seeds, under the Farm Input Support Program (FISP). The program uses vouchers to target farmers who meet certain criteria. These targeted farmers can then redeem the vouchers for inorganic fertilizer at a reduced price and improved maize seed for free. During the 2008/09 growing season (the first year of the data used in this study), the government of Malawi made 202,000 metric tons of subsidized fertilizer and 5365 tons of subsidized seed available to farmers. The program cost an estimated US $265 million (Dorward and Chirwa, 2011). The government paid greater than 90% of the commercial fertilizer cost for farmers who received the subsidy that year. Recipient farmers were officially required to pay the equivalent of US $5.33 for a 50 kg bag of fertilizer that cost between US $40 and $70 at commercial prices, while vouchers for improved maize seed could be redeemed at no charge. From 2008/09 to present, all subsidized fertilizer vouchers had to be redeemed at government depots, while households could redeem their maize seed vouchers at a wide range of large and small input suppliers' stores. Officially each targeted household was supposed to receive two coupons good for two 50-kg bags of fertilizer at a discounted price, and one coupon for a 2 kg bag of hybrid maize seed or a 4 kg bag of OPV seed. In reality, the actual amount of subsidized fertilizer and seed acquired by households varied greatly.

Throughout the years of the subsidy's implementation, the process of determining who received coupons for fertilizer and seed was subject to a great deal of local idiosyncrasies. In 2007/08 and afterward there was a shift in allocation from area under cultivation to allocation based on farm household population and hence as shift in relative allocations from the North and Central regions to the Southern region (Dorward and Chirwa, 2011). At the village level, subsidy program committees and the village heads were supposed to determine who was eligible for the program. In more recent years open community forums were held in some villages where community members could decide for themselves who should receive the subsidy. From about 2008 ''vulnerable households" were officially supposed to be targeted with priority given to resource poor households, including disabled, elderly, female, and child-headed households. However, numerous unofficial criteria may have been used in subsidized seed and fertilizer application, such as a household's relationship to village leaders, length of residence, and social and/or financial standing of the household in the village.

Methods

Conceptual framework

Consider a smallholder household's decision whether or not to plant a piece of land with improved maize varieties that are higher

6 Author's observations through interactions with officials in Blantyre, Zomba, Thyolo, Lilongwe, Nkhotakota, and Mzimba offices in June/July 2011 and Jan/Feb

yielding, but offer less natural protection against insect pests when stored, vs. planting a traditional maize variety that may be lower yielding, but be less susceptible to damage from pests in storage. Assume that the household will plant the improved variety if p(I) > p(L), so that profits p, from planting improved varieties I, are greater than or equal to the profits from planting local varieties, L. Assuming that other inputs besides storage chemicals are held constant, the household understands that p(I) = Pm(XI) - Pc(CI), where Pm represents the market price for the quantity of improved maize produced XI and Pc represents the price of a given quantity of storage chemicals CI applied to improved maize.7 The household also understands that p(L) =Pm(XL) - Pc(CL) where XL represents the quantity of local maize produced, and CL represents a quantity of storage chemicals applied to local maize. If we assume that XI > XL, then the household produces more maize per area of land with improved varieties than with local varieties. However, if CI > CL, then improved varieties require a greater quantity of storage chemicals to be applied to a given quantity of maize than do local varieties. Therefore, the household must decide between higher revenue/higher cost improved varieties, and lower revenue/lower cost local varieties.

In addition, it is widely known that maize prices in many parts of SSA increase greatly after harvest. In fact market price data from the Malawian Ministry of Agriculture and Food Security show that real prices typically increase 50-100% within six months of the harvest season (Government of Malawi, various years; Chapoto and Jayne, 2010). Therefore, the household can increase profits if it is able to hold stocks until later in the marketing year because Pm will rise accordingly as maize becomes scarce. If the household can hold stocks until Pm increases sufficiently, then improved varieties will generate higher revenues than local varieties because XI > XL. However, the household must be able to overcome the potential dry-weight loss caused by pests when grain is placed in storage. This can be achieved through using storage chemicals on improved maize, but this comes at an extra cost which reduces to profitability of improved maize varieties because CI > CL.

The presence of a subsidy for storage chemicals like the one in Malawi beginning in 2008/09 reduces Pc to Pc'. This lowers the input/output price ratio of Pc/Pm, since Pc > Pc0. Therefore, the adoption of improved maize varieties will become more attractive under subsidization because Pc0(CI) will decline faster than Pc0(CL) which will cause a larger increase in p(I) than in p(L).

In our model the farmer considers using 3 interrelated inputs: storage chemicals, inorganic fertilizer, and improved maize seed. The farmer's decision making process is thus examined in accordance with the sequential input adoption literature, which considers that theses inputs form a package that a farmer may choose to adopt entirely at the same time (simultaneous adoption), or in different components at different times (sequential adoption) (Leathers and Smale, 1991; Ersado et al., 2004; Chavas and Di Falco, 2012). The literature cites relative prices of the inputs, risk aversion and understanding of the technology as reasons why farmers may make sequential rather than simultaneous adoption decisions. Leathers and Smale show that in the presence of incomplete information, farmers may make a rational decision to adopt part of an input package, even when it would be more profitable for them to adopt the package as a whole. Ersado et al. present an empirical test of sequential vs. simultaneous adoption. The authors use a likelihood ratio (LR) test to compare a restricted model of simultaneous adoption where adoption of all inputs occurs together, vs. an unrestricted model where the impact of each the technologies on adoption is considered separately in a

7 Anecdotal evidence suggests that in Malawi local maize receives a higher price per kg than improved varieties due to its desirable storage and consumption characteristics. However, we set that aside for parsimony in the conceptual model.

sequential adoption framework. We consider this test of sequential vs. simultaneous adoption in the empirical model presented in the next section.

Empirical model

Improved maize adoption

We operationalize the conceptual model presented above, where household i at time t must decide (i) whether or not to adopt improved maize varieties, (ii) the total area to plant to improved maize varieties, and (iii) the share of its land to plant to improved maize varieties. These decisions are a function of the following factors:

lit = b0 + ftCit-1 + № + bsSit + Aitb4 + Xit b5 + Wit b6 + Ritbi

+ Dit b8 + ai + Sit (1)

where I again represents the household's improved maize adoption decision. The variable for whether or not the household acquired storage chemicals after the previous harvest is represented by C. In the first wave of our data collected after the 2008 harvest, all storage chemicals come from commercial sources, while in the second wave households can purchase from subsidized or commercial sources. We use a variable = 1 if the household used storage chemicals after the previous harvest and 0 otherwise.8 The coefficient ft tests the key hypothesis of whether or not households who used storage chemicals after the previous harvest are more likely to adopt improved maize varieties. The impacts of this study are predicated on the assumption that use of storage chemicals after one season is associated with increased planting of improved maize varieties the next season. Since a household makes the decision to acquire storage chemicals after the harvest that occurs in May, that decision is complete by planting time beginning the following October.9

Kilograms of subsidized fertilizer that the household acquires in year t is represented by F and the kilograms of subsidized improved maize seed that the household acquires in year t is represented by S. Their respective parameters are b2, and b3. Eq. (1) is presented as a sequential adoption model, and including F, and S controls for the extent that use of other inputs affect the decision to plant improved seed at time t. However, we recognize that the decision to adopt all three inputs may be made simultaneously, so we conduct a LR test following Ersado et al. (2004) to compare the model in Table 1 with a model where adoption of improved maize is considered by a single decision where the RHS variable = 1 if the farmer uses storage chemicals, or subsidized fertilizer, or subsidized improved maize seed at time t, and 0 otherwise. Results of the LR test confirm that the simultaneous adoption model can be strongly rejected (p-value = 0.000) in favor of the sequential adoption model presented in Eq. (1). This finding provides evidence that in our context, farmers in Malawi make decisions about storage chemicals, fertilizer and improved seeds in a sequential manner.

In Eq. (1) credit and market access factors that may affect a household's decision to plant improved maize seed are represented by the vector A, while b4 represents the corresponding parameter vector. These factors include (1) distance to paved road in kilometers, (2) distance to the main market in kilometers, (3) distance to

8 We treat C as a binary variable rather than a continuous variable representing the kilograms of storage chemicals acquired because thorough analysis of the data combined with discussions in the field confirm that many households do not know the quantity of storage chemicals that they acquire and apply. Furthermore some households acquire storage chemicals in liquid form, while others acquire it in powder form making it hard to convert to equivalent measures. Therefore to eliminate measurement error we model storage chemical acquisition as a binary decision.

9 Even if insect damage emerges 2-3 months after harvest and thus induces treatment, this occurs well before the next planting season in Malawi.

Table 1

Descriptive statistics of variables used in the analysis.

Variables

2008/09

2010/11

Median

Median

Dependent variables

=1 if household plants improved maize seed Hectares of improved maize seed planted Share of total area planted to improved maize seed Share of total maize area planted to improved maize seeda =1 if HH used commercial storage chemicals after harvest

RHS variables

=1 if HH used subsidized storage chemicals after previous harvest

=1 if HH used storage chemicals after previous harvest (subsidized or commercial)

kgs. of subsidized seed acquired in current year

kgs. of subsidized fertilizer acquired in current year

=1 if farm credit organization in village

Distance to paved roadb (km)

Distance to main marketb (km)

Distance to extension services (km)

Number of dealers who sell subsidized inputs in village

Value of household assetsc ('000 kwacha)

Area cultivated (in ha)

Landholding (in ha)

Age of household head in first survey yearb =1 if female headed household Adult equivalents

=1 if death in the family over past two years =1 if primary (grades 1-4) =1 if upper primary (grades 5-8) =1 if secondary (grades 8-12) =1 if post-secondary

Past year hungry season maize pricec (kwacha/kg) Past year harvest season maize pricec (kwacha/kg) Price of NPK & Urea fertilizer11 (kwacha/kg) Agricultural wage ratec (kwacha/day) Average rainfall, past five growing seasons (in cm) Coefficient of variation on past rainfall

0.69 0.32 0.39 0.44 0.51

0.00 0.51 2.29 65.88 0.32 16.92 39.53 6.11 0.60 48.07 0.96 1.12 44.78 0.32 4.16 0.10 0.25 0.34 0.13 0.01 38.15 45.05 160.33 330.74 822.59 0.11

0.20 0.30 0.33

2.00 50.00

12.00 32.00 5.00 0.00 13.75 0.81 0.81 42.00

39.29 44.40 153.08 283.61 820.20 0.12

0.80 0.41 0.48 0.56 0.48

0.11 0.58 3.69 54.00 0.27 16.43 38.71 4.79 0.26 65.94 0.95 1.17 44.24 0.31 4.17 0.05 0.38 0.34 0.12 0.01 43.53 31.68 97.34 243.14 859.77 0.10

0.40 0.44 0.50

4.00 50.00

10.00 30.00 3.00 0.00 13.50 0.81 0.81 41.00

43.18 32.44 100.00 214.29 861.80 0.09

a Regression results for share of maize area planted to improved maize varieties are not shown for space considerations, as they do not differ fundamentally from share of total area planted to improved maize varieties. These results are available from the authors upon request. b Corresponding variable is time constant and does not vary over time, means and medians may vary over time due to weighting the observations by IPW * survey weights. c Variable is converted to real 2011 kwacha. US $1.00 = 151.55 kwacha in 2010/11 (Chirwa and Dorward, 2013).

extension services in kilometers, (4) number of input suppliers in the village, and (5) whether or not there is a farm credit organization in the village. Household demographics that affect improved seed adoption are represented by the vector X, while b5 represents the corresponding parameter vector. These include (1) value of household assets, (2) household landholding, (3) adult equivalents, (4) if the household is female headed, (5) education of the household head. Factors such as assets, landholding and education of the household head proxy for household understanding and ability to take risks which also influence the adoption decision in a sequential adoption model. Prices that affect the decision to adopt improved seed are represented by the vector w. Relevant prices are (1) commercial price of fertilizer (NPK & urea), (2) agricultural wage rates in the community, (3) previous year hungry season maize price (January to March) and previous harvest season maize prices (May to July), while represents the corresponding parameter vector. Including the previous year's maize price assumes that farmers have naïve expectations about maize prices, but it serves to proxy for the maize price farmers may expect in the coming year. Including prices controls for the exogenous changes that impact relative prices, which affect sequential adoption decisions. Average rainfall over the previous five growing seasons and the coefficient of variation on average rainfall over the previous 5 growing seasons are represented by R, while b7 represents the corresponding parameter vector. These variables are lagged over 5 years in order to proxy for the naïve expectation of what a farmer expects rainfall to be in the coming year, when he or she makes decisions about seed varieties at planting.

Year and region fixed effects are represented by a vector of dummy variables denoted by D, while b8 represents the corresponding parameter vector (see Table 1 for a full list of explanatory variables). The error term in Eq. (1) has two parts. The time constant-unobserved heterogeneity is represented by ai, and the unobserved time-varying shocks are represented by eit.

Crowding out of commercial storage chemicals by storage chemical subsidy

In order to understand the impact that the storage chemical subsidy has on the probability of using commercial storage chemicals, it is important to understand how acquisition of subsidized chemicals may affect a farmer's decision to use commercial storage chemicals. Following the work of Xu et al. (2009), Ricker-Gilbert et al. (2011), and Mason and Jayne (2013) who conceptualize crowding out in the context of subsidized fertilizer, consider the following equation for the probability that a farmer will use storage chemicals, either subsidized or commercial10:

Cit-1 = «0 + a1Sit-i + Ait «2 + Xu a.3 + R^ + Dit «5 + bi + utt (2)

where C is a binary variable representing whether or not a household purchases storage chemicals on the commercial market, and S is a binary variable representing whether or not the household

10 The previous studies that have addressed crowding out in the context of input subsidy programs have done so for fertilizer, which is modeled as a continuous variable. Therefore, in this application we adapt the crowding out framework to the binary decision of whether or not to use storage chemicals.

acquires storage chemicals from subsidized sources. The coefficient estimate, a, tells us the degree to which acquiring subsidized storage chemicals affects the probability that a household will purchase storage chemicals commercially (e.g. the crowd out or crowding in effect). If a > 0, then acquiring subsidized storage chemicals is said to crowd in commercial chemical use. Conversely if a < 0, then acquiring subsidized storage chemicals is said to crowd out commercial chemical use, and if á1 = 0 there is no effect. Since C and S are both binary variables, ái is computed as the average partial effect (APE). The other variable vectors in Eq. (2) are the same as they are in Eq. (1), and the corresponding a's represent the parameters to be estimated. The model in Eq. (2) excludes the vector of prices, denoted by w in Eq. (1), because the prices we have available are determined during the next agricultural season, after the storage chemical purchase decision from the past harvest has been made.

Identification strategy

¡deal identification strategy

In an ideal world we could identify the impacts of storage chemical use, subsidized seed use and subsidized fertilizer use via a randomized control trial (RCT) design where a randomly chosen group of farm households would be given the opportunity to obtain and apply these inputs on their farm. This RCT design would allow us to compare impacts from the treatment group with a control group of farm households who do not receive the inputs. RCTs are now considered the ''gold standard'' of impact evaluation in the development economics literature, because ideally they should allow us to obtain an unbiased average treatment effect (ATE) of using storage chemicals, subsidized seeds and subsidized fertilizer (Duflo et al., 2007). While it would have been ideal to have the ability to evaluate the effects of these inputs in an RCT framework in our context, given the fact that the government of Malawi rolled out the input subsidy programs without conducting any pilot program or considering the need to measure program impacts in an experimental framework, obtaining the ''gold standard'' is impossible.

Nevertheless, for evaluating the impact of storage chemicals on improved maize adoption, the panel dataset used in this study gives us the ability to measure before and after effects of the subsidy program and within household changes over time. Using panel methods with a well specified model can remove some of the end-ogeneity concerns related to using non-experimental methods (Wooldridge, 2010). Furthermore, RCT designs are subject to the critique of limited external validity, as the results obtained in one context may not be relevant in another context (Ravallion, 2009; Barrett and Carter, 2010). In addition, randomizing treatments in agricultural impact assessment has recently come under criticism for failure to control for the effort effect (Bulte et al., 2014). In this situation, participants who are randomly chosen to receive a treatment systematically adjust their level of effort to increase the impact above and beyond the effect of the treatment itself. Failure to account for this effort effect will bias the ATE and overestimate the impact of a treatment.

Current identification strategy

Ultimately given our context and data at hand we do the best possible job we can of identifying consistent impacts of storage chemical use on improved maize adoption in Malawi in a non-experimental context. We need to deal with several modeling challenges in our study to consistently estimate the impact of acquiring storage chemicals on a household's improved maize adoption decision. The first issue is potential reverse causality between the key RHS variable and the dependent variable. The argument in this paper is that accessing storage chemicals affects the household

decision to adopt improved maize varieties. However, one might argue that the relationship goes the other way because households who decide to grow improved varieties may be more likely to acquire storage chemicals if they believe the chemicals are necessary to prevent storage losses. The structure of our analysis should eliminate this concern because the RHS variable that we use is whether or not the household used storage chemicals after the previous harvest, which occurs around May in Malawi. Therefore, that decision is clearly pre-determined before the planting decision is made for the following season, which usually occurs sometime between October and December of the same calendar year in Malawi.

The second potential identification issue is that households are heterogeneous in their ability to acquire storage chemicals. In Malawi, households can go to the market and purchase the quantity of storage chemicals they need or can afford. With the advent of the storage chemical subsidy in 2008/09, any farmer in Malawi could visit an extension office and purchase as many bottles of protectant as they want or need for 100 Kwacha per bottle, subject to availability. To deal with potentially uneven access to storage chemicals by households we include variables such as household assets, distance to the local extension office, distance to roads, and number of dealers who sell subsidized inputs in the village in our empirical models.

The third issue is that even after controlling for observable household-level and community-level access factors, there could be left over unobserved heterogeneity that affects use of storage chemicals and also a households' decision to adopt improved maize varieties. Other studies related to adoption of inputs in SSA have found different estimates when unobserved heterogeneity is controlled for and when it is not (Suri, 2011; Mason and Ricker-Gilbert, 2013). For example, some households may be more motivated to acquire storage chemicals because they are more aware of pest risk than other farmers. Other farmers could just be more talented and know how to manage pests without chemicals. Factors like motivation and talent are unobservable in our models of improved seed adoption and, if ignored, can cause bias coefficient estimates to the extent they are correlated with storage chemical use.

Fortunately, we have panel data that allows us to address the issue of unobserved heterogeneity. We first estimate all models as linear using a household-level first difference (FD) estimator, which measures the difference in the variables of interest between 2008/09 and 2010/11. As such we estimate Eq. (1) in FD form as follows:

Aiu = Aft, + Ab1Ctt_1 + A^F* + AftS* + AAit £4 + AX^s

+ Awit be + ARit £7 + AD it ft + Ae¡t (3)

where D represents the variables in FD form. Eq. (3) demonstrates that the FD estimator removes unobserved time constant heterogeneity, ai in Eq. (1), from the model. In this application the FD estimator is preferable to similar but simpler difference-in-differences (DID) estimator, because the DID only measures changes in the main variable Cit_1, while the FD estimator measures changes for all RHS variables, and thus completely removes any correlation between ai and all RHS variables, including subsidized seed and subsidized fertilizer, from the model. Note that Eq. (2) is operation-alized in an analogous manner using the FD estimator.

The FD estimator offers consistent estimates when models are structured in linear form. However, the dependent variables in our analysis may have non-linear distributions. Unfortunately, FD generates inconsistent parameter estimates when applied to non-linear models due to the incidental parameters problem (Wooldridge, 2010). Fortunately, we can use the Mundlak-Chamberlin (MC) device following Mundlak (1978) and Chamberlain (1984) to deal

with correlation between unobserved heterogeneity and RHS variables in non-linear models.11 The MC device deals with potential correlation between unobserved heterogeneity and RHS variables by decomposing ai in the following way:

ai = Uij + Xi n + rit

The MC device assumes that rit | Xi ~ Normal(0, of); where X¡ is the household time average of all time-varying covariates in Eq. (1). Therefore, to operationalize the MC device, X¡ needs to be included as a covariate in all equations. This specification provides coefficient estimates that are analogous to FD or household fixed effects estimation (Wooldridge, 2010).

For the purposes of this paper the time varying shocks Aeit are assumed to be uncorrelated with the covariates in our models. It may be possible that the quantity of subsidized fertilizer and/or subsidized seed acquired by the households may be correlated with time-varying shocks Aeit. The only way explicitly around this is via instrumental variables (IV), we would need 3 IVs in this context, or an exogenous treatment. However, a recent study in Malawi uses the MC device to deal with correlation between RHS variables and unobserved heterogeneity, along with the control function approach using IV to deal with correlation between subsidized inputs and time-varying shocks (Mason and Ricker-Gilbert, 2013). The study finds that controlling for unobserved heterogeneity affects coefficient estimates, but once unobserved heterogeneity is controlled for correlation between time-varying shocks and subsidized seed and fertilizer is found to not to have as statistically significant effect on the coefficients. Therefore, in the present study we assume that subsidized seed, subsidized fertilizer, and subsidized storage chemicals are uncor-related with time-varying shocks. That being said, as with any study using observational data, our results cannot be considered fully causal.

Functional form and estimator choice12

Adoption of improved maize varieties

The decision whether or not to adopt improved maize varieties is estimated as a binary decision where the dependent variable takes on a value of 1 if the household adopts improved maize varieties and 0 otherwise. We estimate this decision first as a linear probability model (LPM) using the FD estimator and then by a probit with the MC device. The LPM assumes that the marginal effects are linear and ignores any potential non-linear relationships. The probit estimator allows us to consider that the adoption response may be non-linear across the distribution of our data. These models are first estimated in a parsimonious manner only includes the key RHS variables of interest, if the household used storage chemicals, kilograms of subsidized fertilizer acquired, and kilograms of subsidized seed acquired, along with year and region dummies. Results from the parsimonious model are then compared with the full model that includes the entire set of controls presented in Eq. (1) on the RHS.

Area planted to improved maize varieties

The decision of how much area to plant to improved maize varieties potentially takes on the property of a corner solution variable, because a significant number of households do not grow improved varieties. However, beyond that the distribution of area planted is

11 The Mundlak-Chamberlin device is also sometimes referred to as the correlated random effects (CRE) estimator.

12 Results showing the coefficient estimates for the time averages of the covariates, obtained using the MC device are available in Appendix 1 online. Results using pooled estimation that does not control for unobserved heterogeneity are available in Appendix 2 online.

relatively continuous. Therefore, we compare results when the model is estimated linearly with results using a tobit estimator that accounts for the variable's corner solution distribution. First the area planted model is estimated using linear FD and results are compared when the model is estimated via tobit with the MC device. Coefficient estimates using parsimonious models specifications are compared to full models specifications just as in the binary adoption model presented above.

Share of area planted to improved maize varieties

The decision of share of total area to plant to improved maize varieties can be captured in a fractional response because the total share must lie between the 0 to 1 range. We compare results using the linear FD model with results using a fractional probit with the MC device, which explicitly constrains the predicted value between 0 and 1. As in the above models, coefficient estimates using parsimonious models specifications are compared to full models specifications.

Crowding out/in of commercial storage chemicals

The decision whether or not to purchase commercial storage chemicals is estimated as a binary 0 or 1 decision. Therefore, the estimators used in this model are the same as the decision whether or not to adopt improved maize varieties. We estimate this decision first as a linear probability model (LPM) using the FD estimator and then by a probit with the MC device. Parsimonious model results are compared to the results of the full model.

Note that all coefficient estimates that are generated via probit, tobit, and fractional probit are reported as average partial effects (APE) using the 'margins' function in STATA.

Data from this study come from two waves of surveys on smallholders in Malawi. The first round of data comes from the Agricultural Inputs Support Survey II (AISS2) which was conducted after the 2008/09 growing season in Malawi. The second round of data comes from the Agricultural Input Support Survey IV (AISS4) conducted after the 2010/11 growing season. The data were collected by Wadonda consulting and the two data sets give us a balanced panel of462 households in 8 districts, across all 3 regions of Malawi. The sample represents 8 major maize growing livelihood zones covering 77% of all rural households (Wadonda Consulting, 2011).

The AISS2 and AISS4 build upon two earlier nationally representative surveys, the Second Integrated Household Survey (IHS2) in Malawi collected during the 2002/03 and 2003/04 growing seasons, and the 2007 Agricultural Inputs Support Survey (AISS1) conducted after the 2006/07 growing season. Unfortunately, questions related to household storage decisions were only asked during the AISS2 and AISS4 surveys and not in any of the earlier surveys. Therefore we have to treat the data as a two wave panel. However, we use inverse probability weights (IPW) multiplied by the survey weights to deal with household attrition and ensure that our sample which remains in the AISS2 and AISS4 are representative of Malawi's smallholder population. The IPW technique involves three steps: (i) use probit to measure whether observable factors in one wave affect whether a household is re-interviewed in the next wave; (ii) obtain the predicted probabilities (Prit) of being re-interviewed in the following wave; (iii) compute the IPW = (1/Prit) and apply it to all models estimated. For households originally sampled in IHS2, the IPW for household i in AISS1 = 1/PriAISS1. The IPW in AISS2 = 1/(PriAISS1 * PriAISS2), while the IPW for AISS4 is 1/(PriAISS1 / PriAISS2 / PriAISS4). (For more information on IPW see Wooldridge, 2010). We multiply the IPW by the survey sampling weights in the first wave to control for the probability of

the household being selected for interview from the population. The models estimated by OLS, probit, and FD include the IPW / survey weights. The models estimated via tobit do not include this weighting because IPW is not valid in such models. However the results do not differ in any meaningful way when the IPW is used and when it is not, so attrition issues should not be a major concern in this application.

Landholding and area cultivated

The variables for landholding and area cultivated are constructed using the household survey data from farmer estimates of plot sizes. The RHS variable for landholding is based on the amount of land that farmers say that they have the right to cultivate. It is computed as the sum of crop land, fallow land, virgin land, orchards, and land rented out, but excludes land rented in. Landholding is used as a right-hand-side variable in this analysis to proxy for household wealth.

Area cultivated is constructed as the amount of land that a household cultivates for rainy season crop production during the corresponding year. This calculation includes land rented in but not land rented out.13 Area cultivated variable is used to create the dependent variable for area planted to improved maize and share of area planted to improved maize. Since many plots are intercropped in Malawi it is difficult to accurately aggregate exactly how much intercropped land is allocated specifically to maize, and not to other crops on the same intercropped plot. Therefore, for practicality the dependent variables for area planted to improved maize and share of area planted to improved maize should be thought of as area with improved maize cultivated on it.

Prices, wage rates, and rainfall variables Fertilizer prices

Fertilizer prices used in the study are calculated from the survey as Malawian kwacha per kilogram of commercial maize fertilizer. The price is calculated as an average of urea and Nitrogen/Phosphorus/Potassium (NPK) prices, which are the primary fertilizers applied to maize in Malawi. These prices are based on what survey respondents say they pay for commercial fertilizer during the planting season, generally from October to December in Malawi. For those buying fertilizer commercially we use the observed price that they pay, while for those who do not buy commercially we use the district median price to proxy for the price that the household faces for the input.

Maize prices

Data for the variable representing the median hungry season maize price in the household's district during the previous year, and the variable representing the median harvest season maize price in the household's district during the previous year both come from district-level data on maize retail sales, collected by the Malawian Ministry of Agriculture.

Wage rate calculations

Agricultural wage rates are calculated as the price per day of hiring in labor from the household survey. We use the observed price for households who hire in labor, and for those who do not hire in labor, we use the district median wage rate to proxy for the price that they would face to hire workers. The top and bottom 5% of computed wage rates are replaced with district median wage rates to remove outliers.

Rainfall

Locally interpolated time-series data on rainfall come from the University of East Anglia's Climate Research unit (CRU)-TS 3.1 Climate Database (Climate Research Unit, 2011; Mitchell and Jones, 2005).

These data are considered to be one of the most reliable sources of rainfall data that is available. They are geo-referenced to the household's enumeration area. The average past rainfall and coefficient of variation on past rainfall variables are constructed as the average over the past 5 years in the enumeration area, and they vary by survey wave. The variables are season specific and is structured as T-J to represent a farmer's naive expectation at the time of planting as to what he or she expects about the coming seasons rainfall.

All other explanatory variables are constructed from the household survey.

Results

Table 1 presents the means and medians of the variables used in the analysis. The descriptive statistics for the dependent variables indicate that the number of households planting improved varieties, hectares planted to improved varieties, and share of area planted to improved varieties have all increased between 2008/ 09 and 2010/11. In 2008/09, 51% of households purchase storage chemicals commercially, and 0% acquire storage chemicals through the subsidy program. In 2010/11, 58% of households acquire storage chemicals, with 11% of them acquiring the input through the subsidy program. Should the 11% of farmers who receive the subsidized storage chemicals be among those who ordinarily would not purchase storage chemicals in the absence of the subsidy, then one would not expect any impact on the commercial chemical market (in other words, no impact in crowding in or out of chemical retailers). In fact, further analysis of our data indicate that 154 of the 462 respondents in the survey (33% of the sample) bought commercial chemicals in both 2008/09 and 2010/11. Interestingly, 50 households obtained subsidized storage chemicals in 2010/11, and 27 of those households bought commercial chemicals in 2008/09. Since only 3 of those households also bought commercial chemicals in 2010/11 this suggest relatively significant prima facia evidence of crowding out (24/50 = 48%). However these numbers are descriptive and unconditional and do not control for other factors that could affect crowding out.14

Kilograms of subsidized seed acquired by households also increased during that period from an average of 2.29 kgs per household in 2008/09 to 3.69 in 2010/11. At the same time the average amount of subsidized fertilizer acquired per household declined from 65.88 kg in 2008/09 to 54 kg in 2010/11. It is also interesting to note that the average value of household livestock and durable assets increased substantially from 48,070 Kwacha in 2008/09 to 65,940 in 2010/11, while the median value of assets actually declined during that period from 13,750 in 2008/09 to 13,500 in 2010/11. This may indicate that a select few individuals at the top of the distribution are improving their situation, while the vast majority of smallholders are not accumulating any meaningful quantity of wealth.

Table 2 presents the total kilograms of improved maize seed acquired by households in the sample, and the percentage of households acquiring storage chemicals, disaggregated by source (subsidized or commercial) and survey wave (2008/09 or 2010/ 11). The results suggest that from the first to the second wave of our survey the amount of commercial seed use goes down as the seed subsidy goes up. This provides some prima facia evidence

13 Note that the correlation between landholding and area cultivated is 0.61 in our

dataset.

14 We thank an anonymous reviewer for making this point.

100 J. Ricker-Gilbert, M.Jones/Food Policy 50 (2015) 92-105 Table 2

Change in improved maize seed and storage chemical use, by source and survey wave.

Seed (total kgs) Storage chemicals (% adopting)

2008/09 2010/11 Difference 2008/09 2010/11 Difference

Subsidized 431,758 777,098 345,340 0 11 11

Commercial 605,432 499,740 -105,692 51 47 -4

Total 1,037,190 1,276,837 239,647 51 58 7

Table 3

Factors affecting whether or not household plants improved maize varieties.

(1) FD LPM Parsimonious

(2) FD LPM Full

Probit with MC device3 Parsimonious

Probit with MC device3 Full

VARIABLES Coeff. P-value Coeff. P-value Coeff. P-value Coeff. P-value

=1 if HH used storage chemicals after previous harvest 0.0857* (0.052) 0.0901** (0.042) 0.0688* (0.091) 0.0660* (0.089)

kgs. of subsidized seed acquired 0.0117* (0.074) 0.0109* (0.083) 0.0434*** (0.000) 0.0427*** (0.000)

kgs. of subsidized fertilizer acquired 0.0015*** (0.000) 0.0015*** (0.001) 0.0015*** (0.002) 0.0014*** (0.005)

=1 if farm credit organization in village -0.0410 (0.345) -0.0234 (0.566)

distance to paved road (km)1 0.0011 (0.312)

distance to main market (km)1 0.0007 (0.214)

distance to extension services (km) 0.0043 (0.161) 0.0047* (0.075)

number of dealers who sell subsidized inputs in village 0.0433 (0.118) 0.0458** (0.040)

log value of household assets2 -0.0234 (0.218) -0.0106 (0.451)

landholding (in ha) 0.0081 (0.599) 0.0073 (0.586)

age of household head in first survey year1 -0.0013 (0.138)

=1 if female headed household -0.0036 (0.958) -0.0187 (0.758)

log of adult equivalents 0.0199 (0.784) 0.0293 (0.605)

=1 if primary (grades 1 to 4) -0.0187 (0.761) -0.0247 (0.649)

=1 if upper primary (grades 5 to 8) -0.0579 (0.415) -0.0405 (0.541)

=1 if secondary (grades 8 to 12) -0.0041 (0.967) 0.0022 (0.982)

=1 if post-secondary -0.2248 (0.113) -0.1582 (0.324)

past year hungry season maize price (kwacha/kg)2 0.0083 (0.290) 0.0061 (0.399)

past year harvest season maize price (kwacha/kg)2 0.0079 (0.436) 0.0050 (0.565)

price of NPK & Urea fertilizer (kwacha/kg)2 -0.0008 (0.194) -0.0008 (0.202)

agricultural wage rate (kwacha/day) 2 0.0001 (0.405) 0.0000 (0.544)

average rainfall, past five growing seasons (cm) 0.0009 (0.655) 0.0010 (0.589)

coefficient of variation on past rainfall 0.4900 (0.792) 1.2299 (0.431)

Observations 462 462 924 924

R-squared

Coefficients in columns 3 and 4 are Average Partial Effects (APE) estimated via the margins command in Stata; all models include year and region dummy variables; standard errors clustered at the household level; FD = First Difference, LPM = linear probability model, MC = Mundlak-Chamberlain.

* Statistically significant at the 10% level.

** Statistically significant at the 5% level.

*** Statistically significant at the 1% level.

a Corresponding coefficient is time constant and does not vary over time.

b Variable is converted to real 2011 kwacha. US $1.00 = 151.55 kwacha in 2010/11 (Chirwa and Dorward, 2013). c Model includes time-averages of all time-varying covariates; R-squared is correlation-squared in columns 3 and 4.

of crowding out from the seed subsidy, as we might expect. In addition, storage chemical use follows the same trend and also shows some evidence of crowding out. However, the table indicates that there is still a significant market for commercial improved seed, and for commercial storage chemicals in both years of the survey. In 2008/09 commercial seed purchases make up 58% of total purchases in the sample, and in 2010/11 commercial seed purchases make up 39% of total seed purchases; a meaningful amount in both years. In 2008/09 100% of household who use storage chemicals acquire them commercially, because it was the year before the storage chemical subsidy began. In 2010/11, 81% of all households who acquire storage chemicals purchase them from commercial sources, which is still a very significant amount.

Table 3 presents the results for factors affecting the probability that a household adopts improved maize varieties. Across the estimators used in columns 1-4 we can see that acquiring storage chemicals is associated with a positive effect on the probability that a household adopts improved maize varieties. The effect is

statistically significant, with p-value<0.10 in all four columns. The coefficient estimates on the RHS variables of interest in parsimonious model specifications are very similar to those in the corresponding fully specified models. This lends confidence in the stability and consistency of the estimates. The coefficient estimates are consistent and indicate that in the linear FD models presented in columns 1 and 2, acquiring storage chemicals is associated with an increased probability that a household plants improved maize seed by between 8.57 and 9.01 percentage points on average. The results using probit with the MC device in columns 3 and 4 which considers possible non-linear effects, indicate that acquiring storage chemicals is associated with an increased probability that a household plants improved maize varieties by between 6.60 and 6.88 percentage points on average. Looking across columns it is also evident that acquiring an additional kilogram of subsidized seed and subsidized fertilizer are significantly associated with an increased probability that a household plants improved maize varieties by a relatively small amount. The direction of the effect is what we would expect ex ante.

Table 4

Factors affecting area that household plants to improved maize varieties, in hectares.d

(i) (2) (3) (4)

FD Linear FD Linear Tobit with MC device3 Tobit with MC device3

Parsimonious Full Parsimonious Full

VARIABLES Coeff. P-value Coeff. P-value Coeff. P-value Coeff. P-value

=1 if HH used storage chemicals after previous harvest 0.0855** (0.013) 0.0848** (0.021) 0.0754* (0.094) 0.0758* (0.071)

kgs. of subsidized seed acquired 0.0132*** (0.003) 0.0134*** (0.003) 0.0118*** (0.002) 0.0119*** (0.001)

kgs. of subsidized fertilizer acquired 0.0013*** (0.000) 0.0012*** (0.002) 0.0015*** (0.000) 0.0014*** (0.001)

=1 if farm credit organization in village 0.0206 (0.638) 0.0115 (0.774)

distance to paved road (km)1 0.0009 (0.315)

distance to main market (km)1 0.0013*** (0.009)

distance to extension services (km) 0.0049** (0.018) 0.0042 (0.141)

number of dealers who sell subsidized inputs in village -0.0146 (0.573) -0.0069 (0.773)

log value of household assets2 0.0086 (0.633) -0.0043 (0.757)

landholding (in ha) 0.0592* (0.092) 0.0363** (0.031)

age of household head in first survey year1 -0.0007 (0.339)

=1 if female headed household -0.0407 (0.325) -0.0261 (0.668)

log of adult equivalents 0.0097 (0.849) 0.0234 (0.689)

=1 if primary (grades 1 to 4) -0.0155 (0.731) -0.0094 (0.859)

=1 if upper primary (grades 5 to 8) -0.1148* (0.072) -0.1037 (0.102)

=1 if secondary (grades 8 to 12) 0.0019 (0.985) -0.0233 (0.793)

=1 if post-secondary -0.4960*** (0.004) -0.4363** (0.039)

past year hungry season maize price (kwacha/kg)2 -0.0001 (0.991) 0.0044 (0.530)

past year harvest season maize price (kwacha/kg)2 -0.0072 (0.386) -0.0029 (0.710)

price of NPK & Urea fertilizer (kwacha/kg)2 0.0006 (0.404) 0.0004 (0.466)

agricultural wage rate (kwacha/day)2 0.0000 (0.969) 0.0000 (0.640)

average rainfall, past five growing seasons (cm) 0.0001 (0.968) -0.0008 (0.673)

coefficient of variation on past rainfall -1.5971 (0.324) -1.4564 (0.314)

Observations 462 462 924 924

R-squared 0.071 0.134 0.138 0.274

Coefficients in columns 3 and 4 are Average Partial Effects (APE) estimated via the margins command in Stata; all models include year and region dummy variables;

standard errors clustered at the household level; FD = First Difference, LPM = linear probability model, MC = Mundlak-Chamberlain.

* Statistically significant at the 10% level.

** Statistically significant at the 5% level.

*** Statistically significant at the 1% level respectively.

a Corresponding coefficient is time constant and does not vary over time.

b Variable is converted to real 2011 kwacha. US $1.00 = 151.55 kwacha in 2010/11 (Chirwa and Dorward, 2013). c Model includes time-averages of all time-varying covariates; R-squared is correlation-squared in columns 3 and 4.

d For robustness we estimated the unconditional effects of the double hurdle (DH) model with MC device. Results indicate that the main coefficient estimates using the DH are similar to those in table using both FD and tobit with the MC device. In the DH model coefficient estimate for the storage chemical variable is 0.0872, for the subsidized seed variable it is 0.0237 and for the subsidized fertilizer variable it is 0.0012. Full results of the DH model are available to the reader upon request.

Table 4 presents factors affecting the total area in hectares that a household plants to improved maize varieties. Similar to the results in Table 3, Table 4 shows that across columns 1-4 acquiring storage chemicals is associated with a positive and statistically significant effect on the area that households plant to improved maize varieties (p-value < 0.10). Again coefficient estimates from the parsimonious models are very similar to the estimates in the fully specified models. When the model is estimated linearly using FD in columns 1 and 2, the average household who uses storage chemicals is associated with an increase in area planted to improved maize between 0.0848 and 0.0855 hectares. When potential non-linearities are considered in the tobit with MC device estimation in columns 3 and 4, results indicate that using storage chemicals is associated with an increase in area planted to improved maize between 0.0754 and 0.0758 hectares on average. Considering the fact that the average landholding in our sample is only about 1.15 hectares, the impact of acquiring storage chemicals on area planted to improved maize is not huge but is relatively meaningful. Across the columns of Table 4 it is also clear that acquiring and additional kilogram of subsidized seed and fertilizer is associated with a statistically significant and relatively small positive effect on area planted to improved maize varieties.

Table 5 shows the factors affecting share of total area that is planted to improved varieties. The results of Table 5 are consistent with those in Tables 3 and 4. They indicate that acquiring storage chemicals is associated with a statistically significant and positive

effect on share of area that a household plants to improved maize varieties. Again parsimonious model results are similar to full model results for key RHS variables. They indicate that when the models are estimated linearly via FD in columns 1 and 2, using storage chemicals is found to be associated with an increase in the share of area planted to improved maize varieties by between 9.06 and 9.88 percentage points on average. When potential non-linearities are considered in columns 3-4 using fractional probit with the MC device, accessing storage chemicals is found to be associated with an increase in the share of area planted to improved maize varieties by between 7.70 and 8.14 percentage points on average. There is also evidence in columns 3 and 4 to suggest that an additional kilogram of subsidized fertilizer is associated with a statistically significant and small increase the share of area planted to improved maize varieties, which is consistent with previous work by Chibwana et al. (2012).

The results from Tables 3-5 are consistent across models and estimator choice. They show that using storage chemicals after the previous harvest is associated with the household being statistically more likely to plant improved maize varieties, plant more area to improved maize varieties, and plant a larger share of area to improved maize varieties.

Table 6 presents the factors affecting whether or not a household acquires grain storage chemicals on the commercial market following the preceding harvest. The coefficient of the subsidized storage chemical variable provides the estimate of how storage chemicals

Table 5

Factors affecting share of total area planted to improved maize varieties.

(1) FD LPM Parsimonious

(2) FD LPM Full

Fractional Probit with MC device3 Parsimonious

Fractional Probit with MC device3 Full

VARIABLES Coeff. P-value Coeff. P-value Coeff. P-value Coeff. P-value

=1 if HH used storage chemicals after previous harvest 0.0988** (0.015) 0.0906** (0.023) 0.0814** (0.025) 0.0770** (0.031)

kgs. of subsidized seed acquired 0.0075 (0.208) 0.0077 (0.168) 0.0073 (0.221) 0.0078 (0.174)

kgs. of subsidized fertilizer acquired 0.0006 (0.129) 0.0006 (0.131) 0.0007* (0.076) 0.0007* (0.054)

=1 if farm credit organization in village 0.0022 (0.955) 0.0102 (0.773)

distance to paved road (km)1 -0.0017* (0.067)

distance to main market (km)1 0.0009** (0.042)

distance to extension services (km) 0.0084*** (0.002) 0.0078*** (0.001)

number of dealers who sell subsidized inputs in village 0.0044 (0.849) 0.0005 (0.983)

log value of household assets2 -0.0125 (0.468) -0.0139 (0.339)

landholding (in ha) -0.0106 (0.333) -0.0185 (0.146)

age of household head in first survey year1 -0.0006 (0.457)

=1 if female headed household 0.0180 (0.774) 0.0091 (0.873)

log of adult equivalents -0.0165 (0.764) -0.0130 (0.786)

=1 if primary (grades 1 to 4) 0.0376 (0.463) 0.0247 (0.593)

=1 if upper primary (grades 5 to 8) -0.0015 (0.982) -0.0095 (0.868)

=1 if secondary (grades 8 to 12) 0.0785 (0.362) 0.0611 (0.404)

=1 if post-secondary -0.4946*** (0.004) -0.4758*** (0.003)

past year hungry season maize price (kwacha/kg)2 -0.0010 (0.884) -0.0012 (0.840)

past year harvest season maize price (kwacha/kg)2 -0.0034 (0.660) -0.0021 (0.753)

price of NPK & Urea fertilizer (kwacha/kg)2 0.0007 (0.152) 0.0005 (0.304)

agricultural wage rate (kwacha/day)2 0.0001* (0.082) 0.0001* (0.051)

average rainfall, past five growing seasons (cm) 0.0029 (0.106) 0.0029* (0.063)

coefficient of variation on past rainfall -1.6175 (0.251) -1.3689 (0.273)

Observations 462 462 924 924

R-squared

Coefficients in columns 3 and 4 are Average Partial Effects (APE) estimated via the margins command in Stata; all models include year and region dummy variables; standard errors clustered at the household level; FD = First Difference, LPM = linear probability model, MC = Mundlak-Chamberlain. * Statistically significant at the 10% level. ** Statistically significant at the 5% level. *** Statistically significant at the 1% level.

a Corresponding coefficient is time constant and does not vary over time.

b Variable is converted to real 2011 kwacha. US $1.00 = 151.55 kwacha in 2010/11 (Chirwa and Dorward, 2013). c Model includes time-averages of all time-varying covariates; R-squared is correlation-squared in columns 3 and 4.

crowd in or crowd out commercial chemical use. The results across model specifications indicate that acquiring subsidized storage chemicals has a statistically significant (p-value < 0.10) and economically meaningful association with crowding out of commercial storage chemicals. Parsimonious model results are similar to full model results. When the models are estimated linearly using FD LPM columns 1 and 2, access to subsidized storage chemicals is associated with a reduction in the probability that a household will purchase commercial storage chemicals by between 50.56 and 51.81 percentage points on average. When nonlinearities are considered using a probit estimator with the MC device in columns 3 and 4, use of subsidized storage chemicals is associated with a reduction in the probability that a household will purchase commercial storage chemicals by between 58.57 to 63.41 percentage points on average. If we consider 50.56 percentage points to be the lower bound estimate, then our results suggest that the storage chemical subsidy only helps raises the probability that a household will use storage chemicals by about 49.50 percentage points.

The results in columns 2 and 4 of Table 6 show that households where the head has some schooling are significantly more likely to purchase commercial storage chemicals on average, than are households where the head has never been to school. Column 4 also indicates that on average households with a higher value of livestock and durable assets are significantly more likely to purchase commercial storage chemicals on average.

The results section in this article concludes with a discussion of how the findings of storage chemicals' impact on improved maize

adoption can affect household income. In order to do so we use parameter estimates from Bezu et al. (2014) that use the IHS2, AISS1, and AISS2 data sets in Malawi. The findings in Bezu et al. indicate that a 1% increase in area under improved maize cultivation increases average household income by 0.261% per adult equivalent, with statistical significance at the 1% level. The study also finds that increases in improved maize adoption have progressive distributional impacts, as a 1% increase in improved maize area raises household income by 0.296% per adult equivalent on average for the poorest third of the sample, while the effect is not statistically significant for the richest third of the sample.

When we put the estimates from Bezu et al. in the context of our study, we find that when the results from column 4 of Table 4 in the present study are converted into elasticity form, a 1% increase in storage chemical is associated with an increase in area planted to improved maize of 0.1123%. When 0.1123 is multiplied by 0.261 from Bezu et al. we find that a one percent increase in storage chemical adoption is associated with an increase in household income of 0.029% per adult equivalent on average. Based on the estimates from Bezu et al., the effect is larger for the poorest third of households, as a 1% increase in storage chemical use is associated with an increased household income of 0.033% per adult equivalent on average for that sub-sample of the population. These results are not huge but are fairly meaningful for smallholder household income in Malawi. It is also important to note that since the crowding out rate of the storage chemical subsidy program is around 50%, then a 1% increase in subsidized storage chemical

Table 6

Factors affecting crowding out of commercial storage chemicals.

(1) FDLPM Parsimonious

(2) FD LPM Full

VARIABLES

Coeff.

P-value

Coeff.

P-value

Probit with MC

device3 Parsimonious Coeff. P-value

Probit with MC device3 Full

Coeff. P-value

=1 If HH used subsidized storage chemicals after previous harvest -0.5056*** (0.000) -0.5181*** (0.000) -0.6341*** (0.000) -0.5857*** (0.000)

binary=l; if farm credit organization in village 0.0371 (0.455) 0.0324 (0.508)

distance to paved road (km)1 -0.0006 (0.590)

distance to main market (km)1 -0.0005 (0.473)

distance to extension services (km) 0.0017 (0.567) 0.0011 (0.723)

number of dealers who sell subsidized inputs in village -0.0022 (0.933) -0.0029 (0.915)

log value of household assets2 0.0126 (0.492) 0.0296* (0.071)

landholding (in ha) 0.0042 (0.780) 0.0059 (0.763)

age of household head in first survey year1 -0.0009 (0.377)

=1 If female headed household 0.0388 (0.622) 0.0198 (0.794)

log of adult equivalents 0.0809 (0.246) 0.0745 (0.271)

=1 If primary (grades 1 to 4) 0.1576** (0.020) 0.1341** (0.033)

=1 if upper primary (grades 5 to 8) 0.1548* (0.062) 0.1389* (0.065)

=1 If secondary (grades 8 to 12) 0.1007 (0.350) 0.0747 (0.463)

=1 if post-secondary 0.0509 (0.660) 0.0427 (0.742)

average rainfall, past five growing seasons (cm) -0.0007 (0.607) -0.0006 (0.666)

coefficient of variation on past rainfall -1.4609 (0.334) -0.9955 (0.501)

Observations 462 462 924 924

R-squared

Coefficients in columns 3 and 4 are Average Partial Effects (APE) estimated via the margins command in Stata; all models include year and region dummy variables; standard errors clustered at the household level; FD = First Difference, LPM = linear probability model, MC = Mundlak-Chamberlain. * Statistically significant at the 10% level. ** Statistically significant at the 5% level. *** Statistically significant at the 1% level.

a Corresponding coefficient is time constant and does not vary over time.

b Variable is converted to real 2011 kwacha. US $1.00 = 151.55 kwacha in 2010/11 (Chirwa and Dorward, 2013). c Model includes time-averages of all time-varying covariates; R-squared is correlation-squared in columns 3 and 4.

use can be said to help increase household income by only 0.0145% per adult equivalent for all households, and by 0.0165% per adult equivalent for the poorest third of households. As is the case with all types of input subsidy programs, minimizing crowding out of subsidized storage chemicals can help increase total chemical use among smallholders.

Conclusions & policy implications

To date, the relationship between accessing post-harvest technologies and adoption of improved maize varieties in Africa is poorly understood. Using data from Malawi, this article estimates how use of storage chemicals affects a farmer's decision to adopt improved maize varieties, that while being higher yielding, are more susceptible to pest damage during storage than are traditional maize varieties. The article also estimates the extent to which acquiring subsidized storage chemicals crowds out commercial storage chemical acquisition. The implications of this article are important as food security does not end at harvest. With destructive pests like the larger grain borer changing the face of post-harvest grain management in many regions of sub-Saharan Africa, we provide evidence in this article that the consequences even extend to farmers' planting decisions.

The key findings from this article indicate that acquiring storage chemicals after the previous harvest is associated with a statistically significant increase in the probability that the average household will plant improved maize varieties, plant a larger area to improved maize varieties, and increase the share of total area that is planted to improved maize varieties. The magnitude of the effects are not large, but are meaningful for smallholders.

These results are what we might expect, given that Malawian farmers acknowledge available improved maize varieties to be more susceptible to insect pests than local varieties. Therefore, increased access to storage chemicals may increase households' willingness to adopt these higher yielding varieties. Increased adoption of improved maize varieties can have important economic effects for smallholders, because these varieties have the potential to increase yields, and thus household income and food security. Combining our results with findings in Bezu et al. (2014) indicate that an increase in storage chemical use is associated with a small positive effect on household income through its influence on adoption of improved maize varieties. Additional benefits may be also realized by smallholders by maintaining higher quality grain in storage, to sell later at higher prices.

In addition, this study finds that the storage chemical subsidy is significantly associated with crowding out commercial storage chemicals ceteris paribus. These effects are consistent with other studies in both Malawi and Zambia that measure crowding out of commercial fertilizer by subsidized fertilizer (Xu et al. 2009; Ricker-Gilbert et al., 2011; Mason andJayne, 2013), and crowding out of commercial seed by subsidized seed (Mason and Ricker-Gilbert, 2013). The results from our study indicate that the lower bound crowding out estimate for the subsidy is a 50.56 percentage point reduction in the probability that a household buys storage chemicals commercially. This translates into the storage chemical subsidy only raising the probability that a household will use storage chemicals by about 49.50 percentage points on average.

Ultimately, results from this article demonstrate that policies and programs that facilitate access to storage inputs, chemical or otherwise, can advance the adoption of improved maize varieties

that can enhance staple crop production and food security goals for smallholder producers. Failure to account for the production and post-harvest biological constraints which farmers face may result in sub-optimal input use among smallholders. This can undermine the effectiveness of input subsidy programs that seek to promote improved seed adoption by subsidizing seed and inorganic fertilizer. While subsidies for storage chemicals can increase adoption of the input and area planted to improved varieties, they need to be targeted to households who are unable to purchase them on the commercial market, in order to reduce crowding out of commercial inputs. In addition, the supply chain for storage chemical can be strengthened by distributing subsidized chemicals through private agro-dealers, just like the seed subsidy in Malawi, rather than through extension offices.

The goal of this article is to show that there is a clear relationship between access to storage technologies and adoption of improved maize varieties. However, results from this article should not necessarily be used to advocate for increasing storage chemical use among smallholders. Storage chemicals may have other health risks that have not been addressed here, and there are alternative post-harvest technologies such as hermetic (air tight) storage containers that can potentially reduce pest risk in a chemical-free environment. If the government wants to promote storage chemicals to protect against insect damage in the post-harvest season, farmers need to be trained on how to use these chemicals appropriately. While the evidence presented in this study applies to Mal-awian farmers, this relationship is relevant for smallholder farmers in many regions who face destructive storage pests like the larger grain borer. Our result also show that researchers, extension staff, and policy makers should consider post-harvest issues when promoting adoption of improved varieties.

Acknowledgements

The author gratefully acknowledges funding support for the data collection and analysis in Malawi from the Department for International Development (DF1D)/Malawi, the United States Agency for International Development Malawi Mission (USA1D/ Malawi), and from USAlD's Economic Growth and Trade Division (EGAT) to the Food Security Ill Cooperative Agreement at Michigan State University. The authors also gratefully acknowledge funding support from the Bill and Melinda Gates Foundation (BMGF) under the Purdue Improved Crop Storage (PICS 11) Project, and the Guiding Investments in Sustainable Agricultural Intensification in Africa (G1SA1A) project. The authors also wish to thank Andrew Dorward and Ephraim Chirwa for making the A1SS4 data available, along with Corinne Alexander, Megan Sheahan, and Rodney Lunduka for their helpful comments on the manuscript. The authors take responsibility for all remaining errors.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foodpol.2014.10. 015.

References

Adda, C., Borgemeister, C., Biliwa, A., Meikle, W.G., Markham, R.H., Poehling, H.M., 2002.1ntegrated pest management in post-harvest maize: a case study from the Republic of Togo (West Africa). Agric. Ecosyst. Environ. 93 (1), 305-321. Addo, S., Birkinshaw, L.A., Hodges, R.J., 2002. Ten years after the arrival in Ghana of Larger Grain Borer: Farmers' responses and adoption of 1PM strategies. 1nt. J. Pest Manage. 48 (4), 315-325. Barrett, C.B., Carter, M.R., 2010. The power and pitfalls of experiments in development economics: some non-random reflections. Appl. Econ. Perspect. Policy 32 (4), 515-548.

Bezu, S., Tesfahun, G., Shiferaw, B., Ricker-Gilbert, J., 2014. Impact of improved maize adoption on welfare of farm households in Malawi: a panel data analysis. World Development 59, 120-131.

Boxall, R.A., 2002. Damage and loss caused by the Larger Grain Borer Prostephanus truncatus. Integr. Pest Manag. Rev. 7 (2), 105-121.

Bulte, E., Beekman, G., Di Falco, S., Lei,, P., Hella, J., 2014. Behavioral responses and the impact of new agricultural technologies: evidence from a double-blind field experiment in Tanzania. Am. J. Agric. Econ. 96 (3), 813-830.

Chamberlain, G., 1984. Panel data. In: Grilliches, Z., Intriligator, M.D. (Eds.), Handbook of Econometrics, vol. 2. North-Holland, Amsterdam.

Chapoto, A.C., Jayne, T.S., 2010. Maize price instability in eastern and southern Africa: The impact of trade barriers and market interventions. Prepared for the COMESA policy seminar on "Variation in staple food prices: Causes, consequences, and policy options'', Maputo, Mozambique, 25-26 January, 2010.

Chavas, J.P., Di Falco, S., 2012. On the role of risk versus economies of scope in farm diversification with an application to Ethiopian farms. J. Agric. Econ. 63 (1), 2555.

Chibwana, C., Fisher, M., Shively, G., 2012. Cropland allocation effects of input subsidy programs in Malawi. World Dev. 40 (1), 124-133.

Chirwa, E., Dorward, A., 2013. Agricultural Input Subsidies: The Recent Malawi Experience. Oxford University Press.

Climate Research Unit (CRU), University of East Anglia, 2011. CRU-TS 3.1 Climate Database. CRU Time Series (TS) high Resolution Gridded Datasets, NCAS British Atmospheric Data Centre. <http://badc.nerc.ac.uk/view/

badc.nerc.ac.uk__ATOM_dataent_1256223773328276>.

Dorward, A., Chirwa, E., 2011. The Malawi agricultural inputs subsidy programme, 2005/06 to 2008/09. Int. J. Agric. Sustain. 9 (1), 232-247.

Duflo, E., Glennerster, R., Kremer, M., 2007. Using randomization in development economics research: a Toolkit. Handbook Dev. Econ. 4, 3895-3962.

Dugger, C.W., 2007. Ending famine, simply by ignoring the experts. New York Times 2 (12).

Ersado, L., Amacher, G., Alwang, J., 2004. Productivity and land enhancing technologies in Northern Ethiopia: health, public investments, and sequential adoption. Am. J. Agric. Econ. 86 (2), 321-331.

Golob, P., 2002. Chemical, physical and cultural control of Prostephanus truncatus. Integr. Pest Manag. Rev. 7 (4), 245-277.

Golob, P., 2009. On-farm post-harvest management of food grains: a manual for extension workers with special reference to Africa. In: Boxall, R., Gallat, S. (Eds.), Food and Agriculture Organization of the United Nations, Rome, Italy, 2009.

Government of Malawi, Various Years. Ministry of Agriculture market Database for Agricultural Products.

Gyasi, K.O., Abatania, L.N., Paulinus, T., Abdulai, M.S., Langyintuo, A.S., 2003. A study on the adoption of improved maize technologies in Northern Ghana. In: Badu-Apraku, B., Fakorede, M.A.B., Ouedraogo, M., Carsky, R.J., Menkir, A. (Eds.), Maize revolution in West and Central Africa. Proceedings of a Regional Maize Workshop, IITA-Cotonou, Benin Republic, 14-May, 2001. WECAMAN/IITA.

Jones, M., 2012. Measuring the Value of African Smallholder Grain Protection: Two Essays on Storage Economics and Market Valuation of Maize Attributes in Malawi. Unpublished: MS Thesis in Agricultural Economics, Purdue University, West Lafayette, IN, USA.

Katengeza, S.P., Mangisoni, J., Kassie, G., Sutcliffe, C., Langyintuo, A., La Rovere, R., Mwangi, W., 2012. Drivers of improved maize variety adoption in drought prone areas of Malawi. J. Dev. Agric. Econ. 4 (14), 393-403.

Langyintuo, A.S., Mungoma, C., 2008. The effect of household wealth on the adoption of improved maize varieties in Zambia. Food Policy 33 (6), 550-559.

Leathers, H.D., Smale, M., 1991. A Bayesian approach to explaining sequential adoption of components of a technological package. Am. J. Agric. Econ. 73 (3), 734-742.

Lunduka, R., Fisher, M., Snapp, S., 2012. Could farmer interest in diversity of seed attributes explain adoption plateaus for modern varieties in Malawi. Food Policy 37 (5), 504-510.

Mason, N.M., Jayne, T., 2013. Fertilizer subsidies and smallholder commercial fertilizer purchases: Crowding out, leakage, and policy implications in Zambia. J. Agric. Econ. 64 (3), 558-582.

Mason, N.M., Ricker-Gilbert, J., 2013. Disrupting demand for commercial seed: Input subsidies in Malawi and Zambia. World Dev. 45, 75-91.

Mason, N.M., Smale, M., 2013. Impacts of subsidized hybrid seed on indicators of economic well-being among smallholder maize growers in Zambia. Agric. Econ. 44 (6), 659-670.

Mitchell, T.D., Jones, P.D., 2005. An improved method of constructing a database of monthly climate observations and associated high-resolution grids. Int. J. Climatol. 25 (6), 693-712.

Mundlak, Y., 1978. On the pooling of time series and cross section data. Econometrica 46 (1), 69-85.

Ravallion, M., 2009. Should the randomistas rule? Econ. Voice 6 (2), 1-5.

Ricker-Gilbert, J., Jayne, T.S., Chirwa, E., 2011. Subsidies and crowding out: a double-hurdle model of fertilizer demand in Malawi. Am. J. Agric. Econ. 93 (1), 26-42.

Singano, C.D., Nkhata, B.T., Mhango, V., 2008. National Annual Report on Larger Grain Borer Monitoring and Teretrius Nigrescens Rearing and Releases in Malawi. Chitedze Agricultural Research Station Report, P.O. Box 158, (Lilongwe, Malawi).

Smale, M., 1995. Maize is life: Malawi's delayed green revolution. World Dev. 23 (5), 819-831.

Smale, M., Heisey, P., Leathers, H., 1995. Maize of the ancestors and modern varieties: the micro-economics of high-yielding variety adoption in Malawi. Econ. Dev. Cult. Change 43 (2), 351-368.

Suri, T., 2011. Selection and comparative advantage in technology adoption.

Econometrica 79 (1), 159-209. Wadonda Consulting, 2011. Background Document on the 2010/11 Agricultural

Input Subsidy Survey 4 (AISS4). Zomba, Malawi. Wooldridge, J.M., 2010. Econometric Analysis of Cross Section and Panel Data, second ed. MIT Press, Cambridge, MA.

World Bank, 2011. Missing Food: The Case of Post-Harvest Grain Losses in Sub-

Saharan Africa. Report 60371-AFR. (WB, Washington, DC). Xu, Z., Burke, W.J., Jayne, T.S., Govereh, J., 2009. Do input subsidy programs "crowd in'' or "crowd out'' commercial market development? Modeling fertilizer use decisions in a two-channel marketing system. Agric. Econ. 40 (1), 79-94.