You are on page 1of 12

Food Policy 43 (2013) 44–55

Contents lists available at ScienceDirect

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

Impact of metal silos on households’ maize storage, storage losses


and food security: An application of a propensity score matching
Zachary M. Gitonga ⇑, Hugo De Groote, Menale Kassie, Tadele Tefera
International Maize and Wheat Improvement Center (CIMMYT), PO Box 1041-00621, Nairobi, Kenya

a r t i c l e i n f o a b s t r a c t

Article history: Maize is the most important food staple in Eastern and Southern Africa, with a highly seasonal production
Received 6 August 2012 but relatively constant consumption over the year. Farmers have to store maize to bridge seasons, for
Received in revised form 19 April 2013 food security and to protect against price fluctuations. However, the traditional storage methods do
Accepted 5 August 2013
not protect grain well, resulting in large postharvest losses. Hermetically sealed metal silos kill storage
pests by oxygen deprivation without pesticides. Popular in Central America, they are now being pro-
moted in Africa, but their impact here has not yet been studied. This study used propensity score match-
Keywords:
ing to evaluate the impact of metal silos on duration of maize storage, loss abatement, cost of storage, and
Propensity score matching
Metal silo technology
household food security. Metal silo adopters (N = 116) were matched with non-adopting farmers from a
Food security representative sample of 1340 households covering the major maize-growing zones in Kenya. The major
Storage duration effect of the metal silos was an almost complete elimination of losses due to insect pests, saving farmers
Storage cost an average of 150–200 kg of grain, worth KSh9750 (US$130). Metal silo adopters also spent about KSh340
Loss abatement less on storage insecticides. Adopters were able to store their maize for 1.8–2.4 months longer, and to sell
their surplus after five months at good prices, instead of having to sell right after the harvest. The period
of inadequate food provision among adopters was reduced by more than one month. We conclude that
metal silos are effective in reducing grain losses due to maize-storage insects, and that they have a large
impact on the welfare and food security of farm households. The initial cost of metal silos is high
(KSh20,000/1.8 ton) and therefore policies to increase access to credit, to reduce the cost of sheet metal,
and to promote collective action can improve their uptake by smallholder farmers.
Ó 2013 Elsevier Ltd. All rights reserved.

Introduction Causes of postharvest losses are diverse, and include storage-in-


sect pests, postharvest handling, inadequate grain-storage prac-
Maize is one of the major food staples in sub-Saharan Africa tices and absence of storage-management technologies. Reducing
(Mellor et al., 1987), with 57 million tons produced on about 30 PHL could increase food availability by both reducing physical
million ha in the year 2010 (FAOSTAT, 2010; Nweke, 2005). Most losses and increasing income from the improved market opportu-
of it is grown for human consumption (Pingali, 2001), and provides nities that could be used to buy food (FAO and World Bank, 2010).
food and income for more than 300 million smallholder farmers Farmers in developing countries experience postharvest losses
(La Rovere et al., 2010). In Kenya it is the most important food sta- which are highly variable (5–30%) depending on weather condi-
ple crop for over 80% of the population, contributing 65% of total tions at harvesting, storage length and presence of pests (Rembold
staple-food caloric intake and having an annual per capita con- et al., 2011). Postharvest grain losses are estimated to be worth
sumption of 77 kg of maize products (FAO, 2011). US$1.6 billion per year in Eastern and Southern Africa alone (FAO
Postharvest losses (PHL) denote a measurable decrease of food and World Bank, 2010). PHL include losses from harvesting and
grain across the postharvest system which may be quantitative, drying, threshing and selling, transport, on-farm storage, and mar-
qualitative and economic (De Lucia and Assennato, 1994). Food keting. The total PHL in Africa from maize are estimated to range
loss is a subset of PHL and represents the part of the edible share from 14% to 36%, and the on-farm storage losses from 4% to 10%
of food that is available for consumption at either the retail or (Tefera, 2012). In another analysis, based on an extensive review
the consumer levels, but is not consumed for whatever reason of the literature, the storage weight losses for maize in Eastern
(Hodges et al., 2010). and Southern Africa for 2007 was estimated at 17.5%, valued at
$920 million (Rembold et al., 2011). In Kenya, only one rigorous
study is available, from prior to the arrival of the larger grain borer
⇑ Corresponding author. Tel.: +254 2 7224600; fax: +254 2 7224601. (LBG), and this estimated the loss in maize storage at 5% (de Lima,
E-mail address: z.gitonga@cgiar.org (Z.M. Gitonga). 1979).

0306-9192/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.foodpol.2013.08.005
Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55 45

Food security exists when all people have, at all times, access to Maize and Wheat Improvement Center, (CIMMYT) trained a new
sufficient, safe and nutritious food to meet their dietary needs and group of artisans in Kenya and Malawi during the pilot phase of
food preferences in order to lead an active and healthy life the project. The trained artisans make metal silos of different
(Pinstrup-Andersen, 2009). In Eastern and Southern Africa, maize capacities from galvanized metal sheet depending on the need of
is the most important food crop, and its consumption as a basic the farmer, to protect maize and other grains against pest attack.
staple is relatively constant over the year. Unfortunately, maize They are airtight and kill any pest in the grain through oxygen
production is highly seasonal, resulting in strongly fluctuating sup- depletion by the use of a burning candle.
plies. Grain storage is therefore essential to food security by bridg- By 2011, over 40 artisans were trained in Kenya but only four
ing the period between two harvests, as well as by stabilizing have been active, partly due to lack of effective demand and aware-
prices by taking the produce off the market during the peak season, ness among farmers. The active ones usually subcontract other
and releasing it back when the grain is in short supply (Proctor, trained but less active artisans when orders come in. These orders
1994). Good storage methods therefore become an important as- mainly come from institutions, for example schools, who buy the
pect of household food security and rural livelihoods since they en- larger silos, and NGOs who buy smaller silos and supply them to
sure a continuous stable supply of food and better farm incomes farmers outside the project areas. The smaller silos, up to 540 kg,
(Tefera, 2012; Thamaga-Chitja et al., 2004). Although farmers’ pri- can be moved, although particular care needs to be taken not to
mary objective in storing grain is household food security, some break the seals. Larger silos are built inside the room where they
store it for market speculation purposes, and others use it as seed will be kept because the size does not allow them to fit through
in the following season (Proctor, 1994). However, farmers often a standard door of 75 cm.
still sell their produce immediately after harvest at low prices. Metal silos were (for the first time) formally tested and found
When their stocks run out, they have to buy food later in the mar- effective in Kenya (De Groote et al., 2013). On-station trials in dif-
ket at high prices. Apart from on-farm storage, most governments ferent maize agro-ecological zones showed that weevils or larger
in sub-Saharan Africa also engage in establishing strategic food re- grain borers had not damaged grain stored in metal silos after six
serves, to provide for market failures, to stabilize food prices, and months in storage, even without insecticides, while for maize
to prepare for emergencies. However, the reduction of losses be- stored in polypropylene bags with insecticides (Actellic Super)
fore and after harvest at the farm level is probably more efficient good control was only provided for four months. Although metal
but has not received as much attention (Tefera et al., 2011). silos are technically effective against the main grain-storage pests,
While many smallholder farmers in developing countries still their distribution and adoption by smallholder farmers remain lim-
rely on traditional storage methods, storing unshelled maize in tra- ited. The cost of a metal silo with a capacity of holding half a ton is
ditional cribs has proved difficult with the advent of the LGB and about KSh15,000 (US$200), while the cost for a 2.7-ton silo is about
the increase of theft (Boxall, 1997). Traditional storage structures KSh35,000 (US$ 460). A survey of metal silo artisans in 2011
expose grain to rodents and insect attack, and provide favorable showed that over 2300 silos have been fabricated and distributed
climatic conditions for their proliferation, as well as for that of in Kenya. Half of those fabricated have a capacity of 1800 kg each,
microorganisms, thus leading to substantial post-harvest losses and 15% having a capacity of 540 kg each. The relatively high cost
(Ngamo et al., 2007). In Kenya, farmers more commonly store the of the metal silos and lack of awareness dampen the demand and
shelled grain in polypropylene bags. Many smallholder farmers sell widespread adoption by farmers.
off their grain immediately after harvest to avoid damage by stor- Communication media such as television, workshops and dem-
age pests and consequently receive low prices (Boxall, 1998; onstration of the silos in different rural villages are among strate-
Proctor, 1994). The LGB is an invasive storage pest accidentally gies being used by EGSP project partners to promote the
introduced into Africa from Central America, and now recognized technology. So far, credit has not been provided although options
as the most destructive maize pest (Boxall, 2002). The most com- exist. Equity Bank, for example, does provide loans for agricultural
mon traditional maize-storage facilities used by smallholder farm- inputs and equipment, called ‘‘Kilimo biashara’’, where smallholder
ers in Africa include traditional cribs or granaries, baskets, and farmers can borrow between KSh1000 and KSh100,000 over a per-
large pots. iod of 12 months with an annual interest rate of 10%. The EGSP in
In response to challenges posed by the maize weevil and LGB, its second phase is upscaling the fabrication and uptake of metal
improved storage technologies have been developed, including silos in Kenya and Malawi while introducing the new storage tech-
actellic super (a mixture of pirimiphos-methyl and permethrin), nology to Zambia and Zimbabwe.
super grain bags, and the metal silo (De Groote et al., 2013). Metal So far, few impact assessment studies have been conducted on
silos provide a strong barrier against insect pests and rodents and metal silos and most have used descriptive analysis only. In Latin
are hermetically sealed, killing any remaining insects through oxy- America, farmers who acquired metal silos were able to store their
gen depletion (Tefera et al., 2011). They are durable, allow for long maize for up to two years and delay selling it at low prices, hence
storage periods, and are therefore an important storage technology increasing their income (Fischler et al., 2011). Subsistence farmers
in the fight against hunger and food insecurity in developing coun- kept much of their harvested maize for their own consumption and
tries. A metal silo with a capacity of 990 kg can conserve enough hence were able to increase their food security by 30–35 days per
grain to feed a household of five members for one year (FAO, 2008). year. The postharvest loss of stored maize among the users of me-
Metal silos were first developed and disseminated in Central tal silos was negligible (Fischler et al., 2011). Over 500 artisans who
America through the Postcosecha (meaning ‘‘postharvest’’) Pro- received training on metal silo fabrication and entrepreneurship
gram in 1983 by the Swiss Development Cooperation (SDC) also benefitted from the sale of silos to farmers (Hellin and
(Bokusheva et al., 2012; Fischler et al., 2011). Farmers used phos- Kanampiu, 2008). A study in El Salvador, Guatemala, Honduras
toxin pellets to kill the storage pests and the silos were made air- and Nicaragua showed that households who used metal silos
tight to prevent leaking of the fumigant (Fischler et al., 2011). bought maize for fewer months and experienced a significant
In Kenya the metal silos were introduced by the Catholic Relief improvement in their food security and well-being than non-
Services (CRS) in 2000, mainly in the western region (Nguyo, 2007). adopters (Bokusheva et al., 2012).
Following the success story in Central America, an SDC-funded In Africa, the Food and Agriculture Organization of the United
Effective Grain Storage Project (EGSP) was launched in Eastern Nations (FAO) has validated and promoted the household metal
and Southern Africa in 2008 to promote the fabrication and dis- silo in twelve countries (Lorío and Njie, 2011). FAO has also pro-
semination of metal silos to smallholder farmers. The International moted the silos in four countries in Asia, as well as disseminating
46 Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55

silos in eight countries in Latin America. In Kenya, an ex-ante im- Researchers have proposed various methods to avoid selection
pact assessment study was conducted using benefit cost analysis bias. One way of reducing selection bias is random assignment of
(Kimenju and De Groote, 2010). Potential benefits were calculated treatment (de Janvry et al., 2011), but this is not possible for ex-
based on the on-station trials which showed high losses in un- post studies. The use of panel estimators with household fixed ef-
treated polypropylene bags, but negligible losses in the metal silos fects is another option (Wooldridge, 2002), but panel data are
(De Groote et al., 2013), and the potential benefits were compared rarely available for impact evaluation, as in our case. The Heckman
to the costs, based on data from artisans. The results showed that correction (Heckman et al., 1997) and the instrumental variable
the benefits from the silos increased with their size, and that the (IV) approaches (Khandker et al., 2010) are also alternative ways
larger metal silos were profitable if losses and maize prices were to control for selection bias stemming from unobserved heteroge-
sufficiently high. So far, no ex-post impact studies of metal silos neity. These approaches, nonetheless, impose distributional and
in Africa have been conducted. functional form assumptions and require a valid instrument that
This study therefore evaluated the ex-post impact of metal silos determines the treatment status but not the outcome variable,
on maize storage and food security at the household level. Specif- which is a challenge in many empirical studies (Jalan and
ically, we determined the impact of metal silos on the cost of stor- Ravallion, 2003). In addition, the standard IV approach assumes
age, loss abatement, duration of storage and food security. The that the treatment variable has only a parallel shift effect (intercept
study used propensity score matching to account for the possible effect) with respect to the outcome variable, ignoring the interac-
differences in wealth and other socioeconomic characteristics be- tions between technology adoption and other covariates (i.e. ignor-
tween adopters and non-adopters of metal silos. To our knowledge, ing slope effect).
this is the first study that has used the propensity score matching Another econometric method of dealing with the problem of
method to study the impact of storage technologies. selection bias with cross-sectional data in a non-experimental
framework without the assumptions of functional and distribu-
tional form or exogeneity of covariates, is the propensity score
Conceptual framework
matching (PSM) method, that has recently gained popularity
(Diagne and Demont, 2007; Heckman et al., 1997; Jalan and
Adoption model
Ravallion, 2001).

The decision of whether or not to adopt a new storage technol-


ogy such as metal silos depends on the utility the farmer expects to Propensity score matching
derive from the innovation. Farmers only adopt technologies if the
expected utility of adoption (Ua) is greater than non-adoption (Un), PSM is based on the assumption that sample selection bias can
i.e. Ua  Un > 0 (Kassie et al., 2011). be eliminated by conditioning on observable variables, and does so
Random utility models presume that the utility Ua derived by by matching each adopting household with one or more non-
individual households from using the metal silo technology is com- adopting households with similar observable characteristics. In
posed of a deterministic component which can be calculated based essence, matching models simulate the conditions of an experi-
on observed characteristics zi and a stochastic error component e, ment in which adopters and non-adopters are randomly assigned,
which is unobserved, such that allowing for the identification of a causal link between technology
choice and outcome variables. One of the weaknesses of the PSM
T i ¼ bzi þ e1 ; T i ¼ 1 if T i > 0 ð1Þ method is that it does not capture selection bias based on unob-
served heterogeneity. However, Rosenbaum bounds sensitivity
where Ti is a binary indicator variable that takes a value of 1 if analysis can check if the PSM results are sensitive to hidden bias
household i adopts the metal silo and 0 if otherwise, b is a vector (Becker and Andrea, 2002) .
of parameters to be estimated, z is a vector of explanatory variables In the last few years, PSM has been used in several impact stud-
and e is the error term. The error component ei is never observed, ies of agricultural technologies. These include the impact of the
hence we do not have enough information to predict an individual’s adoption of genetically modified cotton on poverty in Pakistan
choice, but we can predict patterns of households’ adoption of the (M’mboyi et al., 2010); the impact of market information access
metal silo from among other alternative storage technologies. on prices received by smallholder farmers in Uganda (Kiiza et al.,
The conditional probability of metal silo adoption by a house- 2010); the impact of improved maize varieties on poverty in Mex-
hold based on the observable characteristics can then be estimated ico (Becerril and Abdulai, 2009); and the impact of the adoption of
using either binary probit or logit models: improved groundnut varieties on rural poverty in Uganda (Kassie
PrðT i ¼ 1Þ ¼ prðT i > 0Þ ¼ 1  Fðbzi Þ ð2Þ et al., 2011). The method was also used to evaluate the welfare im-
pact of pigeonpea-maize intensification in Tanzania (Amare et al.,
where F is the cumulative distribution function for ei, which is as- 2012); the adoption of improved pigeonpea and chickpea varieties
sumed to have a normal distribution for the probit model, or logistic in Ethiopia and Tanzania (Asfaw et al., 2012); and rural poverty in
distribution for the logit model. Bangladesh (Mendola, 2007). This method, however, has not yet
been applied to study the impact of grain storage technology.

Impact evaluation and selection bias


Methodology
The empirical challenge in impact assessment is creating a
counterfactual while addressing selection bias. Directly comparing Estimation strategy of propensity score matching (PSM)
the outcome variables between adopters and non-adopters may be
misleading, because there may also be a systematic difference be- This study used the PSM method to evaluate the impact of me-
tween adopters and non-adopters in terms of observable and tal silo technology on the duration of maize storage, storage losses,
unobservable characteristics. A simple regression model, which food security, and cost of storage, while addressing the economet-
contains a dummy adoption variable as a treatment variable, can- ric issues discussed above. The causal effect of the metal silo on
not control for selection bias stemming from differences in these variables of interest was estimated in two stages: in the first
observed and unobserved characteristics. stage, the propensity scores or the conditional probability of
Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55 47

adoption of the metal silos were estimated using the logit model of Household characteristics considered included household size,
Eq. (2); in the second stage, adopters and non-adopters were the hosting of big events, land cultivated and maize area, expendi-
matched by their propensity scores using three different matching ture and non-agricultural income. Larger households need to store
algorithms: nearest neighbor matching with replacement, kernel more maize to feed all members until the next harvest. Further,
matching and radius caliper matching. In kernel-based matching, large social events like burials or weddings in Africa usually in-
each person in the treatment group is matched to weighted aver- volve the hosting of many people and the consumption of large
ages of individuals who have similar propensity scores, with great- amounts of food. Households that host such events are likely to ex-
er weight being given to people with closer scores. Nearest haust their stored grain reserves sooner.
neighbor matching matches a subject from the control group to a Households with more land are likely to produce more maize
subject in the treatment group, based on the closest propensity and therefore have more need for storage facilities. Land cultivated,
score. Radius matching uses a tolerance level on the maximum off-farm income and expenditure are wealth indicators, and fami-
propensity score distance between a subject in the treatment lies with more wealth and income might have more resources to
group and all individuals in the control group who are within that purchase the silos. Off-farm income opportunities are an important
distance (Chen and Zeiser, 2008). strategy by rural households for overcoming working capital and
The main purpose of propensity score matching is to balance credit constraints in many developing countries. Working capital
the distribution of observed covariates (Lee, 2008), so there should and credit are usually required for investing in improved seeds, in-
be no systematic differences in the distribution and overlap of puts and equipment (Barrett et al., 2001). Off-farm income was
covariates between the two groups after matching. PMS results shown to have a positive and significant impact on adoption of im-
can therefore be tested using different covariates balancing tests proved maize varieties by rural households in Uganda (Diiro,
(Rosenbaum and Rubin, 1985; Sianesi, 2004). In this study, the 2013). Farmers with off-farm income could therefore be more
equality of means of observed characteristics in the treatment likely to purchase metal silos.
and control groups after matching was examined using a two-sam- Institutional factors include having access to a mobile phone ac-
ple t-test: after matching, there should be no significant differ- count or bank account, and access to a passable road. Access to
ences. Further, the matching was tested by comparing the financial services increases the probability of accessing savings or
pseudo R2 and p-values of the likelihood ratio test of the joint insig- receiving credit to finance the adoption of metal silos, and was
nificance of all the regressors obtained from the logit analysis be- approximated by binary variables indicating if the household had
fore and after matching the samples. The pseudo-R2 should be a mobile phone virtual account (M-Pesa, a banking technology pio-
lower and the joint significance of covariates should be rejected neered in Kenya) or a bank savings account. Access to a passable
(or the p-values of the likelihood ratio should be insignificant) road reduces travel costs to the market and to the supplier of the
(Sianesi, 2004). Finally, a propensity score graph was used to check metal silos.
visually if the common support condition was satisfied, i.e. if there Finally, agroecological zones were included in the analysis to
was sufficient overlap. control for different climatic conditions and spatial heterogeneity
Additionally, the balancing property was checked using mean of sampled households. Six maize agroecological zones can be dis-
absolute standardized bias (MASB) between adopters and non- tinguished in Kenya, going from east to west: the low tropics, the
adopters as suggested by Rosenbaum and Rubin (1985), in which dry mid-altitudes, the dry transitional, the moist transitional and
they recommend that a standardized difference of greater than the highlands (Hassan et al., 1998) (Fig. 1). Yields are much higher
20% should be considered too large and an indicator that the in high-potential areas, which include the high and transitional
matching process has failed. The PSM estimation is not robust in zones and the moist zones, than in low-potential areas, which in-
the presence of hidden bias or selection on unobservables. Using clude the drylands and the low tropics. The uni-modal rainfall pat-
the Rosenbaum (2002) bounds test, we checked the sensitivity of tern of the highlands requires that households store more food for
the estimated average adoption effects to hidden bias (unobserved longer periods than the households that live in areas with a bi-
selection). modal rainfall pattern, such as the low, middle and transitional
In estimating the impact of metal silos on the welfare of rural zones.
households, we considered four main outcomes: maize losses in
storage, cost of storage insecticides used, length of storage, and
food security. We postulated that adoption of metal silos would re-
duce postharvest grain losses and the amount of money spent by
households on storage insecticides, while increasing the period
that households are able to store their own grain and thus reducing
food insecurity. Based on previous studies in Latin America
(Bokusheva et al., 2012; Fischler et al., 2011), households adopting
metal silos were expected to keep their maize longer and delay
selling to wait for better prices off-season.
The observable covariates considered, i.e. factors that were
likely to affect the probability of adopting metal silos, were se-
lected based on previous adoption and impact studies and eco-
nomic theory, and included the characteristics of the household
head, characteristics of the household, institutional factors and
agroecological zones. Household head characteristics that could
influence the adoption of the metal silo included age, gender, edu-
cation, maize-farming experience and literacy. In addition, the age
and gender of the primary decision maker, if different from the
household head, might influence the decision to adopt. Literate
and more experienced farmers are more likely to be knowledge-
able on solutions to storage problems and proactive in adopting Fig. 1. Map with selected sub-locations for the adopters and non-adopters of metal
them. silos.
48 Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55

Survey design Data collection

This study used two surveys, a large and representative house- In both surveys, the household head was interviewed with the
hold survey for the control, and a specific metal silo owners’ survey same pretested questionnaire. Characteristics of the household
for the treatment. The first survey was a representative household head included age, gender, literacy, years of formal education
survey of the major maize-growing agroecological zones, con- and years of farming experience. If different from the household
ducted to collect the baseline data for the Aflacontrol project from head, the age and gender of the primary decision maker were also
October 2010 to March 2011 (De Groote et al., 2011). This survey observed. Household characteristics included household size and if
used a stratified two-stage sampling design, with the maize agro- the family had hosted big social events over the last year. Wealth
ecological zones of Kenya (Hassan et al., 1998) as the strata, the indicators included total land cultivated, area planted in maize,
sub-locations (Kenya’s smallest administrative units) as deter- off-farm income and annual expenditure. Respondents were also
mined in the 2009 Kenya national bureau of statistics (KNBS, asked if someone in their household had a mobile phone money ac-
2010), census as the primary sampling units (PSU), and the count or a bank account, and the distance to the closest passable
households as secondary sampling units (SSU). All six maize agro- road (in km).
ecological zones in Kenya were covered: lowland tropics, dry mid- Farmers were asked to estimate their maize production, how
altitude, dry transitional, moist transitional, high tropics and moist much maize they consumed, and how much they sold and when
mid-altitude (Fig. 1). The number of PSU and SSU in each of the they sold it. On storage, they were asked which storage facilities
strata was first determined by optimizing the sample size needed they used for their maize, whether they used insecticides, the cost
to obtain a precision of at least 15% in each stratum, and 8% overall of the insecticides, the amount of maize lost to storage pests over
(calculated for the variable ‘maize area’) (De Groote, 1996). Based the season, and for how many months they stored their maize.
on the results, the required number of PSUs was calculated at Concerning food security, the person responsible for food prep-
120 sub-locations, spread over the different zones. The number aration, usually the wife of the household head, was interviewed.
of households per sub-location was then harmonized to 12, which The respondents were first asked to estimate the number of
was convenient for two teams of six enumerators. Using maps, months in which their household did not have adequate food to
Geographic Information System (GIS) analysis and key informa- feed all the members, to give the months of inadequate household
tion, a list of all rural sub-locations in each of the six zones was food provisioning (MIHFP) score (Bilinsky and Swindale, 2005). The
established, with the number of households in each obtained from second measure of food insecurity was the household food insecu-
the 2009 census. For each zone, the required number of sub-loca- rity access score (HFIAS) (Coates et al., 2007), which reflects the
tions was selected with probability proportionate to size (pps), to respondent’s perception of the food security in the household over
maintain a fixed number of households per sub-location and a the last month and is based on nine questions of increasing food
self-weighted sample (Kalton, 1983). For each sub-location, 12 insecurity from anxiety to being severely food insecure. If they an-
households were selected by random sampling except for the swered any question in the affirmative, they were asked how often
coastal lowlands where six households were selected per sub-loca- (never, rarely, sometimes, and often). From a possible three
tion due to budgetary constraints. The total sample included 1344 positive answers on nine different questions, a score from zero
households, of which four households had adopted metal silos (see (all-negative answers) to 27 can be derived. Based on this score,
Table 1). households can be divided into those who are food secure, and
The second survey, now only of metal silo owners, was conducted those with low, moderate or severe food insecurity.
for the effective grain storage project (EGSP) at the beginning of
2011. Based on information from the project partners and the arti- Results
sans trained by the project in metal silo construction, a list of 124
metal silo owners was compiled , of which 59% were clustered in Households’ social and demographic characterization
Migori and Homa Bay counties (moist transitional and moist
mid-altitude) and 41% in Embu County (moist transitional and dry The four metal silo adopters found among the 1344 households
mid-altitude). All the 124 households were visited and interviewed in the representative survey were combined with the 112 adopters
using the same questionnaire that was administered in the first sur- among the special survey of metal silo owners, leading to a group
vey. Ten households no longer used the silo, either because of insuf- of 116 adopters and 1340 non-adopters.
ficient grain or because of mechanical silo damage, while two Most household heads were male with an average age of
households used it for storing sorghum and beans. The remaining 53 years for both the adopters and the non-adopters of metal silos
112 used it for maize. For the impact analysis, we considered the (Table 2). Literacy and years of formal education of the household
112 users of metal silos for maize storage as the adopters, together head were significantly higher for adopters than for non-adopters,
with the four adopters identified in the first survey. while farming experience was significantly lower. Non-adopters

Table 1
Sampling design for baseline and metal silo adoption survey.

AEZ Baseline survey (2010–2011) n = 1344 Metal silo owners survey (2011) n = 124
Non-adopters Metal silo users Metal silo users Metal silo users (other grains only) Disadopters
Low tropics 90 0 0 0 0
Dry mid-altitude 215 2 20 1 3
Dry transitional 203 0 0 0 0
Moist transitional 354 0 37 0 3
High tropics 240 0 0 0 0
Moist mid-altitude 238 2 55 1 4
Total 1340 4 112 2 10
Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55 49

Table 2
Demographic and social economic characteristics of adopters and non-adopters of metal silo technology.

Group Variable Non-adopters Adopters (N = 116) t-test


(N = 1340)
Mean Std. error Mean Std. error Difference Sig. (2-tailed)
Head Age of the household head (years) 53.4 0.42 53.3 1.06 0.1
Male headed household (%) 81.3 1.07 85.9 3.07 4.7
***
Literacy of the household head (1 if literate; 0 otherwise) 0.8 0.01 0.9 0.02 0.1
***
Household head education (years) 7.1 0.12 10.3 0.38 3.2
**
Farming experience (years) 27.7 0.43 24.6 1.24 3.2
Primary decision maker Primary decision maker is 15–64 yrs male (1 = yes; 0 = No) 0.5 0.01 0.5 0.04 0.0
Primary decision maker is 15–64 yrs female (1 = yes; 0 = No) 0.4 0.01 0.3 0.04 0.1
***
Primary decision maker is >64 yrs male (1 = yes; 0 = No) 0.1 0.01 0.1 0.03 0.1
Primary decision maker is >64 yrs female (1 = yes; 0 = No) 0.1 0.01 0.0 0.01 0.0
***
Household Household size (number of members) 6 0.07 6.9 0.26 0.9
Hosting big social events (1 = yes; 0 = No) 0.2 0.01 0.3 0.05 0.1
***
Total land cultivated (ha) 1.9 0.064 3.1 0.26 1.2
***
Maize area (ha) 1.2 0.04 1.8 0.16 0.6
***
Off-farm income (000’KES) 68.7 6.25 304.6 46.44 235.92
***
Annual expenditure (000’KES) 35.4 1.93 129.9 13.96 94.54
***
Having mobile phone money account (1 = yes; 0 = No) 0.5 0.01 0.8 0.04 0.3
***
Having savings/bank account (1 = yes; 0 = No) 0.7 0.01 1.0 0.02 0.2
***
Distance to the passable road (km) 3.1 0.17 1.5 0.36 1.6
***
Storage chemical users (%) 91.6 0.76 62.5 4.28 29.1
***
Outcome variables Maize storage (months) 4.9 0.09 7.4 0.3 2.5
***
Loss due to storage pests (kg maize) 74.9 6.8 3.4 1.65 71.5
***
Value of loss (Ksh) 3157.5 286.5 142.5 69.75 3015.0
Cost of storage chemicals (Ksh) 531.6 82.3 362.8 57.3 168.8
***
Months of food insecurity 2.3 0.1 0.9 0.2 1.4
***
HFIAS food insecurity scores 6.9 0.2 3.8 0.4 3.1

p < 0.1.
**
p < 0.05.
***
p < 0.01.

had older male decision-makers (over 64 years) but fewer older fe- of Kitale, for example, prices increase from KSh8 per kilogram
male decision-makers than adopters. (US$106/ton), to KSh15/kg (US$200/ton) between seasons. In
The adopting households also differed in many household char- Embu, close to the equator, farmers have two rainy seasons and
acteristics from non-adopting households: the families were larger hence two harvests and two price peaks. Embu is a maize-deficit
(seven members as compared to six); had more land under cultiva- area and hence the prices are higher than in Kitale and Eldoret,
tion (3.3 vs. 1.9 ha) and, especially, grew more maize (1.8 vs. commonly referred to as the grain basket of Kenya. Prices in Embu
1.2 ha). Adopters also had a much higher off-farm income increase from a low of KSh17 to a high of KSh24/kg between sea-
(KSh305,000 vs. KSh69,000), and spent KSh95,900 more each year. sons (Fig. 2). Farmers can increase their earnings from the sale of
They also had higher access to financial services, with more mobile maize by effective storage and by waiting to sell their grain later
phone money accounts (97% vs. 74%) and bank accounts (78% vs. in the season when the prices are more attractive.
46%). Moreover, metal silo adopters were closer to the nearest
passable road (1.5 km vs. 3.1 km). The differences in the means
of the observable characteristics between the adopters and the Maize storage behavior among adopters and non-adopters
non-adopters that could have affected adoption indicated a poten-
tial source of bias, hence the need for matching and selection bias A first comparison between farmers who adopted metal silos
tests. and non-adopters reveals that they differed substantially in their
As expected, the storage behavior and outcomes differed sub- storage-management strategies. While both groups sold maize
stantially between adopters and non-adopters. Almost all non-
adopters (91%) used storage insecticides, compared with only
two thirds (62%) of adopters. Metal silo adopters stored their maize
for two months longer than non-adopters did, and they went with-
out enough food for less than a month as compared to two months
for non-adopters. Adopters also spent less on storage pesticides
(KSh363 vs. KSh532) and experienced much smaller losses (only
3.5 kg vs. 75 kg) than non-adopters. Non-adopters were on average
more food insecure than adopters; they suffered from 2.3 months
of inadequate food provision, compared with 0.9 for the adopters,
and had a mean food insecurity score of 6.9 compared with 3.8 for
the adopters.

Potential benefit and cost of maize storage in metal silos

Analysis of maize prices over the last 20 years reveals that there
are strong seasonal differences as well as differences between
years (Gbegbelegbe and De Groote, 2012). In the surplus market Fig. 2. Average monthly prices of dry maize from 1991 to 2011.
50 Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55

within the first month after harvest (Fig. 3), non-adopters sold In zones where the metal silo project was running, the adopters
most of it in the first month (180 kg per month per household) consistently recorded lower food insecurity scores than the non-
or second month (95 kg) after harvest at low prices, and only small adopters.
amounts in later months. Adopters sold little in the first month The HFIAS followed the same pattern as the MIHFP (Fig. 5). This
(46 kg per household) but gradually increased their sales to a peak index is based on a set of questions indicating increased food inse-
of 150 kg per month in the fifth month, with another peak of al- curity over the last month, from never having to worry about food
most 100 kg at the eight month after harvest when prices were to going a whole day without food, and on this basis, households
much higher. The second peak is probably due to the need to make were classified into four categories, from food secure to severely
space for the new harvest of the second season. Non-adopters were food insecure. The results show that food in security in Kenya’s
probably trying to avoid damage by storage pests as well as meet- maize growing areas is worryingly high. Almost a quarter of the
ing some immediate cash needs, while adopters delayed selling to households are severely food insecure, indicating that in the month
take advantage of improved prices. before the interview they did not have sufficient food. Moreover,
more than a third of households are moderately food insecure,
indicating that they do not eat the quality of food that they would
Maize storage and food security
like, and sometimes cut back on quantity. Food insecurity is high-
est in the drylands, where a third of households are severely inse-
Two measures were used to assess food insecurity: months of
cure, and another third moderately insecure. In the low tropics and
inadequate food provision (MIFP), and the household food insecu-
moist mid-altitudes, a quarter of households are severely food-
rity access score (HFIAS). Food insecurity, as measured by months
insecure. In the high-potential areas, the moist transitional and
of inadequate food provision, was high; on average, households did
the highland tropics, fewer households are affected, but still more
not have sufficient food during two months of the year (Fig. 4).
than 10% are severely food-insecure. Apart from the severely food
Food insecurity is clearly linked to climate: it was highest in the
insecure, in all zones at least a third of households are moderately
low-potential areas, in particular in the dry mid-altitudes
food insecure.
(2.9 months), while households in the other low- to medium-po-
tential areas (the low tropics, the dry transitional and the moist
mid-altitudes), were on average food insecure for 2.5 months. Propensity score matching and test for selection bias
The high-potential areas were clearly more food secure, although
the average was still 1.9 months for the moist transitional and In the first step of the PSM, the logit model of Eq. (2) was esti-
1.6 for the highlands. mated in order to analyze the factors that affect adoption, and to
calculate the propensity to adopt for each household. We used de-
fault nearest neighbor matching method with replacement and
caliper (0.1) to improve the quality of matches. The oversampling
technique radius caliper (0.1) and kernel bandwidth (0.06) match-
ing were also employed for comparison. The results showed that
households with a literate head were more likely to adopt metal si-
los, while farming experience or having a female primary decision
maker between 15 and 64 years old reduced the likelihood of
adoption (Table 3). Household characteristics that increased the
probability of adoption were area of land under cultivation (but
not in maize), off-farm income, expenditure, and access to rural
financial services (owning a mobile phone money account or hav-
ing a bank savings account). Distance to the nearest passable road,
an indicator of transport costs to the farm, reduced the probability
of adoption. Farmers located in the wetter moist transitional and
moist mid-altitude zones were also more likely to adopt metal silos
compared to those located in dry mid-altitude areas.
In the second step of the PSM, adopting and non-adopting farm-
ers were matched on their propensity scores, using three different
Fig. 3. Average maize sales in the months after harvest (kg sold/household).
matching algorithms. The matching procedure was checked to see

Fig. 4. Months of Inadequate Food Provisioning (MIFP). Fig. 5. Household food insecurity access score (HFIAS).
Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55 51

Table 3
Estimation of the logit adoption model on the propensity to adopt.

Dependent variable: silo adoption Coef. Std. err.


Household head Age of the household head (years) 0.01 0.01
Gender of the household head (male) 0.56 0.38
Literacy of the household head (1 if literate; 0 otherwise) 0.91 0.53*
Farming experience (years) 0.01 0.01
Primary decision maker Primary decision maker is 15–64 years female (1 = yes; 0 = No) 0.49 0.28*
Primary decision maker is 15–64 years male (1 = yes; 0 = No) 0.59 0.41
Primary decision maker > 64 years Female (1 = yes; 0 = No) 1.01 0.78
Household Household size (number of members) 0.01 0.04
Hosting big social events (1 = yes; 0 = No) 0.19 0.25
Total cultivated land (ha) 0.29 0.10***
Maize area (ha) 0.18 0.11
Annual expense (000’KES) 0.004 0.001***
Off-farm income (000’KES) 0.001 0.000**
Mobile phone money account (1 = yes; 0 = No) 1.33 0.56**
Savings/bank account (1 = yes; 0 = No) 0.54 0.28**
Distance to a passable road (km) 0.09 0.04***
AEZ = Moist transitional 2.63 0.78***
AEZ = Moist mid-altitude 3.13 0.78***
Constant 7.26 1.22***
Number of obs 892
LR chi2(17) 162.17
Prob > chi2 0.00
Pseudo R2 0.24
Log likelihood 264.74
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.

if it was able to balance the distribution of the relevant covariates all insignificant after matching, indicating that no systematic dif-
in both the treatment and group control, using different methods. ferences remained in the distribution of covariates between the
First, the propensity score test (pstest in Stata), showed a significant adopters and non-adopters of metal silos after matching. The joint
reduction in bias after matching and, most importantly, there were significant effect of the covariates on adoption, as expressed by the
no significant differences in matched non-adopters and adopters significant v2, could not be rejected before matching but was re-
for any of the covariates (Table 4). jected after matching in all the three matching techniques. The
Further, the pseudo R2 of the estimated logit model, which indi- mean and median bias, finally, are all below 20% as required, and
cates how well the covariates explain the probability of adopting are all even below 10%, indicating a very good match.
metal silos, was high before matching, and low afterwards Finally, visual inspection of the distributions of the propensity
(Table 5). Similarly, the p-values of the likelihood ratio test were scores for the metal silo adopters and non-adopters after matching

Table 4
Tests for selection bias after matching.

Variable Matched samplea Bias t-test


Treated Control % Bias % Bias p-Value
(N = 112) (N = 780) reduction
Head Age of the household head (years) 53.3 54.6 9 4516 0.39
Gender of the household head (male) 0.9 0.8 11 18 0.39
Literacy of the household head (1 if literate; 0 otherwise) 1.0 1.0 3 94 0.74
Farming experience (years) 24.6 25.7 7 66 0.53
Primary decision maker Primary decision maker is 15–64 years female (1 = yes; 0 = No) 0.3 0.3 7 53 0.57
Primary decision maker is 15–64 years male (1 = yes; 0 = No) 0.1 0.2 5 77 0.73
Primary decision maker > 64 years Female (1 = yes; 0 = No) 0.02 0.05 12 30 0.31
Household Household size (number of members) 6.9 6.3 22 24 0.11
Hosting big social events (1 = yes; 0 = No) 0.3 0.3 5 61 0.70
Total cultivated land (ha) 8.3 8.8 8 85 0.64
Maize area (ha) 2.0 2.1 5 88 0.79
Annual expense (000’KES) 130.0 120.0 10 88 0.53
Off-farm income (000’KES) 300.0 340.0 8 86 0.76
Having mobile phone money account (1 = Yes, 0 = No) 1.0 1.0 2 96 0.70
Having bank account (1 = Yes, 0 = No) 0.8 0.8 14 80 0.19
Distance to a passable road (km) 1.5 2.2 12 60 0.26
AEZ = Moist transitional 0.5 0.6 15 69 0.25
AEZ = Moist mid-altitude 0.5 0.4 16 76 0.25
a
Those households that lacked suitable matches were dropped during the propensity score matching. Since agroecological zone was used as a covariate, households from
low tropics, dry transitional and highland zones with no metal silos could not be matched.
52 Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55

Table 5 The impact of the adoption of metal silos on reducing maize


Statistical tests to evaluate the matching. losses in storage (loss abatement) was positive and significant for
Matching Pseudo Likelihood ratio p > Chi2 Mean Median all matching algorithms. Metal silo adopters lost on average only
method R2 Chi2 bias bias 3 kg of maize grain in storage, valued at KSh153 (US$ 2), while
Before matching 0.24 175.16 0.000 33 29 non-adopters lost on average 157–198 kg (worth US$104–132).
Neighbor 0.04 14.89 0.604 9 7 The adoption of metal silos also significantly reduced the number
matching of households who use pesticides, and reduced the average cost
Radius matching 0.02 6.35 0.991 4 3
Kernel matching 0.01 5.01 0.998 4 3
of pesticides by KSh1400 (US$ 19). While this is a small amount,
it could be reduced still further since recent research has shown
that maize can safely be stored in silos without the use of pesti-
cides (De Groote et al., 2013).
Households that have adopted metal silos for grain storage are,
on average, able to store their maize for at least nine weeks longer
than non-adopters. The average increase differed slightly by
matching algorithms, from between 1.8 months (radius matching)
to 2.4 months (neighbor matching).
The metal silos lead to a clear and significant reduction in food
insecurity, as measured by both indicators. Households that
adopted metal silos reduced the time of inadequate food provision
by five to six weeks over the year. They also had lower food inse-
curity scores than non-adopters, on average by a score of one (ker-
nel) and 1.5 (neighbor).

Testing for hidden bias with sensitivity analysis

Sensitivity analysis with Rosenbaum bounds was conducted to


check for hidden bias (Table 7). The bounds under the assumption
Fig. 6. Distribution of the propensity scores and common support for propensity
that the true treatment effect was underestimated (sig-) reveal
score estimation for the metal silo adopters and non-adopters.
that the results are largely insensitive to unobserved covariates.
showed that they greatly overlapped (Fig. 6). Metal silo adopters Even for a threefold increase in gamma, the results are still signif-
with appropriate matches from among the non-adopters are icant for treatment effect on losses and duration of storage. The
shown on the graph as ‘treated on support’. odds of differential assignment of treatment due to unobserved
factors (gamma) would have to increase by a factor of 1.4 and
1.3 respectively to change the inference on the effect of metal silos
on cost of storage chemicals and months of food insecurity.
Impact of metal silo technology

To estimate the impact of the technology, the average treatment Conclusion


effect on the treated (ATT), after matching, was calculated (Table 6).
The results showed that the adoption of metal silos significantly re- This study used propensity score matching to evaluate the im-
duced storage loss and cost of insecticide use, increased the time pact of metal silo technology on postharvest loss abatement, cost
for which households stored maize, and reduced household food of pesticide use, length of maize storage and household food inse-
insecurity. All the matching estimators provided similar results. curity, based on a representative household survey and a survey of

Table 6
Impact of metal silo adoption on cost of storage, loss abatement, length of storage and food security.

Outcome variable Matching algorithm Treated Controls ATT SE t-stat Matched observations
Adopters Nonadopters Total
Grain losses in storage (kg/household) Kernel matching 3.07 171.53 168.46 17.95 9.38
Radius 3.07 156.76 153.69 16.49 9.32 112 559 671
Neighbor 3.07 198.47 195.40 49.94 3.91
Cost of storage chemicals (KSh/household) Neighbor 341.25 1748.25 1406.25 1071.00 1.31
Radius 341.25 1468.50 1127.25 267.75 4.22 112 548 660
Kernel 336.00 1737.00 1400.25 274.50 5.1
Duration of maize storage (months) Kernel 7.55 5.67 1.88 0.39 4.78
Radius 7.55 5.75 1.80 0.38 4.08 112 780 892
neighbor 7.55 5.18 2.37 0.56 4.23
Household food insecurity (months) Kernel 0.99 2.07 1.08 0.26 4.09
Radius 0.98 2.12 1.14 0.25 4.47 112 777 889
neighbor 0.98 2.11 1.13 0.38 2.92
Household food insecurity access scale (HFIAS) Score Neighbor 3.77 5.31 1.53 0.82 1.86
Radius 3.77 5.16 1.38 0.63 2.21 112 780 892
Kernel 3.77 4.78 1.01 0.67 1.51
Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55 53

Table 7
Sensitivity analysis with Rosenbaum bounds.

Gamma Loss abatement Cost of storage chemicals Food security Storage duration
sig+ sig- sig+ sig- sig+ sig- sig+ sig-
1 1.50E09 1.50E09 3.10E07 3.10E07 4.80E07 4.80E07 0.00E+00 0.000
1.1 5.60E07 1.30E12 7.40E10 0.000 5.50E10 0.000 0.00E+00 0.000
1.2 0.000 6.70E16 1.10E12 0.001 3.10E13 0.004 0.00E+00 0.000
1.3 0.001 0 1.00E15 0.017 1.10E16 0.049 0 0.000
1.4 0.013 0 0.000 0.095 0 0.226 0 0.000
1.5 0.069 0 0.000 0.286 0 0.537 0 0.000
1.6 0.216 0 0.000 0.553 0 0.811 0 0.000
1.7 0.447 0 0.000 0.784 0 0.948 0 0.000
1.8 0.686 0 0.000 0.920 0 0.990 0 0.000
1.9 0.857 0 0.000 0.977 0 0.999 0 0.000
2 0.948 0 0.000 0.995 0 1.000 0 0.000
2.1 0.984 0 0.000 0.999 0 1.000 0 0.002
2.2 0.996 0 0.000 1.000 0 1.000 0 0.009
2.3 0.999 0 0.000 1.000 0 1 0 0.032
2.4 1.000 0 0.000 1.000 0 1 0 0.086
2.5 1.000 0 0.000 1 0 1 0 0.183
2.6 1.000 0 0.000 1 0 1 0 0.323
2.7 1 0 0.000 1 0 1 0 0.488
2.8 1 0 0.000 1 0 1 0 0.649
2.9 1 0 0.000 1 0 1 0 0.782
3 1 0 0.000 1 0 1 0 0.878

gamma – Log odds of differential assignment due to unobserved factors.


sig+ – Upper bound significance level (overestimation of treatment effect).
sig – Lower bound significance level (underestimation of treatment effect).

metal silo owners. Before matching, substantial differences existed higher prices. Similar results were found in South America where
between adopters and non-adopters. The estimation of the logistic non-adopters of metal silos sold 50% of their harvested maize
model indicated that literacy, wealth, access to financial services shortly after harvest (Bokusheva et al., 2012).
and access to roads positively influenced the adoption of metal si- Because of the reduction in storage losses, combined with an in-
los. After propensity, score matching, however, these differences crease of income due to sales of maize at higher prices, adopting
were no longer significant, so the difference in outcomes between households reduced the period they had inadequate food by more
matched adopters and non-adopters can now be attributed to the than one month. This was also observed in Latin America, where
adoption of metal silos. subsistence farmers who adopted metal silos were able to increase
The major effect of the metal silos was an almost complete the period of their household food security by 30 to 35 days per an-
elimination of losses due to insect pests. These results corroborate num (Fischler et al., 2011).
findings in Central America which showed that metal silos reduced The combined effect of farmers suffering less storage losses,
losses to <1% (Bokusheva et al., 2012). In Kenya, the reduced losses storing more grain and for a longer period, and selling the grain la-
saved the farmers an average of 150–198 kg of grain, worth ter in the season, is likely to buffer maize prices by reducing de-
KSh9750 (US$ 130). The grain saved represents the average annual mand and stabilizing the supply (less after harvest, more later).
consumption of two people, or four months of maize consumption Effective grain storage using metal silos by many farmers could
by an average family of six. The value of saved grain is substantial, therefore help to stabilize food prices across seasons.
especially when compared to the food poverty line KSh988 (US$13) The relatively high cost of the metal silo, however, constitutes
per adult per month of rural Kenyan farmers (Muyanga et al., an impediment to increased demand and to widespread adoption.
2010). An economic analysis could provide guidelines to the circum-
Metal silo adopters also spent about KSh340 on storage insecti- stances under which the technology is economical. Currently, the
cides, compared with KSh1740 for non-adopters, a gain of about supply is also limited. At the time of the survey, 2011, more than
KSh1400. This gain could even be greater, because the use of insec- forty metal silo artisans had been trained in Kenya but only four
ticides in the metal silos is not really necessary, as a recent study were found to be active, partly due to lack of effective demand
has demonstrated (De Groote et al., 2013). However, the use of and awareness among farmers. The active ones usually subcontract
insecticide by the adopters could be attributed to the advice they other trained artisans when orders come in. The artisans are clearly
received from the artisans, while some also kept maize aside in clustered around the pilot areas where training was organized.
bags for consumption to avoid opening the silo and letting oxygen Enhancing adoption of metal silos calls for a deliberate effort to
enter before all the insects had died. Maize stored in bags is prone create awareness and promote metal silos among rural farmers.
to pest attack and the use of insecticides is recommended. This can be achieved through training and educating agricultural
The reduction in storage losses leads to a change in storage and extension officers on how to use the technology, as well as by hold-
marketing behavior. Households who adopted metal silos were ing farmer field days and promoting the technology through the
able to store their maize for 1.8–2.4 months longer than the non- media that are popular with farmers, in particular radio broadcasts
adopters. Further, households without metal silos sold most of in Kiswahili and local dialects. Further, artisans need to be trained
their marketed maize grain within the first few months after stor- in other centers to increase accessibility. In addition, artisans and
age, at the low prices prevalent during that period. The adopters, farmers have to be better informed that the use of insecticides in
on the other hand, knowing that their stored maize was safe, were metal silos is not necessary.
encouraged to wait for better prices. As a result, the adopters only In the second phase of the effective grain storage project (EGSP
sold a small portion of their maize in the months after harvest, but II), national campaigns have been organized to accelerate the diffu-
sold much of their marketed grain after five months, at much sion of metal silos and other storage technologies. In the future,
54 Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55

artisan training will target other areas, and training of extension Diagne, A., Demont, M., 2007. Taking a new look at empirical models of adoption:
average treatment effect estimation of adoption rates and their determinants.
officers and nationwide radio campaigns are planned. Policy inter-
Agric. Econ. 37, 201–210.
ventions such as the provision of affordable credit facilities and Diiro, G.M., 2013. Impact of Off-farm Income on Agricultural Technology Adoption
collective action can help small-scale farmers access the storage Intensity and Productivity. International Food Policy Reasearch Institute.
technology. Government can play a role in reducing the cost of me- FAO. 2008. Household Metal Silos: Key Allies in FAO’s Fight Against Hunger.
Agricultural and Food Engineering Technologies Service, Rome, Italy.
tal silos by taking off the import duty on raw materials and in par- FAO. 2011. FAO Statistical Databases (FAOSTAT). Available by Food and Agriculture
ticular on the metal sheets which cost Ksh2500 per sheet. A silo Organization (FAO) (verified March 11, 2013).
with a capacity of 540 kg costing KSh15,000 would require three FAO, and World Bank. 2010. FAO/World Bank Workshop on Reducing Post-Harvest
Losses in Grain Supply Chains in Africa: Lessons learned and practical
such sheets. guidelines. FAO Headquarters, Rome, Italy, 18–19 March 2010, Rome.
Strengthening artisans’ associations, finally, can enhance the FAOSTAT. 2010. Available by Food and Agriculture Organization of the United
marketing and promotion of metal silos as well as self-regulation Nations <http://faostat.fao.org/site/567/DesktopDefault.aspx>.
Fischler, M., Berlin, R., Bokusheva, R., Finger, R., Marín, Y., País, F., Pavón, K, Pérez, F.,
in maintaining standards. Training more artisans near farmers’ 2011. POSTCOSECHA Programme Central America. SDC.
localities can increase access to metal silos, reduce transport costs Gbegbelegbe, S., De Groote, H., 2012. Spatial and Temporal Maize Price Analysis in
and enable the artisans to offer aftersales service to the farmers. East Africa. Poster presented at the Twenty-eighth International Conference of
Agricultural Economists (ICAE), Foz do Iguaçu, 18–24 August 2012.
Hassan, R.M., Njoroge, K., Corbett, J.D., Njoroge, K., 1998. Combining geo-referenced
Acknowledgements survey data with agroclimatic attributes to characterize maize production
systems in Kenya. In: Hassan, R.M. (Ed.), Maize Technology Development and
Transfer. A GIS Application for Research Planning in Kenya. CAB International,
The authors would like thank the Swiss Agency for Develop- Oxon, UK, pp. 43–68.
ment and Cooperation (SDC) for supporting the adoption survey, Heckman, J.J., Ichimuro, H., Todd, E.P., 1997. Matching as an econometric evaluation
and the Aflacontrol project for supporting the baseline survey used estimator: evidence from evaluating a job training programme. Rev. Econ. Stud.
64, 605–654.
as a control. Authors would also like to thank the Syngenta founda- Hellin, J., Kanampiu, F., 2008. Metal Silos Bring Food Security to El Salvador.
tion for funding a writing workshop, where a first draft of this pa- Appropriate Tech. 35, 69–70.
per was developed, and Liz Lucas for editing. Hodges, R.J., Buzby, J.C., Bennett, B., 2010. Postharvest losses and waste in
developed and less developed countries: opportunities to improve resource
use. J. Agric. Sci. 149, 37–45.
References Jalan, J., Ravallion, M., 2001. Does Piped Water Reduce Diarrhea for Children in Rural
India? Indian Statistical Institute and World Bank, Washington DC.
Jalan, J., Ravallion, M., 2003. Does piped water reduce diarrhea for children in rural
Amare, M., Asfaw, S., Shiferaw, B., 2012. Welfare impacts of maize–pigeonpea
India? J. Econom. 112, 153–173.
intensification in Tanzania. Agric. Econ. 43, 27–43.
Kalton, G., 1983. Introduction to survey sampling. Sage Publications Inc., London.
Asfaw, S., Shiferaw, B., Simtowe, F., Lipper, L., 2012. Impact of modern agricultural
Kassie, M., Shiferaw, B., Muricho, G., 2011. Agricultural technology, crop income,
technologies on smallholder welfare: evidence from Tanzania and Ethiopia.
and poverty alleviation in Uganda. World Dev. 39, 1784–1795.
Food Policy 37, 283–295.
Khandker, S.R., Koolwal, G.B., Samad, H.A., 2010. Handbook on Impact evaluation:
Barrett, C.B., Reardon, T., Webb, P., 2001. Nonfarm income diversification and
Quantitative Methods and Practices. The World Bank, Washington, DC.
household livelihood strategies in rural Africa: concepts, dynamics, and policy
Kiiza, B., Mugisha, J., Lwasa, S., 2010. Effect of access to formal market information
implications. Food Policy 26, 315–331.
on prices received by smallholder farmers in uganda. Int. J. ICT Res. Dev. Africa
Becerril, J., Abdulai, A., 2009. The impact of improved maize varieties on poverty in
1, 1–18.
Mexico: a propensity score-matching approach. World Dev. 38, 1024–1035.
Kimenju, S.C., De Groote, H., 2010. Economic Analysis of Alternative Maize Storage
Becker, O.S., Andrea, I., 2002. Estimation of average treatment effects based on
Technologies in Kenya Cape Town, South Africa. African Association of
propensity scores. Stata J. 2, 358–377.
Agricultural Economists (AAAE) and 48th Agricultural Economists Association
Bilinsky, P., Swindale, A., 2005. Months of Inadequate Household Food Provisioning
of South Africa (AEASA).
(MIHFP) for Measurement of Household Food Access: Indicator Guide, Food and
KNBS, 2010. 2009 Kenya Population and Housing Census: Population Distribution
Nutrition Technical Assistance Project (FANTA), Academy for Educational
by Administrative Units Volume IA. Republic of Kenya, Nairobi Kenya.
Development, Washington, DC.
La Rovere, R., Kostandini, G., Abdoulaye, T., Dixon, J., Mwangi, W., Guo, Z., Bänziger,
Bokusheva, R., Finger, R., Fischler, M., Berlin, R., Marín, Y., Pérez, F., Paiz, F., 2012.
M., 2010. Potential Impact of Investments in Drought Tolerant Maize in Africa.
Factors determining the adoption and impact of a postharvest storage
CIMMYT, Addis Ababa, Ethiopia.
technology. Food Security, 279–293.
Lee, W.S., 2008. Propensity score matching and variations on the balancing test. In:
Boxall, R.A., 1997. Constraints Analysis of the Post-production Sector in Zambia
Third Conference on Policy Evaluation, ZEW, Mannheim, Germany.
FAO-AG–PFL/ZAM/003/PFL. Food and Agricultural Organization of the United
Lorío, M.D.J., Njie, D.N., 2011. The Household Metal Silo: A Helpful Technology For
Nations, Rome.
Food Security. FAO, Rome.
Boxall, R.A., 1998. Grains Post-Harvest Loss Assessment in Ethiopia. Report 2377.
Mellor, J.W., Delgado, C.L., Blackie, M.J., (Eds.) 1987. Accelerating Food Production in
Natural Resources Institute Chatham, UK.
Sub-Saharan Africa. The Johns Hopkins University Press, Baltimore and London.
Boxall, R.A., 2002. Damage and loss caused by the Larger Grain Borer Prostephanus
Mendola, M., 2007. Agricultural technology adoption and poverty reduction: a
truncatus. Integr. Pest Manage. Rev. 7, 105–121.
propensity-score matching analysis for rural Bangladesh. Food Policy 32, 372–
Chen, W.V., Zeiser, K., 2008. Implementing Propensity Score Matching Causal
393.
Analysis with Stata. Penn State University, Population Research Institute.
M’mboyi, F., Mugo, S., Mwimali, M., Ambani, L., 2010. Maize Production and
Coates, J., Swindale, A., Bilinsky, P., 2007. Household Food Insecurity Access Scale
Improvement in Sub-Saharan Africa African Biotechnology Stakeholders Forum
(HFIAS) for Measurement of Household Food Access: Indicator Guide (v. 3),
(ABSF). Nairobi, Kenya.
Food and Nutrition Technical Assistance Project (FANTA), Academy for
Muyanga, M., Jayne, T.S., Burke, W.J., 2010. Pathways into and out of Poverty: A
Educational Development, Washington, DC.
study of Rural Household Wealth Dynamics in Kenya WPS 39/2010. Tegemeo
De Groote, H., 1996. Optimal survey design for rural data collection in developing
Institute of Agricultural Policy and Development, Nairobi, Kenya.
countries. Quart. J. Int. Agric. 35, 163–175.
Ngamo, T.S.L., Ngassoum, M.B., Mapongmestsem, P.M., Mapongmestsem, P.M.,
De Groote, H., Narrod, C., Kimenju, S., Bett, C., Scott, R., Tiongco, M., 2011. Measuring
Haubruge, E., Lognay, G., Hance, T., 2007. Current post harvest practices to avoid
rural consumers’ willingness to pay for quality labels using experimental
insect attacks on stored grains in Northern Cameroon. Agric. J. 2, 242–247.
auctions: the case of aflatoxin free maize in Kenya. In: Poster paper presented at
Nguyo, R., 2007. Grain storage: keeping weevils at bay with metal silos. In:
the European Association of Agricultural Economics Congress 2011. Zurich,
Rootheart, R., Taylor, P. (Eds.), Creating Sustainable Agricultural Impact for
Switzerland, August 30th to September 2nd, 2011.
Smallholders: MATF Innovative Partnerships and Approaches 4th MATF
De Groote, H., Kimenju, S.C., Likhayo, P., Kanampiu, F., Tefera, T., Hellin, J., 2013.
Experience Sharing Workshop. Mombasa, Kenya, pp. 19–23.
Effectiveness of hermetic systems in controlling maize storage pests in Kenya. J.
Nweke, F.I., 2005. A Review of Cassava in Africa with Country Case Studies on
Stored Prod. Res. 53, 27–36.
Nigeria, Ghana, the United Republic of Tanzania, Uganda and Benin 2. FAO and
de Janvry, A., Dustan, A., Sadoulet, E., 2011. Recent Advances in Impact Analysis
IFAD 2005.
Methods for Ex-post Impact Assessment of Agricultural Technology: Options for
Pingali, P.L., (Ed.) 2001. CIMMYT 1999-2000 World Maize Facts and Trends. Meeting
the CGIAR. Independent Science and Partnership Council Secretariat, Berkeley,
World Maize Needs: Technological Opportunities and Priorities for the Public
California, USA.
Sector. CIMMYT, Mexico, DF.
de Lima, C.P.F., 1979. The assessment of losses due to insects and rodents in maize
Pinstrup-Andersen, P., 2009. Food security: definition and measurement. Food
stored for subsistence in Kenya. Trop. Stored Prod. Inform. 38, 21–26.
Security 1, 5–7.
De Lucia, M., Assennato, D., 1994. Agricultural Engineering in Development: Post-
Proctor, D.L., (Ed.) 1994. Grain storage techniques. Evolution and Trends in
Harvest Operations and Management of Foodgrains. Food and Agriculture
Developing Countries. FAO Agricultural Services Bulletin No. 109. FAO, Rome.
Organization of the United Nations, Rome.
Z.M. Gitonga et al. / Food Policy 43 (2013) 44–55 55

Rembold, F., Hodges, R., Bernard, M.H.Knipschild., Léo, O., 2011. The African Tefera, T., Kanampiu, F., De Groote, H., Hellin, J., Mugo, S., Kimenju, S., Beyene, Y.,
Postharvest Losses Information System (APHLIS). European Union, Luxembourg. Boddupalli, P.M., Shiferaw, B., Banziger, M., 2011. The metal silo: An effective
Rosenbaum, P.R., 2002. Observational Studies, second ed. Springer-Verlag, New grain storage technology for reducing post-harvest insect and pathogen losses
York. in maize while improving smallholder farmers’ food security in developing
Rosenbaum, R.P., Rubin, B.D., 1985. Constructing a control group using multivariate countries. Crop Protection 30, 240–245.
matched sampling methods that incorporate the propensity score. Am. Stat. 39, Thamaga-Chitja, J.M., Hendriks, S.L., Ortmann, G.F., Green, M., 2004. Impact of maize
33–38. storage on rural household food security in Northern Kwazulu-Natal. J. Family
Sianesi, B., 2004. An Evaluation of the Active Labour Market Programmes in Sweden. Ecol. Consum. Sci. 32, 8–15.
Rev. Econom. Statistics 86, 133–155. Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. MIT
Tefera, T., 2012. Post-harvest losses in African maize in the face of increasing food Press, Cambridge, MA.
shortage. Food Security 4, 267–277.

You might also like