Adoption of Drought Tolerant Maize Varieties under Rainfall Stress in
Malawi
Samson P. Katengeza*1, Stein T. Holden1 and Rodney W. Lunduka2
Paper prepared for presentation at the Fifth Conference of the African Association of
Agricultural Economists (5th CAAAE), 23-26 September 2016, Addis Ababa, Ethiopia
1
Norwegian University of Life Sciences, School of Economics and Business, P.O. Box 631, 1432 Ås, Norway
CIMMYT- Southern Africa Regional Office (SARO), PO Box MP 163, Mount Pleasant Harare, Zimbabwe
* Corresponding author, samkatengeza@gmail.com, Phone: +265 888 705 351, +47 486 433 426
2
Adoption of Drought Tolerant Maize Varieties under Rainfall Stress in Malawi
Abstract
This paper examines adoption of drought tolerant maize varieties under rainfall stress in Malawi
using correlated random effects Probit and Tobit models with control function approach.
Drought tolerant maize is a promising technology that has the capacity to help smallholder
farmers adapt to drought risks. Using 2009, 2012 and 2015 data from six districts, results show
adoption has increased from 46% in 2009 to 59% in 2015. The likelihood of adoption is
significantly increased by drought with early droughts having greater impact (31%) than late
droughts (20%). Early droughts are also associated with an increased acreage of land allocated
to drought tolerant maize and quantity of seed planted. However lagged drought variables
appear to negatively affect adoption. The possible explanation is that the years preceding the
surveys were associated with good rains such that farmers responded by buying less of drought
tolerant maize anticipating similar rainfall pattern. Another important driver of adoption is the
farm input subsidy programme. However, while access to subsidised seed increases both
adoption and intensity of adoption, previous year’s access has a negative impact. This suggests
that the increased adoption is due to availability of cheap seed as opposed to farmers’ previous
exposure to the varieties. This may indicate limited awareness on the benefits of drought tolerant
maize varieties. This is also consistent with extension visits positively affecting adoption. Good
extension messages and promotion of drought tolerant maize varieties should be improved to
allow farmers make informed decisions.
Key words: Drought tolerant maize, drought exposure, farm input subsidy programme,
correlated random effects, Malawi
Page 2 of 20
Introduction
Recurrent extreme weather events such as droughts and floods undermine crop yields and
aggregate production thereby reducing food availability and agricultural incomes especially
among smallholder farmers in developing countries (Davies et al., 2009; Kassie et al., 2009;
Kato et al., 2011; Pauw et al., 2011). Failure by farm households to adapt to such weather shocks
worsens negative effects of these extreme weather events and inhibits further investment and
economic growth both by households and national level (Kassie et al., 2014; Kassie et al., 2009;
Kato et al., 2011; Nangoma, 2007). The extreme weather events kick start a knock-on effect that
start from low production to food insecurity and local and national economic shock (Devereux,
2007). Malawi is one of many countries in developing world greatly affected by negative
impacts of weather extremes. In past two decades, the country has experienced several adverse
climatic hazards that have led to severe crop losses, infrastructure damages and occasional
displacement of people (Nangoma, 2007; Pangapanga et al., 2012; Pauw et al., 2010). The most
recent shocks include droughts of 2004/05 and 2011/12 (Holden & Fischer, 2015; Holden &
Mangisoni, 2013) and 2014/15 flash floods early in the growing season and droughts thereafter.
Investing in agricultural production methods that boost farmers’ resilience against weather
shocks through climate change adaptation and disaster risk reduction approaches is a key
strategy to reduce negative impacts (Davies et al., 2009; Pangapanga et al., 2012). In a country
with poor or missing markets for insurance and credit and little off-farm activities, adoption of
agricultural management strategies that reduce production risks is the only realistic option for
smallholder farmers (Kassie et al., 2014). Drought tolerant (DT) maize variety is one potential
technology that has the capacity to help smallholder farmers adapt to drought risks. It is
estimated that DT maize can produce up to 30% of their potential yield after six weeks of water
stress, before and during flowering and grain-formation (Magorokosho et al., 2009; Mangisoni et
al., 2011). In Malawi maize is life such that the absence of the commodity is synonymous to food
insecurity (Katengeza et al., 2012). Production is predominantly rain fed and prone to frequent
droughts which may result in 50% yield loss (Fisher et al., 2015).
The government of Malawi has consequently taken a leading role in promoting and
disseminating DT maize varieties through the farm input subsidy programme (FISP) and the
Page 3 of 20
Agricultural Sector Wide Approach programme (ASWAp). Through the ASWAp, the
government's long-term objective is to promote sustainable and climate-smart agriculture
development (Asfaw et al., 2014) and shift from drought and flood prone farming systems to
methods that improve farmers’ adaptive capacity, enhance resilience and resource use efficiency,
increase crop yield and reduce yield variability in the face of weather extremes (Garrity et al.,
2010; Lipper et al., 2014). FISP has consequently been reported as a major determinant of DT
maize adoption in Malawi (Holden & Fisher, 2015). Fisher et al. (2015) cited unavailability of
seed and high seed price as barriers to adoption of DT maize emphasising the importance of farm
input subsidy program in enhancing accessing to DT maize seed.
Adoption of DT maize varieties has been previously studied (Fisher et al, 2015: Holden and
Fisher, 2015). Fisher et al. (2015) used cross sectional data from six countries in Africa where
Drought Tolerant Maize for Africa (DTMA) project is promoting dissemination of drought
tolerant maize varieties including Malawi while Holden and Fisher (2015) used three year panel
data (2006, 2009 and 2012) for Malawi. These studies looked at general adoption of the DT
maize varieties. We build on these two studies using three year panel data (2009, 2012 and 2015)
for Malawi to examine adoption of DT maize varieties under rainfall stress conditions. We
believe that exposure to drought condition and knowledge of the benefits of the DT maize
varieties may trigger a change in farmers’ behaviour to adopt to the new varieties. Our data
covers three important weather variations namely normal rainfall in 2009, early droughts in 2012
and a combination of flash and riverine floods and late droughts in 2014/15. The paper addresses
the following hypotheses: (1) Exposure to drought increases farmers’ likelihood of adopting
drought tolerant maize varieties; (2) access to farm input subsidy program increases adoption of
drought tolerant maize varieties; (3) past exposure to drought tolerant maize varieties increases
adoption in the following years.
Model Specification and Estimation Strategy
Adoption methods
Farmers’ adoption decision of DT maize is modelled as in Holden and Fisher (2015) based on
three year panel data (2009, 2012 and 2015). The model is specified as follows:
Page 4 of 20
𝐷𝑇𝑖𝑝𝑡 = 𝛼0 + 𝛼1 𝑅𝑑𝑡 + 𝛼2 𝐷𝑟𝑖𝑡 + 𝛼3 𝑆𝑖𝑝𝑡 + 𝛼4 𝐻𝑖𝑡 + 𝛼5 𝑃𝑖𝑝𝑡 + 𝛼6 𝐷𝑠𝑖𝑡 + 𝛼𝑖 + 𝜀𝑖𝑝𝑡
1
where DTipt is the dependent variable and is a dummy on whether household i grew DT maize
on plot p in year t or not. The explanatory variables captured as 𝑋𝑖𝑝𝑡 are defined as follows: R dt
is a vector of variables capturing rainfall stress (times of dry spells, times of early and late dry
spells) in the farmer’ district d in year t. Drit is farmer i perception on exposure to drought in
year t. Sipt is a vector of institutional variables such as a dummy for access to subsidized inputs
and used them on the plot, visits by extension workers and whether farm household accessed
input credit. Hit denotes household characteristics. Pipt controls for observable farm plot
characteristics such as soil type, slope, fertility status, plot size and distance to plot from home
while Dsi𝑡 controls for location variables (survey districts). αi captures unobservable time-
invariant characteristics of households and farms such as managerial ability and unobservable
land quality. εipt is normally distributed error term and we assume is independent of 𝑋𝑖𝑝𝑡 .
Parameters in equation (1) can be estimated by either fixed effects (FE) Probit or correlated
random effects (RE) Probit. The FE method removes the unobserved effect (𝛼𝑖 ) by time
demeaning the data. The fixed effects Probit thus sweeps away all explanatory variables that are
constant over time (Wooldridge, 2014). Again fixed effects estimation may cause incidental
parameters problem especially when unobserved effects (αi ) are taken as parameters to be
estimated (Wooldridge, 2009). The alternative method is the random effects estimator. The
traditional RE Probit model assumes that unobserved effects (αi ) and explanatory variables (𝑋𝑖 )
are independent, i.e.
𝐶𝑜𝑣(𝑋𝑖𝑡 , 𝛼𝑖𝑡 ) = 0, 𝑡 = 1, 2, … , 𝑇, & 𝑖 = 1, … , 𝑛
2
𝛼𝑖 |𝑋𝑖 ~𝑁(0, 𝜎𝛼2 )
3
and that αi is normally distributed, i.e.:
The validity of this unconditional normality depends on some restrictive assumptions but
becomes more reasonable as T gets large (Wooldridge, 2009). Thus, Arslan et al. (2014)
proposes testing the unconditional normality within conditional maximum likelihood (CMLE)
Page 5 of 20
framework. The CMLE approach allows 𝛼𝑖 and 𝑋𝑖 to be correlated (Chamberlain, 1980;
Wooldridge, 2010) assuming that
𝛼𝑖 |𝑋𝑖 ~𝑁(𝜑 + 𝛿𝑋̅𝑖 , 𝜎𝛾2 )
4
where σ2α is the variance of αi in the equation αi = |(φ + δX̅i + γi )
and 𝑋̅𝑖 ≡ 𝑇 −1 ∑𝑇𝑡=1 𝑋𝑖𝑡 is the 1 × K vector of time averages.
This approach enables the paper to estimate partial effects of 𝑋𝑖 on response probability at the
average value of 𝛼𝑖 (𝛾𝑖 = 0). The approach also avoids incidental parameters problem. Assuming
possible interdependency on adoption decisions of different technologies.
Intensity of adoption
We define intensity of adoption as the size of the plot in hectares (ha) under DT maize variety
and the quantity of DT seed planted. With the possibility of a censored plot size and DT maize
seed at plot level, we use correlated random effects Tobit to analyse the intensity of adoption.
Let the size of land that is allocated to a DT maize variety by farmer i at time t be 𝐿𝑖𝑡 . The
unobserved effects Tobit model for a corner at zero for Lit can be specified as:
𝐿∗𝑖𝑡 = 𝑚𝑎𝑥 (0, 𝛼𝑋𝑖𝑡 + 𝜀𝑖𝑡 + 𝛼𝑖 )
𝐷(𝜀𝑖𝑡 |𝑋𝑖𝑡 , 𝛼𝑖 ) = 𝑁(0, 𝜎𝜀2 )
5
6
where the dependent variable (𝐿𝑖𝑡 ) is the size of the plot in ha (and quantity of DT maize seed
planted on the plot) and the explanatory variables are as defined in equation 1(Wooldridge, 2010
pp; 540-542).
Study Areas, Data and Sampling Procedure
The paper uses three-year panel data from six districts in Malawi namely Lilongwe, Kasungu,
Chiradzulu, Machinga, Thyolo and Zomba. Agro-ecologically, Chiradzulu, Kasungu, Lilongwe,
Machinga, and Zomba districts are located in medium altitude zone which enjoys high average
rainfall ranging from 800 – 1,200 mm annually with an altitude of 1,000 to 1,500 metres above
sea level although Machinga is partly drought-prone district (Katengeza et al., 2012; Mangisoni
Page 6 of 20
et al., 2011). Thyolo district on the other hand belongs to the high plateaux and hilly areas which
lie in an altitude over 1,500 above sea level and receives over 1,200 mm of rainfall annually.
Agro-ecological and location variables affect adoption of agricultural technologies such as DT
maize. Such variables capture variations in rainfall, soil quality, production potential,
infrastructure development as well as availability of input and output markets (Doss & Doss,
2006).
The data is based on an original sample of 450 households surveyed in 2006 and 2007 and 376 in
2009, 350 in 2012 and 2015 (Table 1). The initial sample was randomly selected following the
2004 Integrated Household Survey Two (IHS 2). Data collection involves detailed farm plot
level information measured with GPS on plot sizes of which a total of 1076, 1387 and 1281 plots
are reported in 2009, 2012 and 2015, respectively. The plot is defined in this paper by Holden
and Lunduka (2012) as a “uniform crop stand that received homogenous input treatment”. The
quality of the data is better with minimal measurement errors because (1) data is collected from
all farm households’ plots unlike larger surveys which normally collect data from one plot (2) all
plots are measured with GPS as opposed to relying on farmers’ estimates which are prone to big
errors (Holden & Mangisoni, 2013). This data is also of interest as it captures weather extremes
namely, 2009 with good rains, 2012 early drought and 2015 flood and late drought.
Table 1: Study areas
District
Thyolo
Zomba
Chiradzulu
Machinga
Kasungu
Lilongwe
Total
2009
Households
50
41
79
45
90
71
376
Plots
145
115
117
146
356
197
1,076
2012
Households
47
76
37
47
82
61
350
Plots
162
264
163
185
388
225
1,387
2015
Households
47
79
34
45
80
65
350
Plots
168
272
120
161
331
229
1,281
Total
Households
144
196
150
137
252
197
1,076
Plots
475
651
400
492
1,075
651
3,744
Descriptive Statistics: Dependent and Explanatory Variables
The dependent variables: Drought tolerant maize varieties
Presented in Table 2 are descriptive statistics for the variables. Plot level data was collected on
maize varieties planted on the plot. Adoption is measured by whether an individual farm
Page 7 of 20
household planted DT maize variety in at least one of the plots while intensity is measured by the
area allocated to DT maize. Also included in intensity of adoption is the quantity of DT maize
seed planted. We expect an increase in level of adoption over time especially after 2011/12
drought, farmers should adopt more of DT maize as a response to drought shock. The continued
implementation of farm input subsidy programme is also expected to further increase adoption.
The issue of adoption is however subjective as it may depend on farmers’ perceptions. If farmers
perceive drought tolerant maize as climate-smart then adoption may increase in response to
drought shocks. Contrary, farmers may view DT maize as low yielding compared to other
improved maize (OIM) varieties hence adoption may decrease over time. Availability and
development of institutions such as agricultural extension services and credit markets also plays
important roles in adoption decisions.
Our results show 46% adoption of DT maize in 2009, 54% in 2012 and 59% in 2015. The results
are as expected with an upward sloping over the survey years. The question however is whether
the increase is due to farmers’ response to drought or other factors. Holden and Fisher (2015)
reported that the increase in adoption is mainly due to farm input subsidy programme which has
over the years disseminated DT maize varieties. The DT maize seed has been an integral
component in the FISP package and this has made it easy for farmers to access the seed. In terms
of plot area allocated to DT maize there is an increase from 0.224 ha in 2009 to 0.272 ha in 2015.
For seed there is also a slight increase from 5.0 kilograms in 2009 to 5.6 kilograms in 2015.
Page 8 of 20
Table 2: Definitions and summary statistics of variables by year
Variables
Description
Dependent variables
DTmaize
1 if household adopted drought tolerant maize variety
DTarea
Area in ha under DT maize
DTseed
Quantity of DT maize seed in Kg
Independent variables
Drought variables
Drought1yrFmr 1 =Farmers perceive drought occurred previous year
Drought0yrFmr 1 =Farmers perceive drought occurred survey year
Drought1yr
Times dry spells occurred (actual rainfall) previous year
Drought0yr
Times dry spells occurred (actual rainfall) survey year
Earlydrought
Times early dry spells (actual rainfall) previous year
Latedrought
Times late dry spells (actual rainfall) in survey year
Institutional variables
Seedfisp1yr
1=Household received seed subsidy coupon previous year
Seedfisp0yr
1=Household received seed subsidy coupon survey year
Extension
1=Household was visited by an extension worker
Creditinput
1=Household accessed farm input credit
Plot characteristics
Logplotdist
Log(plot distance in km + 1)
Landtenure
1 if operated by plot owner
Sandy soil
1=Farmers perceive sandy soil
Loam soil
1=Farmers perceive loam soil
Clay soil
1=Farmers perceive clay soil
Flat slope
1=Farmers perceive flat slope on plot
Moderate slope
1=Farmers perceive moderate slope on plot
Steep slope
1=Farmers perceive steep slope on plot
High fertility
1=Farmers perceive high soil fertility on plot
Medium fertility 1=Farmers perceive medium soil fertility on plot
Low fertility
1=Farmers perceive low soil fertility on plot
Plotsize(gps)
Plot size measured by GPS (ha)
Logplotsize
Log(Plot size in ha)
Plotsize(farmer) Plot size (reported by farmer) (ha)
Household demographic characteristics
Age
Age of household head (years)
Education
Education of household head (years)
Family size
Total family size (number)
Sex
1 = If gender of household head is male
Marital status
1 = If household head is married
Flabour
Family labour (no of persons)
Hlabour
Hired labour (no of persons)
2009
Year
2012
2015
0.456
0.224
4.969
0.538
0.257
4.837
0.587
0.272
5.607
0.531
0.253
5.139
0.082
0.091
2.219
0.757
0.000
0.000
0.273
0.441
1.280
2.130
0.794
0.000
0.239
0.740
2.736
3.035
0.869
0.391
0.207
0.446
2.048
2.045
0.591
0.134
0.337
0.370
0.474
0.105
0.598
0.555
0.149
0.063
0.696
0.649
0.290
0.091
0.561
0.538
0.290
0.085
5.455
0.913
0.244
0.479
0.273
0.582
0.363
0.050
0.178
0.612
0.204
0.338
-1.424
0.623
5.755
0.947
0.198
0.542
0.255
0.648
0.297
0.048
0.208
0.679
0.102
0.296
-1.582
0.341
6.278
0.931
0.206
0.673
0.121
0.518
0.415
0.068
0.087
0.705
0.208
0.301
-1.592
0.366
5.848
0.932
0.214
0.569
0.214
0.584
0.356
0.056
0.158
0.669
0.167
0.310
-1.540
0.430
46.88
4.572
5.554
0.809
0.765
2.614
0.915
51.06
4.983
5.468
0.763
0.731
2.870
0.817
48.94
4.838
5.715
0.731
0.679
2.623
1.297
49.15
4.817
5.578
0.765
0.722
2.706
1.023
Mean
Page 9 of 20
Explanatory variables
The choice of explanatory variables is based on our hypotheses, previous studies and available
data. Such variables include (1) rainfall stress variables, (2) plot-level factors (e.g. plot size,
perceived soil fertility, slope, soil type, and distance from home. (3) Household level factors (e.g.
sex of household head, age, education, family size, family labour, hired labour, and marital
status). (5) Institutional factors (e.g. access to extension, input credit, and farm input subsidy
programme).
Rainfall stress variables
We define rainfall stress variables in this analysis as those capturing dry spells. Exposure to dry
spells is a key variable in the analysis and we assess the extent to which the sampled households
were exposed to dry spells in each of the survey years (2009, 2012, and 2015) as well as lagged
variables. 74% of the farm households reported drought shock in 2015 while 44% reported
drought shock in 2012. This represents severity of drought shock in 2015 and 2012 than in 2009.
There is however an element of subjectivity in assessment of dry spell exposure using farm
household perceptions (Holden & Quiggin, 2015; Holden & Fisher, 2015). This may result in
endogeneity because more pessimistic farmers tend to overestimate the probability of a negative
outcome and therefore perceive higher probability of drought shocks. These farmers might also
be more risk-averse and more likely to adopt. We therefore constructed an objective drought
measure using daily rainfall data from meteorological services department to test whether
drought impact adoption of DT maize varieties. In 2015 dry spells occurred at least three times.
A dry spell is defined as a period of 10 – 15 days with a total rainfall of less than 20mm
following a rainy day of at least 20mm.
Institutional variables
Key institutional variables considered are agricultural extension services, credit access, and
access to the Farm Input Subsidy Programme (FISP). Agricultural extension services may
remain an important channel for agricultural technologies in Malawi. We measure access to
extension services as a dummy variable on whether farm households were visited by an
extension officer or not. On average 29% reported being visited by an agricultural extension
Page 10 of 20
worker at least once in a growing season. In defining the credit access variable we used the Feder
et al. (1990) approach which distinguishes between farmers who choose not to participate in
credit markets and those who do not have access to credit. Credit-constrained farmers are those
who need credit but are unable to get it while credit-unconstrained farmers are those who decide
not to participate as well as those who need and are given. Only 35% are credit unconstrained.
We, however, note that credit unconstrained may not be enough as farmers may access credit for
different reasons, hence we consider only those who accessed credit for farming reasons (e.g.
buying inputs). Only 8% accessed input credit. On subsidy, we find that the share of households
receiving seed subsidy coupon increased from 37% in 2009 to 65% in 2015. We also included
the lagged seed subsidy variable to assess whether previous access to DT maize seed can
enhance adoption of the same in the following years.
Plot level variables
Plot-specific variables include, perceived soil fertility, slope, soil type, plot size, fertiliser used
on plot, and distance from home. Plot distance from household residence is an important factor
that can influence adoption of CSA practices. Longer distances increase transaction costs, for
walking and monitoring hence less adoption (Kassie et al., 2015).
Household characteristic variables
Household level factors control for household heterogeneity and these include education of
household head, age, sex, marital status, family size, family and hired labour. These variables
may influence adoption decisions in countries such as Malawi which have high market
imperfections and institutional failures (Kassie et al., 2015). Education increases understanding
of shocks such as droughts and floods and the adaptive measures hence increases adoption
(Katengeza et al., 2012; Mangisoni et al., 2011). The average educational attainment of
household heads is 4.8 years of education in the sampled districts. The average age of the
household head is 49 years while about 72% of the sample households are male-headed. In terms
of family size, on average, there are five members in each of the sampled households with an
active labour force of 2.7. An active labour force is an important variable to explain adoption
decisions as some production activities require more labour.
Page 11 of 20
Results and discussion
DT maize adoption
Table 3 are results of determinants of adoption of drought tolerant maize varieties. Our primary
objective is to examine whether exposure to drought enhances adoption of DT maize. Farmers in
Malawi were exposed to early droughts in 2012 and late droughts in 2015 in addition to dry
spells in other years. The first Probit model (probit 1) uses two drought variables, namely, farmer
perception dummy variable given a value of one if a farmer perceives drought occurred in a
given year and a variable capturing number of times dry spells occurred in each of the three
years and their lagged variables. Although the coefficients of the probit model were not different
from the marginal effects, presented here are marginal effects. The results show a positive and
significant relationship between farmers’ exposure to drought and adoption of drought tolerant
maize varieties. Both the subjective (farmers’ perception) and objective drought variables
increase the probability of adopting DT maize. This result is consistent with our expectation and
the findings of Holden and Fisher (2015) that farmers who previously were exposed to drought
are more likely to adopt DT maize as an adaptive mechanism. Ding et al. (2009) also reported
that farmers’ experience with drought increases their likelihood of adopting conservation tillage
systems.
However lagged drought variables for both farmer perception and rainfall data are associated
with negative impact on adoption. The possible explanation is that the years preceding survey
years were normal years with no serious reported droughts. Therefore the drought tolerant maize
varieties would not have been necessary as farmers expected a normal year as previous.
The second probit (probit 2) expand the first model by replacing the aggregate drought variable
with early and late dry spells. We define early drought as a period between December and early
January which is planting time while late drought is a period between February and early March
which is a period of maize flowering and grain formation. The early drought appears to have a
greater impact on DT maize adoption than late drought. Farmers who are exposed to early and
late drought are 31% and 20% more likely to adopt DT maize, respectively. The possible
explanation is that early drought acts as early warning to farmers such that farmers are more
likely to buy and plant maize varieties which are drought tolerant. Another explanation is that
Page 12 of 20
early drought affects germination of maize forcing farmers to replant. Replanting implies farmers
buying more of early maturing varieties to fit into the growing season as Malawi has a unimodal
type of rainy season which ends by late March or early April. Although other hybrids are also
early maturing, the 2012 experience shows that most farmers opt for DT early maturing maize
varieties (Holden & Fisher, 2015) e.g. SC403 (Kanyani). The impact of drought on adoption of
DT maize is also supported by the district dummy variables. Farmers in Machinga a drought
prone district (Katengeza et al., 2012) are 35% more likely to adopt drought tolerant maize and
likely to increase plot size planted with DT maize seed by 58% than farmers in Thyolo who
receive more and stable rainfall.
Tobit models are for intensity of adoption using land sizes (in hectares) allocated to drought
tolerant maize and quantity of DT maize seed planted. In Tobit 1 we use the farmer perception
variable of drought as well as early and late drought variables on land under DT maize. Tobit 2
and Tobit 3 uses the control function approach on proportion of land and quantity of seed,
respectively. The results are consistent with the probit results where exposure to drought is
associated with likelihood of increasing acreage of land under DT maize as well as increasing
quantity of seed planted. Early droughts are more likely to increase acreage of land allocated to
drought tolerant maize by 33% and the quantity of drought tolerant maize seed bought by 86%.
The paper also tests the impact of access to farm input subsidy programme on adoption of DT
maize varieties. Access to FISP increases adoption by 34-37%. The results are in agreement with
Holden and Fisher (2015) who reported FISP as a strongest driver of adoption of DT maize.
However while access to seed subsidy input increases both adoption and intensity of adoption,
previous year’s access has a negative impact on adoption. This suggests that the increased
adoption is due to the availability of cheap seed as opposed to farmers’ previous exposure with
the drought tolerant maize varieties. A plausible explanation is the lack of information on the
benefits of the DT maize varieties. Fisher et al. (2015) reported that about 40% of small holder
farmer did not grow DT maize varieties because of poor labelling of DT maize packages. There
is yet to be a very clear labelling of DT maize varieties that provide enough information to
farmers to make informed decision. This has been achieved in early maturing varieties like
SC403 by SEEDCO that have used a symbol of a monkey to show speed and fast maturity of the
early maize variety.
Page 13 of 20
Visits by agricultural extension workers, flat slope, high soil fertility, are also associated with
high adoption while distance to plot and plot sizes reduces the likelihood of adoption. The
positive significance of extension visits confirms the importance of increased awareness of the
varieties to enhance adoption. Controlling for household heterogeneity, the results show that
education, age and being married are associated with less probability of adoption while
household size and family labour increases the likelihood of both adoption and adoption
intensity. Hired labour is also associated with the increased probability of allocating more land to
DT maize cultivation as well as increasing DT maize seed.
Page 14 of 20
Table 3: Correlated Random Effects Probit (marginal effects) and Tobit Models with Control Function Approach
Variables
Drought variables
Drought1yr (Times of dry spell occurrence previous year)
Drought0yr (Times of dry spell occurrence survey year)
Drought1yrFmr (1 if farmer perceives drought occurred previous year)
Drought0yrFmr (1 if farmer perceives drought occurred survey year)
Probit 1
(b/se)
Tobit with CFA
(b/se)
-0.170***
-0.060
0.211***
-0.050
0.309***
-0.060
0.199**
-0.090
-0.103***
-0.030
0.158***
-0.030
0.102***
-0.030
0.100**
-0.050
-0.246***
-0.030
0.385***
-0.030
0.329***
-0.030
0.313***
-0.040
-0.440***
-0.060
1.009***
-0.050
0.863***
-0.060
0.823***
-0.080
-0.101*
-0.060
0.375***
-0.060
0.159***
-0.050
-0.029
-0.080
-0.101*
-0.060
0.341***
-0.060
0.187***
-0.050
0.003
-0.080
-0.058*
-0.030
0.130***
-0.030
0.116***
-0.030
-0.088*
-0.050
-0.145***
-0.030
0.399***
-0.030
-0.471***
-0.060
1.099***
-0.060
-0.159***
-0.040
-0.139*
-0.080
-0.226***
-0.050
-0.403***
-0.100
-0.154**
-0.070
-0.087
-0.090
-0.015
-0.060
-0.048
-0.070
-0.170***
-0.050
-0.240**
-0.110
-0.149**
-0.070
-0.096
-0.090
0.049
-0.060
0.018
-0.070
-0.123***
-0.030
-0.197***
-0.060
-0.017
-0.040
-0.054
-0.050
0.048
-0.030
0.099**
-0.040
-0.242***
-0.020
-0.407***
-0.050
-0.160***
-0.030
-0.130***
-0.040
0.119***
-0.030
0.148***
-0.030
-0.502***
-0.050
-0.999***
-0.110
-0.492***
-0.060
-0.331***
-0.080
0.391***
-0.060
0.105
-0.070
Early drought (Dec to early Jan)
Late drought (Feb to early March)
Institutional variables
Access to seed subsidy previous year
Access to seed subsidy in survey year
Extension visits
Input credit
Plot characteristics
Moderate slope
Steep slope
Medium fertility
Low fertility
Loam soil
Clay soil
Tobit with CFA
(Quantity dt seed)
(b/se)
Tobit (Plot size)
(b/se)
-0.044*
-0.020
0.065***
-0.020
-0.070
-0.060
0.353***
-0.050
Probit 2
(b/se)
Page 15 of 20
Log plot distance
Log plot size
Household characteristics
Education
Sex of household head (1=male)
Household size
Age
Age squared
Marital status (1=married)
Family labour
Hired labor
-0.050***
-0.010
0.002
-0.030
-0.051***
-0.010
-0.003
-0.030
-0.019**
-0.010
0.064***
-0.010
-0.074***
-0.010
0.091***
-0.010
-0.210***
-0.010
0.061**
-0.030
-0.020***
-0.010
0.507***
-0.100
0.041***
-0.010
-0.012*
-0.010
0.000**
0.000
-0.480***
-0.090
0.038**
-0.020
0.024**
-0.010
-0.019***
-0.010
0.483***
-0.100
0.045***
-0.010
-0.011
-0.010
0.000*
0.000
-0.464***
-0.090
0.048***
-0.020
0.020**
-0.010
-0.004
0.000
0.216***
-0.050
0.036***
-0.010
-0.004
0.000
0.000
0.000
-0.232***
-0.050
0.032***
-0.010
0.009*
-0.010
0.000
0.000
0.075***
-0.010
0.027***
0.000
0.171***
-0.020
0.075***
-0.010
-0.021
-0.120
0.127
-0.100
0.354***
-0.110
0.051
-0.090
-0.103
-0.120
0.034
-0.070
0.024
-0.060
0.165***
-0.060
0.159***
-0.050
-0.006
-0.070
0.171
-0.250
0.000
3300
-0.147
-0.140
0.000
3300
0.026
-0.060
0.210***
-0.050
0.581***
-0.050
0.301***
-0.040
-0.081
-0.060
1.287***
-0.030
-0.808***
-0.100
0.000
3300
0.274**
-0.120
0.681***
-0.100
1.887***
-0.110
0.770***
-0.090
-0.238*
-0.120
3.437***
-0.070
-1.436***
-0.200
0.000
3300
Location variables (district dummies)
Zomba
Chiradzulu
Machinga
Kasungu
Lilongwe
Error from adoption equation
Constant
Prob > chi2
Number of household (plot) observations
0.330
-0.240
0.000
3300
Page 16 of 20
Conclusions and implications
Weather extremes especially recurrent droughts threaten agricultural productivity and food
security in Malawi whose population largely depend on maize for food. Drought tolerant maize
is one promising technology to minimise the grinding impact of drought. In recent times several
drought tolerant maize varieties have been developed by national research institutions in
collaboration with CIMMYT researchers and have been disseminated across the country.
Examining determinants of adoption of this promising technology is increasingly becoming
important. Following the work of Holden and Fisher (2015) and Fisher et al. (2015) this paper
has used correlated random effects probit and tobit models with control function approach to
understand adoption of DT maize in Malawi under rainfall stress. The data is from farm
households in six districts collected in 2009, 2012 and 2015 using a sample size of 376 in 2009
and 350 for 2012 and 2015. The year 2009 is used as control since no serious drought shock was
reported.
Holden and Fisher (2015) reported a substantial increase in adoption of DT maize from 2006 to
2012, and this study also finds a significant increase from 46% in 2009 to 59% in. The paper has
found strong evidence of the impact of drought on increased adoption. This implies that farmers
learn from exposure to drought and respond by adopting risk reducing technologies such as DT
maize varieties. Farmers in drought prone districts such as Machinga are more likely to adopt DT
maize varieties than their counterparts in districts with high and stable rainfall such as Thyolo.
Lagged variables of drought are however associated with less likelihood of adoption. This could
be due to the fact that the years preceding the surveys were associated with normal rains such
that farmers responded by adopting less of DT maize in anticipating of similar good rains.
Farmers may thus respond by adopting more of improved hybrids as opposed to DT maize.
Another important driver of adoption also reported by Holden and Fisher (2015) is the farm input
subsidy programme. However while access to seed subsidy input increases both adoption and
intensity of adoption, the lagged variable of access to seed subsidy has a negative impact on
adoption. This suggests that the increased adoption is due to the availability of cheap seed as
opposed to farmers’ previous exposure with the drought tolerant maize variety. This may
Page 17 of 20
indicate limited awareness on the benefits of drought tolerant maize varieties. This is also
consistent with extension visits positively affecting adoption.
The understanding that farmers respond to exposure to weather shocks is an important
observation to maize seed breeders, agricultural extension workers and other development
partners to further promote the climate risk reducing technologies. Promotion of technologies
which are perceived by farmers themselves as climate-smart based on their experience are more
likely to receive high adoption rates and make an impact to the general livelihood. In Malawi
with FISP contributing significantly to the adoption, extension messages should be intensified
with empirical evidence so that farmers can continue using the DT seed even after FISP. It is
imperative however to understand that farmers in Malawi respond more to early droughts which
acts as early warning by adopting more of early maturing varieties. Breeders should thus respond
by breeding and disseminating more early maturing DT maize as opposed to late maturing DT
maize seed. More importantly good extension messages and promotion of drought tolerant maize
varieties should be improved to allow farmers make informed decisions.
References
Arslan, A., McCarthy, N., Lipper, L., Asfaw, S., & Cattaneo, A. (2014). Adoption and intensity of
adoption of conservation farming practices in Zambia. Agriculture, Ecosystems & Environment,
187, 72-86. doi: 10.1016/j.agee.2013.08.017
Asfaw, S., McCarthy, N., Lipper, L., Arslan, A., Cattaneo, A., & Kachulu, M. (2014). Climate variability,
adaptation strategies and food security in Malawi. ESA Working Paper No. 14-08. Rome, FAO.
Chamberlain, G. (1980). Analysis of Covariance with Qualitative Data. The Review of Economic Studies,
225-238.
Davies, M., Guenther, B., Leavy, J., Mitchell, T., & Tanner, T. (2009). Climate change adaptation,
disaster risk reduction and social protection: complementary roles in agriculture and rural
growth? IDS Working Papers, 2009(320), 01-37.
Devereux, S. (2007). The impact of droughts and floods on food security and policy options to alleviate
negative effects. Agricultural Economics, 37(s1), 47-58.
Ding, Y., Schoengold, K., & Tadesse, T. (2009). The Impact of Weather Extremes on Agricultural
Production Methods: Does drought increase adoption of conservation tillage practices? Journal of
Agricultural and Resource Economics, 395-411.
Doss, C. R., & Doss, C. (2006). Analyzing Technology Adoption, Challenges and limitations of micro
studies. Agricultural economics, 34, 207-219.
Page 18 of 20
Feder, G., Lau, L. J., Lin, J. Y., & Luo, X. (1990). The relationship between credit and productivity in
Chinese agriculture: A microeconomic model of disequilibrium. American Journal of
Agricultural Economics, 1151-1157.
Fisher, M., Abate, T., Lunduka, R. W., Asnake, W., Alemayehu, Y., & Madulu, R. B. (2015). Drought
tolerant maize for farmer adaptation to drought in sub-Saharan Africa: Determinants of adoption
in eastern and southern Africa. Climatic Change, 133(2), 283-299.
Garrity, D. P., Akinnifesi, F. K., Ajayi, O. C., Weldesemayat, S. G., Mowo, J. G., Kalinganire, A., . . .
Bayala, J. (2010). Evergreen Agriculture: a robust approach to sustainable food security in Africa.
Food security, 2(3), 197-214.
Holden, S., & Fischer, M. (2015). Can Adoption of Improved Maize Varieties Help Smallholder Farmers
Adapt to Drought? Evidence from Malawi: Centre for Land Tenure Studies, Norwegian
University of Life Sciences.
Holden, S., & Lunduka, R. (2012). Do fertilizer subsidies crowd out organic manures? The case of
Malawi. Agricultural Economics, 43(3), 303-314.
Holden, S., & Mangisoni, J. (2013). Input subsidies and improved maize varieties in Malawi: What can
we learn from the impacts in a drought year? : Centre for Land Tenure Studies, Norwegian
University of Life Sciences.
Holden, S., & Quiggin, J. (2015). Climate risk and state-contingent technology adoption: The role of risk
preferences and probability weighting.
Holden, S. T., & Fisher, M. (2015). Subsidies promote use of drought tolerant maize varieties despite
variable yield performance under smallholder environments in Malawi. Food Security, 7(6),
1225-1238.
Kassie, M., Teklewold, H., Jaleta, M., Marenya, P., & Erenstein, O. (2015). Understanding the adoption
of a portfolio of sustainable intensification practices in eastern and southern Africa. Land Use
Policy, 42, 400-411.
Kassie, M., Teklewold, H., Marenya, P., Jaleta, M., & Erenstein, O. (2014). Production Risks and Food
Security under Alternative Technology Choices in Malawi: Application of a Multinomial
Endogenous Switching Regression. Journal of Agricultural Economics.
Kassie, M., Yesuf, M., & Köhlin, G. (2009). The role of production risk in sustainable land-management
technology adoption in the Ethiopian Highlands. rapport nr.: Working Papers in Economics 407.
Katengeza, S. P., Mangisoni, J. H., Kassie, G. T., Sutcliffe, C., Langyintuo, A., La Rovere, R., &
Mwangi, W. (2012). Drivers of improved maize variety adoption in drought prone areas of
Malawi. Journal of Development and Agricultural Economics 4(14), 393-403.
Kato, E., Ringler, C., Yesuf, M., & Bryan, E. (2011). Soil and water conservation technologies: a buffer
against production risk in the face of climate change? Insights from the Nile basin in Ethiopia.
Agricultural Economics, 42(5), 593-604.
Lipper, L., Thornton, P., Campbell, B. M., Baedeker, T., Braimoh, A., Bwalya, M., . . . Henry, K. (2014).
Climate-smart agriculture for food security. Nature Climate Change, 4(12), 1068-1072.
Page 19 of 20
Magorokosho, C., Vivek, B., & MacRobert, J. (2009). Characterization of maize germplasm grown in
eastern and southern Africa: Results of the 2008 regional trials coordinated by CIMMYT.
Mangisoni, J. H., Katengeza, S., Langyintuo, A., Rovere, R., & Mwangi, W. (2011). Characterization of
maize producing households in Balaka and Mangochi Districts in Malawi. Country Report–
Malawi. Nairobi: CIMMYT.
Nangoma, E. (2007). National adaptation strategy to climate change impacts: A case study of Malawi.
Human Development. UNDP. Report, 2008.
Pangapanga, P. I., Jumbe, C. B., Kanyanda, S., & Thangalimodzi, L. (2012). Unravelling strategic
choices towards droughts and floods' adaptation in Southern Malawi. International Journal of
Disaster Risk Reduction, 2, 57-66.
Pauw, K., Thurlow, J., Bachu, M., & Van Seventer, D. E. (2011). The economic costs of extreme weather
events: a hydrometeorological CGE analysis for Malawi. Environment and Development
Economics, 16(02), 177-198.
Pauw, K., Thurlow, J., & van Seventer, D. (2010). Droughts and floods in Malawi: assessing the
economywide effects: International Food Policy Research Institute (IFPRI).
Wooldridge, J. M. (2009). New Developments in Econometrics, Lecture 6: Non-linear Panel Data
Models. Cemmap Lectures. University College London. Retrieved from
http://www.cemmap.ac.uk/resources/imbens wooldridge/slides 6.pdf
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data: MIT press.
Wooldridge, J. M. (2014). Introduction to Econometrics (Europe, Middle East & Africa ed.): Cengage
Learning EMEA, Hampshire, UK
Page 20 of 20