Académique Documents
Professionnel Documents
Culture Documents
net/publication/321018364
CITATIONS READS
0 514
1 author:
Edmund Yeboah
Technische Universität München
2 PUBLICATIONS 0 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Edmund Yeboah on 12 November 2017.
Thesis
Yeboah Edmund
2
Table of Contents
Introduction ................................................................................................................ 6
Review of Existing Literature...................................................................................... 8
Understanding the Concept of Innovation ........................................................... 8
Household/Individual Factors......................................................................16
Institutional Factors.....................................................................................17
3
Impact Evaluation Strategies .............................................................................27
Methodology .............................................................................................................27
Problem Statement ............................................................................................28
Hypothesis ..................................................................................................28
Results .....................................................................................................................45
Tests of Endogeneity .........................................................................................46
Discussion ................................................................................................................48
Conclusion and Recommendations...........................................................................50
Appendix...................................................................................................................51
References ...............................................................................................................52
4
List of Tables and Figures
List of Tables
Table 1: Reasons for School absenteeism categorized ....................................................32
Table 2: Comparison of differences in outcome measures comparing OLS and IV methods
taken from (Baser,2009). .................................................................................................38
Table 3: Innovation by Gender .........................................................................................40
Table 4: Innovation by level of education .........................................................................41
Table 5:Descriptive statistics of adopters and non-adopters ............................................42
Table 6:Estimated coefficients and standard errors for household school absenteeism
among primary school-aged children in Tanzania ............................................................45
Table 7: Results of instrument strength test .....................................................................47
Table 8: How children commute to school........................................................................51
Table 9: Why child was absent from school. ....................................................................51
List of Figures
Figure 1. The bell-shaped frequency curve and the s-shaped cumulative curve for an
adopter distribution. (Rogers, 2003) .................................................................................13
Figure 2. Adopter categorization on the basis of innovativeness (Rogers, 2003) .............13
Figure 3: Illustration of instrumental variable approach modified from (Okui, Small, Tan, &
Robins, 2012) ..................................................................................................................36
Figure 5. Mean of household school absenteeism by gender of household head.............43
Figure 6. Mean of household school absenteeism by level of education of household head
........................................................................................................................................43
Figure 7. Mean of household school absenteeism by location of household ....................44
Figure 8.Mean of household school absenteeism by farm holding of household ..............44
5
Introduction
The agricultural sector in Tanzania employs 75% of the population and contributes 26% to
the national economy (Chongela, 2015). Farming in Tanzania is dominated by traditional
farming methods and techniques such as the use of traditional (low yielding) seeds, less
use of fertilizers and pesticides, reliance on rains instead irrigation, slash and burn among
others (Beyene & Kassie, 2015; Lotter, 2015). It is increasingly becoming difficult to sustain
these traditional methods of farming in the face of increasing population with high demand
for food (Lotter, 2015). Drought and food insecurity persist in many parts of rural Tanzania
(Lotter, 2015). Lotter (2015) observed in the region of Dodoma (Tanzania) that food stress
due to low productivity in this region weakens the strength of people’s immune system to
combat death-causing diseases that could easily be prevented with good nutrition. His point
is to buttress the need for strategies that can improve food productivity, reduce poverty and
hunger. One such strategy is the promotion of innovation in agricluture. Agricultural
innovations are new methods, techniques and technologies that produce results that are
superioir to that of their predecessors (Diederen, van Meijl, Wolters, & Bijak, 2003).
Agricultural innovation 1 is a tool that has been widely used to improve food security,
eradicate hunger and alleviate poverty and through increased productivity and efficient use
of resources (more yield per unit of land) (Verkaart, Munyua, Mausch, & Michler, 2017).
Innovation is regarded as a vital step that gives farmer’s competitive edge in a changing
and challenging agricultural scene. It is seen as the main driver of productivity (Läpple,
Renwick, & Thorne, 2015). Different studies have found positive association between
agricultural innovation and different household welfare measures such as consumption
expenditure, crop income, own crop consumption, household consumption, nutritional
variation, poverty alleviation, household food security and educational improvements (Bezu,
Kassie, Shiferaw, & Ricker-Gilbert, 2014; Khonje, Manda, Alene, & Kassie, 2015; Sanginga,
Adesina, Manyong, Otite, & Dashiell, 1999). In the area of educational improvements, most
studies focus on school enrolments (also referred to as attendance in some studies) and
completion (Admassie, 2003; Beegle, Dehejia, & Gatti, 2009; Beegle, Dehejia, Gatti, &
Krutikova, 2008; Burke & Beegle, 2004; Henley, McAlpine, Mueller, & Vetter, 2010; Jewitt
& Ryley, 2014). There is, however, relatively low attention given to school absenteeism in
general (Epstein & Sheldon, 2002) and the impact of innovation on school absenteeism.
Early monitoring of school absenteeism is improtant for maintaing the enrolled students and
reducing dropping out of school (Epstein & Sheldon, 2002). The focus of this thesis is the
effect of innovation on school absenteeism among primary school children in a household.
7
To uncover the relationship between school absenteeism and innovation, I will use the
cross-sectional national survey data from Tanzania collected between 2010 and 2011
farming season. School absenteeism will be analyzed on household level considering the
proportion of children absent in school to the total number of children in the household who
attends primary school. I treated innovation as a composite variable that includes the use
of at least an improved seed variety, fertilizer (organic or inorganic), pesticide/herbicide and
irrigation technique. The rest of the thesis is structured as follows: the next section reviews
the existing literature on agricultural innovation and school absenteeism, followed by the
method used in analyzing my work. Section four describes the data I used. Section five
outlines the main results of the study. Section six is where I discuss my results. In the last
section, I conclude and provide recommendations for policy makers based on my results.
This section is organized into different sub-sections. The first sub-section discusses the
definition of innovation, innovation-decision processes, attributes of innovation and
categories of innovation adopters. The sub-section two to four briefly reviews findings of
different studies conducted on agricultural innovation, highlighting the determiners and
impacts of innovation particularly in Tanzania. The fifth sub-section highlights on school
participation and absenteeism. Sub-section six focuses on the link between innovation,
education and child labour and the last section briefly describes impact evaluation
strategies.
Innovation Definition
Agricultural innovations are new methods, technologies and techniques of farming (Wheeler
et al., 2016). The world bank considers innovation as a system and defines innovation
system as “a network of organizations, enterprises, and individuals focused on bringing new
products, new processes and new forms of organizations into social and economic use,
together with the institutions and policies that affect their behaviour and performance”
(World Bank, 2006). Innovation differs from place to place based on needs(Rogers, 2003).
Innovation brings technological progression in agriculture (Diederen et al., 2003)
8
Innovation Process
Innovation adoption is not a one-time event. It is a process over time that also involves
series of actions that flows in a sequence. For one to adopt an innovation, it usually entails
these phases:
• Knowledge
• Persuasion
• Decision
• Implementation
• Confirmation
In the knowledge phase, the individual encounters the innovation and learns how it
functions. The relationship between knowledge of innovation and the need for innovation is
explained differently. Whiles some scholars claim that the knowledge phase is a passive
process; in that knowledge of innovation evoke the need of that innovation. Other scholars
use the theory of selective exposure to argue that individual with need for innovation actively
search for the knowledge that can solve their problem ( check Rogers, 2003). Rogers (2003,
p. 169) states that “early knowers of innovation have more education, higher social status,
more agent contact, more exposure to mass media channels and interpersonal channels of
communication than late knowers. However, knowledge of innovation does not necessarily
result in adoption of innovation even though innovation knowledge and innovation adoption
share the same individual characteristics (more education and more social status) (Rogers,
2003).
The second phase of the innovation process is persuasion. Rogers (2003) defines
persuasion not as the conventional manipulation or convincing an individual but rather an
individual influencing him/herself after being aware of the favourable and the unfavourable
components of an innovation. Relative advantage, compatibility, and complexity are some
of the attributes that shapes individual’s perception about an innovation. At this stage, the
individual simulates the outcomes of the innovation and tries to predict the future situation
with the innovation along with the uncertainties. Rogers (2003) suggest that peer advice is
vital at this stage. Future innovators value the experience and advice of already innovators
more than what might be in the science journals. This stage creates an impression on the
individual but might not lead to adoption decision.
The decision phase in the decision-making process describes the willingness to adopt an
innovation. Uncertainty remains a key factor in determining adoption decision (Ghadim &
Pannell, 1999; Rogers, 2003). One way to overcome uncertainty is adopting an innovation
progressively. Innovations that can be trialled on a small scale stands a higher chance of
adoption than those innovations without any trials. Farmers obtain knowledge about the
9
profitability of an innovation during the trial period which reduces uncertainty and increases
their willingness to adopt (Ghadim & Pannell, 1999). Opinion leaders in the society are
pivotal in influencing the decision of others at this stage of the innovation process. At the
end of the decision phase, innovation is adopted or rejected. The innovation might be
rejected actively or passively. Active rejecters consider the innovation, might take part in
trials and decide not to adopt it. Passive rejecters do not consider using the innovation at
all.
The decision to adopt might be immediately implemented or held back for implementation
later due to logistic problems. Technical difficulties and problems with innovation manifest
itself at this phase, Rogers (2003) suggests that change agents need to provide information
and technical assistance to help adopters at this stage.
The last phase of the innovation decision-making process is confirmation. Implementation
phase is not the end phase as shown by researchers. Adopters seek for further information
after implementation to strengthen their position. Discontinuance is as important an issue
in innovation as the initial adoption (Rogers, 2003). Rogers (2003) identifies two types of
discontinuances: (1) replacement and (2) disenchantment. New innovations with better
characteristics replace old ones and result in discontinuance of the old innovation.
Disenchantment is when unsatisfied adopters abandon an adopted innovation due to low
performance or low relative advantage over the predecessor.
Attributes of Innovation
Rogers (2003) shows that innovations differ in nature and therefore it is difficult to obtain a
set of standardized attributes that fit for all innovations. Nonetheless, the author combed
through innovation literature to come out with five attributes that might run into each
empirically but each can stand as a conceptually distinct attribute describing innovation.
The five attributes listed by Rogers (2003) include:
• relative advantage
• compatibility
• complexity
• trialability
• observability
10
observe that farmers considered marginal gains in deciding between improved and
traditional maize varieties.
Compatibility is defined by (Rogers, 2003, p. 223) as “the degree to which an innovation is
perceived as consistent with the existing values, past experiences, and needs of potential
adopters”. Compatibility also depend on the socio-cultural factors, existing ideas and needs
of the adopter.
Complexity describes how easy or difficult an innovation is perceived to be emulated.
Rogers (2003) suggest a negative association between complexity and rate of innovation
adoption.
Trialability was mentioned before in discussing the innovation-decision process. It is the
ability of an innovation to be replicated on a small scale. The easier the innovation is to be
trialled, the higher the likelihood of adoption (Rogers, 2003). Farmers expectation about the
profitability of an innovation is verified in the trial stage. The ability to trial an innovation
provides farmers the platform to gather knowledge about the innovation which boost their
adoption decision. Presumably, ‘hard’ 2
irrigation technology, which requires costly
equipment and cannot be adopted progressively face high rejection compared to improved
maize varieties which farmers can trial on a small parcel of land. Those who do not take
part in the trial, decide based on the outcome of participants (Rogers, 2003).
Observability is the extent to which the outcome of an innovation can be observed by others.
Individuals base their adoption decision on what they have observed on trial fields as well
as the outcome of other adopters. Rogers (2003) states the observability affects the rate of
adoption in a social system.
Rate of Innovation
The rate of innovation indicates how long it takes for innovation to diffuse in a social system.
It is defined as the speed of innovation adoption in a social group (Rogers, 2003). The
speed and extent of innovation determines the impact of the innovation (Diederen et al.,
2003). Diederen et al. (2003) identify two approaches in explaining the diffusion of innova-
tion. The first approach is regarded as a disequilibrium approach where lack of transpar-
ency, imperfect information and 'uncertainty about the operating conditions, risk and perfor-
mance' of innovation explains why not all potential adopters jump at once on board the
innovation train. The number of adopters of an innovation increases or decreases as infor-
mation generated after its implementation spreads. The second approach is regarded as
an equilibrium approach where the variation of adoption benefits dictates the diffusion of
innovation. The benefits of innovation differ per adopter’s structural characteristics such as
3Innovators as used here is different from as used in other sections of this paper. In other sections,
subjects are divided into two groups. Those who has adopted any form of innovation are called in-
novators and non-innovators otherwise.
12
Figure 1. The bell-shaped frequency curve and the s-shaped cumulative curve for an adopter distribution.
(Rogers, 2003)
Fertilizer Application
The use fertilizer can increase crop yield in Africa by 64% (Vanlauwe et al., 2014). Both
organic and synthetic fertilizers increase crop yield (Lotter, 2015). Synthetic fertilizers
produce instantaneous results, whiles the benefits of organic fertilizers might not be
instantaneous but have a long lasting and healthy effect on plants, soil macro and micro-
organisms as well the environment. Chemical residues, potential poisoning of ground water
by synthetic fertilizers present challenges that must be dealt with (Lotter, 2015). However,
Lotter (2015) argues that soils in Africa are in deplorable state and would require a lot of
time for yield to respond to only organic fertilizers. Therefore, despite the negative effects
of synthetic fertilizers, he claims that the rapid increase in productivity, which reduces
hunger and improves food security, outweighs the negative effect of inorganic fertilizer use
in Africa. Vanlauwe et al. (2014) present a synergistic relationship between the use of
organic and synthetic fertilizers in Africa. They point out that, for the benefit of synthetic
fertilizer to be fully realised, the depleting organic matter in African soils needs to be
replenished. Organic matter binds the soil together to hold up the synthetic fertilizers to
provide a good yield response (Vanlauwe et al., 2014). Lotter (2015) claims that both
synthetic and organic fertilizer are needed to ensure food security in Tanzania. He debunks
the outright avoidance of synthetic fertilizers, per him, the recommended 60kg/ha of
elemental nitrogen (in Africa) will not poison the soil. Chemical fertilizer increases yield
whether the variety is traditional or improved (Kajisa & Nakano, 2012). Kajisa and Nakano
(2012) found that the use of chemical fertilizer was beneficial to both adopters and non-
adopters of modern rice varieties in Tanzania.
14
Irrigation Innovation
Irrigation innovations are classified into ‘hard’ and ‘soft’ innovation (Wheeler et al., 2016).
Hard irrigation innovation involves building irrigation infrastructures whiles soft irrigation
innovation includes “change in skills and management (e.g. alternate wetting and drying
techniques)” (Wheeler et al., 2016, p. 6). Examples of hard irrigation found in Tanzania
include pumping surface-water, pumping groundwater, water buckets or other devices, and
run-off harvesting. The soft irrigation innovation include growing cover crops, mulching, crop
rotation, planting leguminous crops, alternate crop mix, use of charcoal and wood to control
run-off (Wheeler et al., 2016). Kajisa and Nakano (2012) showed that irrigation was very
important in realizing the improvements in modern rice varieties in Tanzania.
Herbicide/Pesticide Innovation
Lotter (2015) remarks that the perennial extensive rooted grasses in Africa can be
effectively cleared either by tillage or the use of herbicides. Tillage destroys the soil structure
and exposes it to erosion. Herbicides protects the soil from continuous disturbances such
as soil erosion, it promotes zero tillage and eliminates the need for slash-and-burn that
results in loss of soil nutrients. The use of herbicides save farmers from the back-breaking
act of weeding (for clearing the land) predominant in many sub-Saharan African countries.
The use of herbicides also increase rain water capture up to five folds (Lotter, 2015).
Innovation Packages
The innovations listed above may be used in combination with one another. Kaliba et al.
(2000) and other studies present evidence for this observation in their study. For example,
Kaliba et al. (2000) found IMVs and fertilizer as ‘innovation package’ to some extent. Their
complementarity depends on the land holding of the adopter. Adopters with small land
holdings view the two as complimentary since input efficiency is their only alternative to
improve their yield, whereas farmers with large land holdings use less fertilizer as they
increase the land size for IMVs (Kaliba et al., 2000; Smale, Just, & Leathers, 1994). Kajisa
and Nakano (2012) observed that modern rice varieties alone had limited impact on income
and productivity. Without water management techniques, adopters of modern rice varieties
even with irrigation experienced no positive impact on their income. However, modern rice
in combination fertilizer or bund (water management technique) increased yield and income.
This support the argument that agricultural innovations rarely function in isolation but rather
as packages. For this reason, I treated innovation as a composite binary measure which
indicates household adoption of one or more of the innovations discussed.
15
Determinants of Adoption of Agricultural Innovation
To promote innovation adoption, researchers have carried out different studies to identify
the factors that affect the adoption of innovation. The determinants of innovation may be
stepping stones or stumbling blocks to innovation. Here, I discuss the different determinants
of innovation especially as found in research work in Tanzania. Determinants are organised
into household/individual factors, institutional factors, and innovation factors.
Household/Individual Factors
Household and individual factors such as age, education level, farm holding, income,
household size, farmers experience among other factors determines the adoption decision
of farmers towards an innovation (Ghadim & Pannell, 1999). As reported by Kafle (2010),
older farmers are more likely to clinch to their traditional methods explaining the negative
association between age and innovation. Etoundi and Dia (2008) on the contrary found
positive and significant association between age and the adoption of some maize varieties.
A year increase in education results in 8 times more likelihood to adopt IMVs in the Hai
district of Tanzania as shown by Mmbando and Baiyegunhi (2016) using a logistic
regression method. The positive impact of education on adoption was explained by Kafle
(2010) as due to the ease of comprehending new technologies among educated household
heads. Fisher et al. (2015) using a multinominal logit model found a similar positive effect
of years of education on innovation adoption, attributing it to the ability of educated farmers
to effectively and quickly assimilate information on new technologies. Amare, Asfaw, and
Shiferaw (2012) using a double-hurdle model found that the years of schooling of the
household head had a positive effect on adoption of IMVs and pigeon peas up to 5 years.
After 5 years, there found a negative effect, which they attributed to off-farm activities of
educated households. They said educated farmers may have other jobs aside farming. Off-
farm activities mean less time for farming activities, which affects farmer’s willingness to
invest in innovations. Farmers’ experience with an earlier innovation also plays a role in
subsequent innovation adoption. Kathage, Kassie, Shiferaw, and Qaim (2015) found that
farmer’s experience in improved maize cultivation had a negative effect on the likelihood of
adoption in the East of Tanzania whiles it is insignificant in the North. This contradicts the
general notion that farmers who have had exposure with improved maize are likely to adopt
new varieties compared to inexperienced farmers (Kathage et al., 2015). The contradiction,
per the authors, was due to the low returns on hybrid maize encountered by farmers who
cultivated it. Large land holding meant more allocation for improved maize varieties
(Ransom, 2003). Kathage et al. (2015) found a non-significant positive effect of land holding
on adoption of hybrid maize in Tanzania. Whiles, Etoundi and Dia (2008) found that high
labour intensity for improved maize production hinders large holders of land from switching
16
from traditional to improved varieties. Kaliba et al. (2000) found that the amount of land
allocated to improved maize negatively affected the adoption of inorganic fertilizer.
Verkaart et al. (2017) observed that large land owners were more diversified resulting in
less adoption of chick pea per land allocation compared to small land owners (Verkaart et
al., 2017). Wealth, income (both from farm and off-farm) and livestock holding also
determine the adoption of innovation. Kaliba et al. (2000) reported that farmer’s wealth had
a negative impact of fertilizer adoption. This observation reflects the risk behaviour of poor
and rich farmers. The authors explain that “poor farmers use input intensification ( use more
fertlisers and improved seeds that can increase productivity on the same piece of land) to
manage risk whereas relatively rich farmers use extensive production and diversification
(e.g. increase in the farm holding and labour) for the same effect” (Kaliba et al., 2000, p. 44).
Mmbando and Baiyegunhi (2016) identified off-farm income as one of the factors that has
a positive influence on improved maize adoption. Farmers with off-farm income are three
times more likely to adopt improved maize than those without off-farm income. They
explained that farmers with off-farm income can afford the required investment in innovation
(seeds and fertilizers) than their colleagues with no additional income. Livestock holding
has a positive effect on improved maize adoption through more household income and the
supply of organic manure (Kafle, 2010; Kassie et al., 2014). Kassie et al. (2014) found that
livestock ownership increase the likelihood of improved maize adoption, because,
households with livestock can handle the risk of adoption. Also, Livestock holders can sell
their livestock to purchase inputs for innovation (Kassie et al., 2014). Conversely, Alumira
and Rusike (2005) in Zimbabwe, reported negative effect of livestock holding on adoption
of IMV. They attributed their observation to the disperse and spatial cultivation among draft
cattle-owning households and their inability to afford IMV seeds. Kathage et al. (2015) found
no significant effect of assert ownership on adoption of hybrid maize in Tanzania.
Institutional Factors
Determinants that will be treated under institutional factors include access to extension,
awareness, group participation, credit and market access. These determiners are
innovation enabling environments that can be improved by institutions that promote
innovation.
Agricultural extension is the medium through which information is carried from researchers
and experts to farmers. Extension work is useful in overcoming barriers in innovation faced
by farmers and gathering feedback back to researchers to improve innovation (Wheeler et
al., 2016). In the Hai district of Tanzania, Mmbando and Baiyegunhi (2016) found that
access to extension increase the likelihood of adopting IMVs over six times compared to
those who had no access to extension. Kaliba et al. (2000) also found intensity of extension
17
services as the main factor that influence adoption of synthetic fertilizers in Tanzania. To
avoid incomplete information that undermines adoption, Beyene and Kassie (2015)
recommended that extension agents (whose work have positive impact on the speed of
adoption) must be trained and equipped to improve adoption of IMVs. Wheeler et al. (2016)
reviewed 53 articles on irrigation innovation and extension. 60% of the articles found
positive association between extension and irrigation adoption, 34% found no effect whiles
6% found negative association. Their results did not differ between developed and
developing countries. Banful, Nkonya, and Oboh (2010) studied that impact of extension on
fertilizer use in Nigeria. They found that limited extension services affected the adoption of
fertilizer. The farmers pointed out inadequate extension information on application of
fertilizers as a hindrance to their adoption decision. Kathage et al. (2015) found that the
yield benefit of IMVs in the north of Tanzania sped up the awareness of IMVs. Fisher et al.
(2015) recommends that information about adoption must be communicated clearly,
resolving negative perceptions and highlighting the positive attributes. The relevance of
awareness in innovation adoption also differs with different innovations. Easy to use
innovations spread quickly, whilst knowledge-intensive innovation may require information
improvement strategies beyond farmer-to-farmer transfer (Fisher et al., 2015).
Barrantes and Yagüe (2015) emphasize the effectiveness of social learning and the use
small groups in Peru. In the presence of inadequate information, high transaction cost and
imperfect markets, farmer groups play a key role in disseminating information on inputs and
credit availability (Kassie et al., 2014). Mmbando and Baiyegunhi (2016) found that farmers
with membership in groups are two times more likely to adopt IMVs compared to those with
no membership in groups. Mmbando and Baiyegunhi (2016) also found that participation in
field trials and experiments increase the likelihood to adopt IMVs four times. This is because
farmers who participate in such trials are more aware and knowledgeable with IMVs
compared to their counterparts without such an exposure (Mmbando & Baiyegunhi, 2016).
In Niger, Abdoulaye and Sanders (2005) observed that field trials ranked as the top
determinant of fertilizer use. Farmers want to see the results of the innovation before they
buy into the idea (Abdoulaye & Sanders, 2005). Beyene and Kassie (2015) highly
recommend that local institutions must be promoted, strengthened and supported to aid
innovation adoption.
Credit was found as an important determiner of IMVs adoption in Ethiopia (Feleke
& Zegeye, 2006). Mmbando and Baiyegunhi (2016) found a similar positive effect of credit
on adoption of IMVs. They reported that farmers with self-reported access to credit are 3
times more likely to adopt IMVs compared to those who reported no credit access. On the
contrary, Kathage et al. (2015) found credit to be not significant in all their models in
determining the adoption of hybrid maize in Tanzania, based on which they challenge the
18
notion that non-adoption is a result of market failures relating to rural finance. They assert
that farmers face less issues with saving for seed costs and therefore may not find access
to credit as a challenge in adopting hybrids because already they purchase some of the
traditional seeds they use. Ransom (2003) found that the nearer a plot is to the market, the
higher the likelihood of the farmer adopting IMVs. “Poor market access increases production
costs and reduces profits, and therefore, is expected to result in reduced adoption of new
agricultural technologies” (Fisher et al., 2015, p. 296). Ransom (2003) found a positive
relationship between distance to market and IMVs adoption in Nepal. Fisher et al. (2015)
point out the labour intensity in cultivating IMVs. They found availability of labour as a key
determiner for sustained cultivation of drought tolerant maize.
Innovation Factors
Difficulty in accessing IMVs hinders farmers adoption decision (Amare et al., 2012).
Difficulty in acquiring drought tolerant maize seeds in Tanzania, Ethiopia, Zambia and
Uganda had a negative effect on its adoption (Fisher et al., 2015). Apart from availability,
the cost of the seed also deters farmers from adopting it. Fisher et al. (2015) showed that
farmer’s inability to switch from traditional varieties to drought tolerant maize in Tanzania,
Malawi and Uganda was due to high seed cost. However, Kathage et al. (2015) argue that
seed cost is a mask effect for low return on adoption as they found in their work. They argue
that not all seeds used by the farmers who cultivate traditional varieties are from their own
farm. The farmers buy some of the traditional seeds and the main reason that will motivate
farmers to invest in improved seeds, is when the return on the investment is high.
Agro-ecological characteristics (such as rainfall pattern and soil fertility) also determine the
adoption of innovation. Comparing maize yield between the north(highlands) and the east
(lowlands) of Tanzania, Kathage et al. (2015) found that there was substantial difference
between the yield of IMVs and traditional varieties in the North whiles the difference is
insignificant in the east. They attributed this to differences in agro-ecological conditions of
the two areas. This further explains why the adoption gap (difference between adopters and
non-adopters) is higher in the east (-0.23) compared to the North (-0.08). . Kaliba et al.
(2000) found that farmers in high rainfall area (compared to those in low rainfall) are more
likely to adopt IMVs. Fisher et al. (2015) also found that farmers at lower altitude were more
likely to adopt drought tolerant maize than those at higher altitude. Beyene and Kassie
(2015) noted that in addition to rainfall patterns, soil fertility also increase the speed of
improved maize adoption in Tanzania. Kaliba et al. (2000) expected use of inorganic
fertilizers to be higher in high rainfall areas but they found that intensive extension programs
in lowlands (with low rainfall) may have influenced high inorganic fertilizer use in lowland
areas they studied.
19
Impact of Adoption of Agricultural Innovation
The aim of innovation is to solve a problem, fulfil a need and therefore creates an impact.
An impact that will be felt individually by households, socially by the community and broadly
by the country. Innovation impact assessment steers research, informs policy and
determines the allocation of public investment (Maredia, Shankar, Kelley, & Stevenson,
2014).
20
methods (propensity score matching, continuous treatment effects and binary propensity
score matching) the authors used, positive adoption impact was obtained across the
different methods.
Amare et al. (2012) showed that consumption expenditure for improved maize and pigeon
pea adopters was 15% and 22% higher, respectively, compared to non-adopters with
propensity score matching techniques. Using the endogenous switching technique, Amare
et al. (2012) found that the mean of consumption expenditure was 120% higher for maize
adopters and 95% higher for pigeon pea adopters compared to non-adopters. They
demonstrated that “the consumption expenditure per capita of improved maize and pigeon
pea non-adopters would have been increased by 163% and 40%, respectively, if they
adopted the technology” Amare et al. (2012, p. 40). The evidence presented supports
investment in innovation as a remedy for food insecurity.
21
chemical fertilizers increased yield significantly irrespective of the variety (whether
traditional or modern). IMVs increased the yield of adopters in Ethiopia (Feleke & Zegeye,
2006). Larsen and Lilleør (2016) assessed that impact of agricultural interventions on the
nutrition of children in innovation households. They found that” intervention had a significant
positive impact of about 0.9 SD6 on the height-for-age z-scores of young children who had
been fully exposed to the project in their early life” (Larsen & Lilleør, 2016, 23). The
prevalence of stunting also reduced to about 18 percentage points (Larsen & Lilleør, 2016).
Return on Investment
Anandajayasekeram et al. (2007) used the cost-benefit approach to show that technology
development and transfer in maize in Tanzania accrued US$ 1.1 billion net present value
(NPV) at 5% discount rate between 1974 to 1994. The internal rate of return (IRR) as
calculated by Anandajayasekeram et al. (2007) with the same approach was 19%. An
increase in improved maize research and extension was shown to be profitable. Increasing
cost (investment in research and extension) by 20% projected an estimated IRR of 20.1%
using the Akino-Hayami method and NPV of US$ 1.6 billion at 5% discount rate. A 50%
increase in cost, however, projected a declined IRR of 16.6% and NPV of US$ 1.4 billion at
5% discount rate. Despite the projected decline, the authors asserted that it will still be a
profitable investment for society considering the opportunity cost of capital in Tanzania
(Anandajayasekeram et al., 2007).
Environmental Impact
Positive environmental impacts are positive externalities of innovation technologies. Row
planting that comes with the cultivation of IMVs controls soil erosion (Anandajayasekeram
et al., 2007). Pest and disease resistant varieties reduce the chemical burden on the soil
and its negative effects on ground water. Slash and burn systems of farming found to be
detrimental to the environment are also controlled with the use of herbicides and improved
varieties (Anandajayasekeram et al., 2007).
Child labour is a global phenomenon, in 2012, out of the 26.2% of children in sub-Saharan
Africa involved in employment, 21.7% of them were identified as child labour (Diallo,
Etienne, & Mehran, 2013). Nearly one out of every two children aged 14-17 while 70.4% of
children aged 5-17 in Tanzania are involved in economic activities, yet only 20.7% are
classified as child labour. Most of the cases are found in rural areas, where boys are
engaged in child labour more than girls (Tanzania National Bureau of Statistics, 2016). Most
of the children engaged in child labour world-wide are involved in the agricultural sector
(58.6%). The work they do mostly (68.45%) contributes to family work compared to other
sources (Diallo et al., 2013). In Tanzania, it is reported that 89.3% of the children found in
child labour are engaged in agricultural and fishery work (Tanzania National Bureau of
Statistics, 2016). Beegle et al. (2008) found in Tanzania that the “Economic activities for
children consist predominately of farming, including tending crops in the field, processing
24
crops, and tending livestock” (Beegle et al., 2008, p. 6). Also, among vulnerable school
children (children affected by abject poverty and diseased parents) in Tanzania, Anangisye
(2011) found that the majority of them were engaged in income generating activities to
support their education. Charcoal making and selling, carrying raw timber, farm work
activities, fetching water, carrying bricks, stone crushing and involvement in commercial sex
were the main income generating activities that the children were involved in. In one school
interviewed, some children had to break school during the farming season-August to
December- to raise money for their education (Anangisye, 2011). Admassie (2003) asserts
that the magnitude of child labour may be underestimated because official estimates do not
capture all the children working in unpaid domestic and agricultural activities, because of
underreporting due to the legal deterrence of using children as part of labour force. Different
views on child labour in different cultures also affect how child labour is defined and
therefore reported (Admassie, 2003). For example, Admassie (2003) noted that in rural
areas, child labour is often perceived as equipping children with basic life skills. Abid et al.
(2015) report that resource constrained household, and households with limited job
opportunities engage their children in child labour to augment household income. Admassie
(2003) found that work related issues are the main cause hindering school attendance in
Ethiopia. He remarks that “Child labour is dis-investment in human capital formation and
has a detrimental effect on the subsequent private and social returns from it” (Admassie,
2003, p. 168). As stated by Beegle et al. (2008, p. 16), “one standard deviation increase in
child labour hours (5.7 hours) is associated with a decrease of nearly half a year of schooling
and an 8.8 percentage point reduction in the chance of completing primary school”. Being
engaged in domestic or farm work differs in its impact on school attendance since some
activities are compatible with schooling whiles others are not (Admassie, 2003). However,
the time struggle between education and income generating activities results in truancy and
high dropout rate as observed by Anangisye (2011).
Methodology
The method section begins with the problem statement, the research question, hypothesis
and objective of the study. Section two continues with a background study of primary
education in Tanzania and highlights the major policy changes and developments in
Tanzania’s primary education. The next section describes the data and variables I used in
this thesis. The estimation method is described in section four.
27
Problem Statement
Agriculture modernization through different innovations have been used in Tanzania to
improve productivity, household welfare and food security. These improvements can have
both positive and negative outcomes on school attendance of primary school children. The
positive impact is expected to be reflected in low school absenteeism. For example,
increased income enables innovators to hire the needed labour instead of using children as
part of household labour (which takes them out of the classroom). Whiles, absenteeism is
expected to be high in households where child labour is needed for labour intensive
innovations.
Research Question:
What is the effect of agricultural innovation on school absenteeism among primary school
children?
Hypothesis
Agricultural innovation reduces household school absenteeism through increased
household welfare of innovation adopters.
30
Data Sources and Variables Description
The Tanzania National panel survey data is based on a stratified, multi-stage cluster sample
design that contain wide range of information on household characteristics, agricultural
activities, education, health, governance, subjective welfare, housing and sanitation, labour,
food consumption, credit information, finance information, shocks to household welfare,
death information and anthropometric information at household and community level. Data
collection was initiated in 2008 and is being repeated every two years, to provide policy
makers with a longitudinal data to track individuals/households over long period to aid in
better policy making. It currently has three waves readily available, the latest wave (wave
3) was collected in 2012/2013.The data was collected by the Tanzania government with
assistance from the World Bank. Trained enumerators were used to conduct face to face
interview in collecting the data (Tanzanian National Bureau of Statistics, 2011). I used the
second wave of the data, collected in 2010/11 farming season. My analysis was done on
household level, aggregating individual responses to household responses. The wave 2
data contain information on 3924 households selected from 26 regions and 239
enumeration areas across the country. I scaled the sample down to 2971 of households
that cultivated on at least one plot in the 2010 long rainy season.
The outcome variable is school absenteeism measured as the proportion of total children
in a household who missed school in the last two weeks before the survey. I chose the age
of 5-15 years including both pre-primary and primary school aged children. This follows the
classification of ILO that indicates children from 5 years are involved as part of household
labour especially in farming household (Tanzania National Bureau of Statistics, 2016). The
proportion of school absenteeism was constructed from a binary variable, where 1 means
the respondent was absent at least once in the two weeks before the survey and 0
represents otherwise.The reasons attributed to absenteeism include: (1) public holiday, (2)
school closed but not in break, (3) school closed in break, (4) absence of teacher, (5) illness
of child, (6) illness of household member, (7) funeral, (8) disciplinary action, (9) parent’s
inability to meet educational cost, (10) child refused to go to school, (11) child had to work
and (12) others that respondents had to specify. Public holidays, school out of session,
absence of teacher were excluded in this thesis because they are not household or child
driven reasons. According to Balfanz and Byrnes (2012),all the other reasons are worthy of
investigating because they contribute to chronic absenteeism. The reasons are further
categorised into if the child ‘cannot’, ‘will not’ or ‘do not’ attend school. A child who ‘cannot’
go to school is compelled by obligations and circumstances to be at a different place during
school time. If the child is avoiding an unpleasant situation at school such as bullying,
embrassment, inability to fulfill assigmnents requirements, in this case the child ‘will’ not to
go to school. A child who ‘do’ not want to attend school is not be hindered but based on his
31
decision or parents decision fails to attend school (Balfanz & Byrnes, 2012). Table 1 put the
the reasons of school absentation stated above into ‘cannot’, ‘will not’ and ‘do not’
categories.
Variable Description
Proportion of total school aged children in an household who
AbPs
missed school in the last two weeks before the survey
32
Table 3: Description of endogenous (Treatment) and instruments used to estimate Treatment
Variable Description
Household innovation decision used as treatment variable
Innovation
(1=adopted any of the four innovations described above, 0=
(Treatment)
adopted none of the four innovations)
Mean of village level innovation (Range: 1-0, 1=all households
Vinnov selected in the village are innovators, 0= none of the
households adopted an innovation)
Variable Expected
Description
Sign
Gender Household head gender (1 = male; 2 otherwise) +/-
Age Household head age (in years) +/-
HHEduc Household head Level of education +
FSizeh Farm size (hectares) +
HH_Size Number of people in household +
Num. child Number of children (5-15) in the household +
SchTime Average time to school for each household (in hours) -
TLU7 Livestock ownership (Total tropical livestock units) +
Education, sex, age - of household head -, household size, asset ownership (Tropical
livestock unit per household), farm size per household and average traveling time to school
per household were the covariates in the regression analysis. Female household heads are
expected to have less proportion of absenteeism compared to household with male head.
Beegle et al. (2009) stated the women are more likely to channel their resources in more
child-oriented ways such as educating their children. Al-Samarrai and Peasgood (1998)
found that number of children in a household has a positive effect on attending and
completing school. They explained that domestic and farm work will be better distributed
among many children ensuring less burden compared to household with few children.
Average traveling time is the mean of the time taken by all children in the household to
travel to school. Kana et al. (2008) found negative association between distance to school
and educational attainment. They explained that poor households could not afford vehicle
to transport their children to school resulting in less time spent in the classroom per child if
7 Tropical Livestock Units (TLU) was estimated using FAO conversion factors.
33
the distance of their village to the nearest school is far. The location of the village, whether
in a rural or an urban area, is used as an indicator of school characteristics. Schools in
urban areas tend to have better facilities and qualified teachers compared to rural areas. I
assume that households in rural areas will have higher rate school absenteeism compared
to rural areas.
Estimation Specifications
Non-experimental observational studies which infer causality often suffer from
unrandomized treatment selection of groups and unmeasured confounders (Okui et al.,
2012). In randomized controlled trials, treatment is assigned to the subjects by the
investigator. The investigator can control the experimental environment and be assured the
confidence to interpret statistically significant difference as due to the treatment (Peck,
Olsen, & Devore, 2008). Peck et al. (2008) explain that, it is difficult to draw conclusions
based on observational studies because statistical significance may be caused by
uncontrolled underlying factors instead of conditions that define the groups. Baser (2009)
points out that studies of causal inference are affected by overt and hidden biases. Overt
biases result from observed differences between treatment and control group that likely
affect the response of the treatment. Hidden bias are unobservable factors that affect the
outcome of the treatment. Observable bias can be dealt with using propensity score
matching and ordinary least square (OLS) regression analysis whiles instrumental variables
(IV) regression and switching regression models have been popularly used in dealing with
unobservable bias (Baser, 2009; Maredia et al., 2014; Wooldridge, 2010). IV regression is
deployed when there are measurement errors, simultaneity (y affects x, x affects y),
unmeasured confounders and omitted variables (Okui et al., 2012). I used IV regression in
this thesis due to its ability to control observed and unobserved differences between
individuals (Baser, 2009; Okui et al., 2012; Wooldridge, 2010). The IV method measures
the average treatment effect (ATE) of the individual treatment of the whole population of
interest (Baser, 2009). Baser (2009) asserts that any valid instrument can identify the ATE
if the effect of the treatment is the same for everyone. On the contrary, if the effects are
different, different instruments will identify different effects. In this instance, the ATE that
can be estimated by the IV methods is thus among individuals who change their behaviour
towards the treatment through the instrument, which is called local average treatment effect
(LATE). Emsley, Lunt, Pickles, and Dunn (2008) assert that measuring household casual
effects of a treatment using IV regression requires two “exchangeable” groups. The two
groups need to have identical distributional characteristics. The effects of the treatment
against the control is evaluated by comparing group A after adopting the treatment and in
the same period had they received the control (counterfactual). The counterfactual of the
34
group A cannot be observed in the same period, therefore a control group B that did not
adopt the treatment but have the same characteristics as the adopters are exchanged as
the counterfactual group to study the household causal effects (Emsley et al., 2008). The
authors remark that exchangeability is however affected by selection effects such as
missing data, nonadherence in a randomized study and confounding. “Confounding occurs
when there are variables that are risk factors for the outcome of interest, correlated to the
exposure under investigation, but which do not lie on the causal pathway between them. A
variable meeting these criteria is called a confounder” (Emsley et al., 2008, p. 335). Emsley
et al. (2008) suggest that the selection effects be adjusted and conditional exchangeability
assumed.
The equation of interest is:
The assumption is that 𝑥 ∗ must be exogenous and do not correlate with 𝑢 . Potential
correlation between 𝑢 and 𝑥 ∗ comes from many sources notably measurement errors or 𝑢
containing an unobservable variable that correlates with one of 𝑥 ∗ . I assume that 𝑥𝑗 is a
set of exogenous variable but my 𝑥𝑘 (innovation) cannot be considered as exogenous. If 𝑥𝑘
correlates with 𝑢 due to unobserved variables (Cov (𝑥𝑘 ,u)≠0), this means 𝑥𝑘 is endogenous,
therefore estimates 𝛽𝑘 produced from (1.2) will be inconsistent using OLS. However,
innovation adoption decision is likely to suffer from endogeneity (Shiferaw et al., 2014). This
is because farmer’s decision to innovate is likely to be influenced by household
characteristics which can also affect school participation in the household.
Wooldridge (2010) writes that instrumental variable method provides a general solution to
the endogenous variable problem. Baser (2009) asserts that in order to obtain consistent
estimate of 𝛽𝑘 , new information is needed. The new information comes in the form a new
variable 𝑧𝑖 . 𝑧𝑖 is referred to as the instrumental variable. For a variable to qualify as an
instrument, it must satisfy some conditions. Figure 3 below to illustrates these conditions.
The dotted pathways marked with X violates the conditions of the instrumental variable.
35
y
𝑥𝑘
𝑧𝑖 𝑥𝑗
Figure 3: Illustration of instrumental variable approach modified from (Okui, Small, Tan, &
Robins, 2012)
The first condition is that, 𝑧𝑖 must be uncorrelated with the error term 𝑢 in equation (1.2).
This is Expressed as; 𝑐𝑜𝑣(𝑧𝑖 , 𝑢) = 0 (1.4)
This implies that 𝑧𝑖 like 𝑥𝑗 must be exogenous (Wooldridge, 2010).
The second assumption is that the instrument 𝑧𝑖 “must be related, either positively or
negatively, to the treatment indicator” (Baser, 2009, p. 1202). Wooldridge (2010) explains it
with the linear projection of 𝑥𝑘 onto all the exogenous variable in equation (1.2) as well as
the instrument.
𝑥𝑘 = 𝛿0 + 𝛿1 𝑥1 + 𝛿2 𝑥2 + ⋯ + 𝛿𝑘−1 𝑥𝑘−1 + 𝜃𝑖 𝑧𝑖 + 𝑟𝑘 (1.5)
Where 𝐸(𝑟𝑘 ) = 0 and 𝑟𝑘 is uncorrelated with 𝑥𝑗 and 𝑧𝑖 .
This linear projection equation is referred to as the reduced-form equation of the
endogenous variable 𝑥𝑘 . Wooldridge (2010) asserts that the key assumption is that the
coefficient of 𝑧𝑖 (𝜃𝑖 ) should be nonzero.
𝜃≠0 (1.6)
Wooldridge (2010) interprets it as the partial correlation between 𝑥𝑘 and 𝑧𝑖 giving that 𝑥𝑗 is
‘netted out’. When these two conditions are met 𝑧𝑖 qualifies to be called an instrument of
𝑥𝑘 .
36
The first condition (𝑐𝑜𝑣(𝑧𝑖 , 𝑢) = 0) cannot be tested because the covariance in the condition
involves the unobservable 𝑢 , however, the second condition (𝜃 ≠ 0) can be tested
statistically (Wooldridge, 2010).In this thesis, I used the two-stage least squares (2SLS)
instrumental variable approach to obtain the estimates. The first stage regression obtains
the fitted values of the treatment variables from the regression of the treatment variable on
the instrument and the exogenous regressors in the estimation as described in equation
(1.5). The obtained treatment variable estimate from the first stage is used in the OLS
regression in the second stage as described in equation (1.7) (Wooldridge, 2010).
Baser (2009) highlights the difficulty of finding a credible instrument that satisfies the
conditions stated above. Innovation was instrumented using the mean of innovation per
village. Abdulai and Huffman (2005) used cumulative village/regional-level adoption
decision and distance to extension as instruments to examine the diffusion of crossbreed-
cow in Tanzania. They found that farmers learn from each other and this influences diffusion
of profitable innovation. The village level innovation in this thesis is the average innovation
value for a village. This is calculated as the total of the number of innovators divided by the
total number of farmers in the village. Farmers from the same village have the same village
level innovation value. Shiferaw et al. (2014) also used access to extension information as
an instrument in their work. However, adding access to extension (measured in this survey
by the distance of the plot from the district headquarters) as an instrument was ruled out by
the post-estimation overidentification test of my instrumental variable regression model.
Wooldridge (2010) shows that using a weak instrument produces inconsistent estimates
that are misleading. Wooldridge (2010) therefore suggests that it is better to use OLS than
to use a weak instrument in a 2SLS . According to Wooldridge (2010), 2SLS produce large
standard errors compared to OLS and a weak instrument can exaggerate the size of the
standard errors of the 2SLS estimation. Wooldridge (2010) points out the difficulty between
using OLS that produce inconsistent results if any of the variables is endogenous and the
large size of 2SLS standard errors which results in insignificant estimates. The accuracy of
the estimates produced by the IV regression depends on the size of the sample (Baser,
2009; Wooldridge, 2010). Baser (2009) asserts that small sample size may produce
inaccurate IV estimates (Baser, 2009) . Following the caution of Wooldridge (2010) on the
misuse of the IV method, I conducted the endogeneity and weak instrument test to ascertain
the need for the IV method instead of OLS or other methods.
38
explains the endogenous regressor after partialling out the effect of the exogenous
regressors. The rule of thumb as suggested by Stock and Staiger (1997) is that in the case
of one endogenous regressor, instruments can be regarded as weak if the first-stage F is
less than 10. However, Stock and Yogo (2003) clarifies that the F statistics overstates the
relevance of the instruments if the fit contains more than one endogenous regressor and
instrumental variables. The second part of the test presents the minimum eigenvalue of the
Craig-Donald statistics, critical values for the 2SLS relative bias and critical values of Stock
and Yogo (2005) second characterization of weak instruments. The 2SLS relative bias
compares the bias of the 2SLS estimator to the bias of the OLS estimator. Critical values
are produced for different bias tolerance levels (5%, 10%, 20% and 30%). A weak
instrument has lower minimum eigenvalue than the 2SLS relative bias critical value at a
selected bias acceptance level. The stata13 manual explains that the Stock and Yogo
(2005) second characterization of weak instruments “defines a set of instruments to be
weak if a Wald test at the 5% level can have an actual rejection rate of no more than 10%,
15%, 20%, or 25%” (StataCorp, 2013, p. 9).
The statistical measures of choice used were descriptive statistics and summary measures
represented in graphs and tables. Stata 13 statistical software package was used for the
analysis. Section one describes how adopters and non-adopters differ in mean of the
household characteristics. Section two compares the mean of school absenteeism for
various household characteristics. In all the analyses, P < 0.05 was used as the criterion of
statistical significance.
Table 3 categorize adopters and non-adopters of innovation per gender. 28 percent of the
households studied are headed by females compared to 72 percent headed by males.
Households that adopted innovation constitutes 40 percent of the total household. Among
the non-adopters, almost one-third of households are headed by females. 22 percent of the
households that adopted innovation are headed by females whiles 78 percent are headed
by males. Majority of the household heads studied have primary level education. Only 6 out
of 2948 household heads have attained tertiary level of education. Except for household
heads with tertiary education, non-adopters are more than adopters in all the levels of
education. In comparing adopters and non-adopters within the levels of education (row%),
half of household heads with tertiary level education adopted innovation, followed by 47
percent of household heads with secondary level education, 45 percent of household heads
with primary level education and at the bottom is household heads with no education (25
percent) (Table 4).
40
Table 4: Innovation by level of education
Table 5 compares the mean of adopters and non-adopters of innovation for different
variables along with the mean difference and its statistical significance. Almost 15 percent
of school going children surveyed in my sample did not attend school in the last two week
before the survey was conducted. This number excludes school absenteeism due to school
break and absent teacher (see Table 9 in appendix). Burke and Beegle (2004) found that
a week before their data was collected 17 percent of the surveyed children did not attend
school.Households that adopted innovation reported less household school absenteeism
compared to non-adopters. Adopters had almost 1% less household school absenteeism
compared to non-adopters but this difference is not statistically significant. Adopters are
younger than non-adopters and have more years of education. Household size is the total
number of people in the household whiles the numbr of children as used in table is the
number of children who satisfies that age range (5-15) in the study and are currently in
primary school. Innovative households have on average one household member more than
non-innovative households. Adopters and non-adopters have almost the same number of
children in primary school. It takes children of adopters more time to commute to school
41
than that of non-adopters. Adopters own more livestock, have higher income and bigger
farm size than non-adopters. Adopters hold about 3 hectares more land than non-adopters.
Non-adopters live about 11km closer to a district headquarter. The means of innovation as
measured in different households and in village level are closer to each other.
Non-
Full sample Adopters
adopters
Mean Mean Mean Mean diff
Variables
(SE) (SE) (SE) (SE)
AbsP 0.146 0.134 0.127 0.007
(.0122) (.0111) (.0118) (.0162)
Age 48.02 51.16 49.87 1.29***8
(.5336) (.3939) (.4751) (.6191)
HH_size 7.52 4.73 5.32 -0.59***
(.1718) (.0870) (.1020) (.1353)
Num. of kids 2.58 2.32 2.33 -0.08
(.0609) (.0440) (.4745) (.0651)
SchTime(hrs) 0.48 0.45 0.48 -0.03
(.0179) (.0154) (.0166) (.0227)
TLU 2.07 0.97 1.77 -0.80***
(.3201) (.1363) (.2093) (.2469)
Log income 15.88 15.04 15.50 -0.46***
(.0986) (.0799) (.0892) (.1218)
Fsize (ha) 8.83 5.97 8.32 -2.65***
(.8395) (.4839) (.6700) (.8066)
Extdt (km) 44.89 34.93 45.89 -10.96***
(3.2642) (1.507) (2.250) (2.603)
Innov 0.40
(.0090)
Vinnov 0.41
(.0022)
School absenteeism is low in households where the head is a female (Figure 4). School
absenteeism by level of education shows no clear trend (Figure 5). Household heads with
tertiary education had the highest school absenteeism compared to no education, primary
and secondary level household heads. It must be noted that only 0.2% of the household
8
Ho: Mean diff = 0 Ha: Mean diff ≠ 0 Confidence level=95%
42
heads studied had tertiary education level. It can be observed, that households with
secondary education level had the least household school absenteeism.
43
Rural households differ from households in urban areas in their rate of household school
absenteeism. Absenteeism is lower in urban areas than in rural areas (Figure 6). Figure 7
explores the difference in school absenteeism among households with less than 2 hectares
of farm land (regarded as small-holder farmers) and those cultivating on more than 2
hectares. School absenteeism is high in small-holder farmers compared to the rest.
44
Results
Table 6:Estimated coefficients and standard errors for household school absenteeism among primary school-
aged children in Tanzania
OLS IV
Table 6 reports the outcome of the OLS and IV regression models. Innovation adoption
significnatly decreases household school absenteeism in both models. However the
45
magnitude of the effects and significance differ, the OLS model reported that innovation is
likely to result in 5 percent (p<0.1) less absenteeism whiles the IV model reported almost
30 percent (p<0.01) less absenteeism due to innovation adoption. The age of the household
head has a significantly positive effect on household school attendance among primary
school children. The older the household head the lower the proportion of household
absenteeism. The coeffcient of the two models were closer in magnitude but IV model was
significant at 10 percent whiles the OLS was significant at 5 percent significance level.
Female household heads are likely to experience lower absenteeism compared to the male
households, however the outcome is not statistically significant. The higher the education
level of the household head, the lower the household absenteeism. The effect of education
on household absenteeism was not statistically significant. The higher the farm and
household size the higher the proportion of school absenteeism. In both models having a
higher farm and household size likely increases household absenteeism but this outcome
is not statistically significant. The number of children per household has a negative
relationship with school absenteeism. The more children in a household, the less the
proportion of absenteeism. Household school absenteeism increases as the time from
home to school increases. An hour increase in average time taken to school by children in
a household likely results in 9 percent (OLS) and 10 percent (IV) increase in household
absenteeism. The outcome is statistically significant at p<0.01 in the two models. The higher
the number of livestock possessed by the households, the higher the rate of school
absenteeism. This outcome is statistically not significant in both the OLS and IV models.
The test for endogeneity and strength of the instrument used in the IV model is presented
below.
Tests of Endogeneity
To verify the claim that innovation is endogenous, I conducted the endogeneity post-
estimation test using the ‘estat endogenous’ command available in stata13. The test
hypothesizes that innovation adoption in the 2SLS model is exogenous. The results of the
test are as follows: Durbin (score) chi2(1) = 6.41294 (p = 0.0113); Wu-Hausman F (1,696)
= 6.36187 (p = 0.0119). The null hypothesis can be rejected (at p<0.05) which implies that
innovation is endogenous and therefore OLS estimation might not be considered accurate
and consistent than the IV estimates. This result supports the choice of instrumental
variable regression as the method for exploring the relationship between innovation and
school attendance.
46
Test of Instrument Strength
A weak instrument has been shown by Baser (2009) as more harmful, because it produces
inaccurate and inconsistent estimates worse than the problem (OLS) it is meant to solve.
To test the strength of village level innovation (the instrument), I used the Stata 13 post
estimation command ‘estat firststage’. The null hypothesis is that the instrument is weak.
The test results are as follows:
The instrument can be established as a strong instrument because the minimum eigenvalue
(33.3624) is greater than the value of the norminal 5% Wald test at 10% bias tolerance
level.
47
Discussion
The basic question investigated in this thesis explores the relationship between innovation
and household school absenteeism. Two sides of innovation evoke the interest of this
question. The first side is that innovation improves household welfare which may reflect in
low household absenteeism. The other side is that innovation may draw upon children as
part of household labour, the demand for children on the farm will clearly interrupt school
attendance resulting in higher household school absenteeism among adopter households.
The results in Table 5 indicate that adopters and non-adopters of innovation differ in their
characteristics. It confirms the well documented observation that adopters of innovation are
younger, possess more land and livestock, have higher income and large household size.
This outcome is in line with what Shiferaw et al. (2014) found in Ethiopia. One interesting
observation from Table 5 is that non-adopters on average reside closer to a district
headquarters than adopters. In most developing countries, the district headquarters is a
building complex that contains majority of the government institutions including the
agricultural extension office. Therefore, one would assume that residing closer to a district
headquarter means easy access to extension services (where many agricultural innovations
in developing countries are promoted). The contra result can be explained that, despite
adopters living far from a district headquarter, adopters are wealthier (have higher income,
bigger farm size, more livestock) which makes it easy for them to afford extension services
which sometimes requires expenses on the side of the farmer. Based on the outcome of
the endogeneity and instrument strength test , I establish that the IV model is the most
accurate and consistent model of the two (OLS and IV) for explaining the effect of innoavtion
on household school absenteeism (See Table 6 & Table 6). Therefore,the discussion will
center on the outcome of IV model . From Table 6, the results reveal that innovation benefits
primary school children by improving their school attendance. The exact channel through
which innovation impacts school attendance was not the focus of this thesis. However,
inference drawn by comparing my results and the results of Kana et al. (2008) and Shiferaw
et al. (2014) explains one channel. Shiferaw et al. (2014) found that adoption of improved
wheat improves household food security, meanwhile Kana et al. (2008) found that children
from food insecure households suffer from low school attendance. Therefore, if innovation
improves household food security and food security affects school attendace, then it can
infered that innovation reduces school absenteeism through its improvement in household
food security. The effect of innovation on household absenteeism (in Table 6) must however
be treated with caution because the analysis excluded majority of the households due to
missing observations in the parameters used. Also from table 6, the older the household
48
head , the lower the occurrence of school absenteeism among primary school children.
Beegle et al. (2009) demonstrated using a random effects tobit estimation that increase in
household head age increases child’s hours of school attendance significantly. Abid et al.
(2015) also observed that older household are more likely to send their children to school
compared to younger household heads. Beegle et al. (2009) found that the gender of the
household head do not explain why children are not attending school which is similar to
what was found in this thesis. Children’s commuting time to get to school affects their
participation in school. The household average time it takes to reach school in the sample
used in this thesis is 48 minutes (table 5). And, the longer the time the higher the proportion
of household school absenteeism ( β=0.10, p<0.001; in both models) (Table 6). Admassie
(2003) shows in rural Ethiopia that 15 percent of the surveyed children cited distance of
school from home as one of the reasons for not attending school. Kana et al. (2008) also
found that the shorter the distance from home to school, the better the educational
attainments of the child. The distance and time to get to school depends on the means of
transportation. In this thesis, 90% of the surveyed children walk on foot to school, 5.2
percent used public transport, 2.4 and 1.4 percent used bike and private vehicle
respectively. The large number of children commuting to school on foot accounts for the
higher average in time it takes to get to school. The distance problem can be tackled by
building more schools in the community or improving the means of transportation (state-
sponsored buses) which will reduce the school commuting time and therefore reduce school
absenteeism.
49
Conclusion and Recommendations
The focus of this thesis was on exploring the relationship between agricultural innovation
and school absenteeism. Chronic absenteeism described as more than 20 days of being
absent for a school year has a negative consequence on students (Olson, 2014). The
negative consequences of poor attendance are felt as early as 3rd grade where children’s
reading proficiency is affected. It counts as a leading indicator of school drop-out in the 6th
grade and eventually predicts the performance of students (Attendance Works, NA).
Household welfare improvements are undisputed benefits of agricultural innovation ( See
Shiferaw et al., 2014). Most often, innovation studies do not consider children as direct
potential beneficiaries of innovation. I investigated if adopting innovation can reduce the
incidence of household school absenteeism in primary school children in Tanzania. To be
able to do this, I used cross-section data from the Tanzanian national survey 2010/2011.
The sample of household studied were those that have primary school children aged within
5-15 years. I excluded non-farming households and households where absenteeism was
‘legitimately’ beyond the influence of the household head or the child. Empirical results from
an OLS and IV models indicate that innovation reduces household school absenteeism.
Innovation affects school absenteeism positively; 5 percent (OLS) and 30 percent (IV) less
absenteeism is associated innovation adoption. Despite the positive effect of innovation on
school attendance, the results must be interpreted with caution due to large number of
missing data. The thesis also found age and school commuting time as predictors of
household absenteeism. Older household heads are likely to experience less absenteeism
whiles longer commuting time increases the occurrence of household absenteeism.
The reults show promoting agricultural innovation can also drive down school absenteeism.
Policy makers may also consider improving the availability of schools in communities in
Tanzania to reduce the time it takes for children to commute to school. Alternatively,
communities without schools or schools located far from the residence of most students
should be provided with buses to reduce the time children take in walking on foot to school.
I conclude that the positive impact of innovation can be directly experienced by children as
improvement in their school attendance. Investigating school absenteeism on household
level as indicated in the limitation section percluded child specific characteristics like age,
birth order, relationship of the child to the household head found in other studies to influence
school attedance. Further studies can consider individual child characteristics as the unit of
analysis instead of the household as a unit of analysis. Panel analysis is also recommended
to track the changes in household absenteeism due to innovation adoption.
50
Appendix
51
References
Abdoulaye, T., & Sanders, J. H. (2005). Stages and determinants of fertilizer use in
semiarid African agriculture: The Niger experience. Agricultural Economics, 32(2), 167–
179. https://doi.org/10.1111/j.0169-5150.2005.00011.x
Abdulai, A., & Huffman, W. E. (2005). The diffusion of new agricultural tecnologies: The
case of crossbreed-cow technology in Tanzania. American Journal of Agricultural
Economics, 87(3), 645–659.
Abid, G., Rafiq, Z., Khan, B., & Ahmed, A. (2015). Child Trade-Off Theory: A Theoretical
Discussion on the Structure, Causes, Consequences and Eradication of Child Labor.
Bulletin of Business and Economics, 4(1), 24–34.
https://doi.org/10.1080/09645292.2011.641271
Admassie, A. (2003). Child labour and schooling in the context of a subsistence rural
economy: Can they be compatible? International Journal of Educational Development,
23(2), 167–185. https://doi.org/10.1016/S0738-0593(02)00012-3
Al-Samarrai, S., & Peasgood, T. (1998). Educational attainments and household
characteristics in Tanzania. Economics of Education Review, 17(4), 395–417.
https://doi.org/10.1016/S0272-7757(97)00052-6
Alumira, J., & Rusike, J. (2005). The Green Revolution in Zimbabwe. Electronic Journal of
Agricultural and Development Economics, 2(1), 50–66.
Alvi, E., & Dendir, S. (2011). Sibling Differences in School Attendance and Child Labour in
Ethiopia. Oxford Development Studies, 39(3), 285–313.
https://doi.org/10.1080/13600818.2011.598923
Amare, M., Asfaw, S., & Shiferaw, B. (2012). Welfare impacts of maize-pigeonpea
intensification in Tanzania. Agricultural Economics, 43(1), 27–43.
https://doi.org/10.1111/j.1574-0862.2011.00563.x
Anandajayasekeram, P., Kaliba, A. R., Mwangi, W., Martella, D., & Shao, F. M. (2007).
Economic impact of maize research in Tanzania. Retrieved from
https://www.researchgate.net/publication/286677015
Anangisye, W. (2011). Bottlenecks in the access to primary education in Tanzania: The
struggles of Vulnerable School Children in Makete District. KEDI Journal of Educational
Policy, 8(2&primo_mobile=1), 323–344.
Asadul, I., & Chongwoo, C. (2013). Child labor and schooling responses to access to
microcredit in rural Bangladesh. Economic Inquiry, 51(1), 46–61.
https://doi.org/10.1111/j.1465-7295.2011.00400.x
Attendance Works. (NA). 10 Facts About School Attendance - Attendance Works.
Retrieved from http://www.attendanceworks.org/facts-stats-school-attendance/
Balfanz, R., & Byrnes, V. (2012). Chronic Absenteeism: Summarizing What We Know
From Nationally Available Data. Baltimore.
Banful, A., Nkonya, E., & Oboh, V. (2010). Constraints to Fertilizer Use in Nigeria: Insights
from Agricultural Extension Service. Retrieved from
https://www.researchgate.net/publication/46442050
Barrantes, C., & Yagüe, J. L. (2015). Adults’ Education and Agricultural Innovation: A
Social Learning Approach. Procedia - Social and Behavioral Sciences, 191, 163–168.
https://doi.org/10.1016/j.sbspro.2015.04.387
52
Barrett, C. B., Sherlund, S. M., & Adesina, A. A. (2001). Shadow Wages, Allocative
Inefficiency, and Labor Supply in Smallholder Agriculture. SSRN Electronic Journal.
Advance online publication. https://doi.org/10.2139/ssrn.257330
Baser, O. (2009). Too much ado about instrumental variable approach: Is the cure worse
than the disease? Value in health : the journal of the International Society for
Pharmacoeconomics and Outcomes Research, 12(8), 1201–1209.
https://doi.org/10.1111/j.1524-4733.2009.00567.x
Beegle, K., Dehejia, R., & Gatti, R. (2009). Why Should We Care About Child Labor?: The
Education, Labor Market, and Health Consequences of Child Labor. Journal of Human
Resources, 44(4), 871–889. https://doi.org/10.1353/jhr.2009.0025
Beegle, K., Dehejia, R., Gatti, R., & Krutikova, S. (2008). The Consequences of Child
Labor: Evidence from Longitudinal Data in Rural Tanzania. Policy Research Working
Paper, 4677. Retrieved from https://www.researchgate.net/publication/23723534
Beyene, A. D., & Kassie, M. (2015). Speed of adoption of improved maize varieties in
Tanzania: An application of duration analysis. Technological Forecasting and Social
Change, 96, 298–307. https://doi.org/10.1016/j.techfore.2015.04.007
Bezu, S., Kassie, G. T., Shiferaw, B., & Ricker-Gilbert, J. (2014). Impact of Improved
Maize Adoption on Welfare of Farm Households in Malawi: A Panel Data Analysis.
World Development, 59, 120–131. https://doi.org/10.1016/j.worlddev.2014.01.023
Burke, K., & Beegle, K. (2004). Why Children Aren't Attending School: The Case of
Northwestern Tanzania. Journal of African Economies, 13(2), 333–355.
https://doi.org/10.1093/jae/ejh011
Chongela, J. (2015). Contribution of Agriculture Sector to the Tanzanian Economy.
American Journal of Research Communication, 3(7), 57–70. Retrieved from
http://www.usa-journals.com/wp-content/uploads/2015/06/Chongela_Vol37.pdf
Diallo, Y., Etienne, a., & Mehran, F. (2013). Global child labour trends 2008 to 2012.
Geneva: ILO.
Diederen, P., van Meijl, H., Wolters, A., & Bijak, K. (2003). Innovation adoption in
agriculture : innovators, early adopters and laggards. Cahiers d’Economie et de
Sociologie, 67, 29–50. Retrieved from https://hal.archives-ouvertes.fr/hal-
01201041/document
Emsley, R., Lunt, M., Pickles, A., & Dunn, G. (2008). Implementing double-robust
estimators of causal effects. The Stata Journal, 8(3), 334–353. Retrieved from
http://www.stata-journal.com/article.html?article=st0149
Epstein, J. L., & Sheldon, S. B. (2002). Present and Accounted for: Improving Student
Attendance Through Family and Community Involvement. The Journal of Education
Research, 95(5). Retrieved from http://www.attendanceworks.org/wordpress/wp-
content/uploads/2017/03/Present_and_Accounted_For_Improving_Student_Attendanc
e_Through_Family_and_Community_Involvement.pdf
Etoundi, S., & Dia, B. K. (2008). Determinants of the adoption of improved varieties of
Maize in Cameroon: case of CMS 8704. Proceedings of the African Economic
Conference, 397–413.
Feleke, S., & Zegeye, T. (2006). Adoption of improved maize varieties in Southern
Ethiopia: Factors and strategy options. Food Policy, 31(5), 442–457.
https://doi.org/10.1016/j.foodpol.2005.12.003
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. https://doi.org/10.1007/s10584-015-1459-2
53
Garbarino, S., & Holland, J. (2009). Quantitative and qualitative methods in impact
evaluation and measuring results. Retrieved from Governance and Social Development
Resource Centre website: http://www.gsdrc.org/docs/open/eirs4.pdf
Ghadim, A. A., & Pannell, D. J. (1999). A conceptual framework of adoption of an
agricultural innovation. Agricultural Economics, 21(2), 145–154.
https://doi.org/10.1016/S0169-5150(99)00023-7
Henley, R., McAlpine, K., Mueller, M., & Vetter, S. (2010). Does school attendance reduce
the risk of youth homelessness in Tanzania? International journal of mental health
systems, 4, 28. https://doi.org/10.1186/1752-4458-4-28
International Labour Oganization (ILO), & International Programme on the Elimination of
Child Labour (IPEC). (2017). What is child labour (IPEC). Retrieved from
http://ilo.org/ipec/facts/lang--en/index.htm
Jewitt, S., & Ryley, H. (2014). It’s a girl thing: Menstruation, school attendance, spatial
mobility and wider gender inequalities in Kenya. Geoforum, 56, 137–147.
https://doi.org/10.1016/j.geoforum.2014.07.006
Kadzamira, E., & Rose, P. (2003). Can free primary education meet the needs of the
poor?: Evidence from Malawi. International Journal of Educational Development, 23(5),
501–516. https://doi.org/10.1016/S0738-0593(03)00026-9
Kafle, B. (2010). Determinants of adoption of improved maize varieties in developing
countries: A review. International Research Journal of Applied and Basic Sciences,
1(1), 1–7. Retrieved from http://www.ecisi.com
Kajisa, K., & Nakano, Y. (2012). How does the Adoption of Modern Variety increase
Productivity and Income?: A Case of the Rice Sector in Tanzania.
Kaliba, A. R., Verkuijl, H., & Mwangi, W. (2000). Factors Affecting Adoption of Improved
Maize Seeds and Use of Inorganic Fertilizer for Maize Production in the Intermediate
and Lowland Zones of Tanzania. Journal of Agricultural and Applied Economics,
32(01), 35–47. https://doi.org/10.1017/S1074070800027802
Kana, M., Phoumin, H., & Seiichi, F. (2008). Does child labor have a negative impact on
child education and health? A Case Study in Rural Cambodia. GSICS Working Paper
Series. (19).
Kassie, M., Jaleta, M., & Mattei, A. (2014). Evaluating the impact of improved maize
varieties on food security in Rural Tanzania: Evidence from a continuous treatment
approach. Food Security, 6(2), 217–230. https://doi.org/10.1007/s12571-014-0332-x
Kathage, J., Kassie, M., Shiferaw, B., & Qaim, M. (2015). Big Constraints or Small
Returns?: Explaining Nonadoption of Hybrid Maize in Tanzania. Applied Economic
Perspectives and Policy. Advance online publication.
https://doi.org/10.1093/aepp/ppv009
Khana, R. E. A., Khan, T., & Satter, R. (2010). A comparative analysis of rural and urban
child labor in Pakistan. European Journal of Economics, Finance and Administrative
Sciences, 27. Retrieved from https://www.researchgate.net/publication/289388627
Khonje, M., Manda, J., Alene, A. D., & Kassie, M. (2015). Analysis of Adoption and
Impacts of Improved Maize Varieties in Eastern Zambia. World Development, 66, 695–
706. https://doi.org/10.1016/j.worlddev.2014.09.008
Läpple, D., Renwick, A., & Thorne, F. (2015). Measuring and understanding the drivers of
agricultural innovation: Evidence from Ireland. Food Policy, 51, 1–8.
https://doi.org/10.1016/j.foodpol.2014.11.003
54
Larsen, A. F., & Lilleør, H. B. (2016). Can Agricultural Interventions Improve Child
Nutrition?: Evidence from Tanzania. The World Bank Economic Review, lhw006.
https://doi.org/10.1093/wber/lhw006
Lotter, D. (2015). Facing food insecurity in Africa: Why, after 30 years of work in organic
agriculture, I am promoting the use of synthetic fertilizers and herbicides in small-scale
staple crop production. Agriculture and Human Values, 32(1), 111–118.
https://doi.org/10.1007/s10460-014-9547-x
Maredia, M. K., Shankar, B., Kelley, T. G., & Stevenson, J. R. (2014). Impact assessment
of agricultural research, institutional innovation, and technology adoption: Introduction
to the special section. Food Policy, 44, 214–217.
https://doi.org/10.1016/j.foodpol.2013.10.001
Mmbando, F., & Baiyegunhi, L. (2016). Socio-economic and institutional factors
influencing adoption of improved maize varieties in Hai district, Tanzania. Retrieved
from https://www.researchgate.net/publication/304773194
Ntwenya, J., Kinabo, J., Msuya, J., Mamiro, P., Mamiro, D., & Katalambula, L. (2015).
Household food insecurity and associated factors in rural communities: A case of
Kilosa District, Tanzania. African Journal of Agricultural Research, 10(52), 4783–4794.
https://doi.org/10.5897/AJAR2012.1718
Okui, R., Small, D. S., Tan, Z., & Robins, J. M. (2012). Doubly robust instrumental variable
regression. Statistica Sinica, 22(1), 173–205. Retrieved from
http://www.jstor.org/stable/24310144
Olson, L. S. (2014). Why September Matters: Improving Student Attendance. Retrieved
from Baltimore Education Research Consortium website: http://baltimore-berc.org/wp-
content/uploads/2014/08/SeptemberAttendanceBriefJuly2014.pdf
Peck, R., Olsen, C., & Devore, J. L. (2008). Introduction to statistics and data analysis
(Edition 3).
Ransom, J. (2003). Adoption of improved maize varieties in the hills of Nepal. Agricultural
Economics, 29(3), 299–305. https://doi.org/10.1016/S0169-5150(03)00057-4
Rogers, E. M. (2003). Diffusion of innovations (5th ed.). New York, N.Y.: Simon &
Schuster.
Sanginga, P. C., Adesina, A. A., Manyong, V. M., Otite, O., & Dashiell, K. E. (1999).
Social impact of soybean in Nigeria's southern Guinea savanna. Agricultural
Economics.
Seid, Y., & Gurmu, S. (2015). The role of birth order in child labour and schooling. Applied
Economics, 47(49), 5262–5281. https://doi.org/10.1080/00036846.2015.1047086
Shiferaw, B., Kassie, M., Jaleta, M., & Yirga, C. (2014). Adoption of improved wheat
varieties and impacts on household food security in Ethiopia. Food Policy, 44, 272–
284. https://doi.org/10.1016/j.foodpol.2013.09.012
Sifuna, D. N. (2007). The Challenge of Increasing Access and Improving Quality: An
Analysis of Universal Primary Education Interventions in Kenya and Tanzania since the
1970s. International Review of Education, 53(5-6), 687–699.
https://doi.org/10.1007/s11159-007-9062-z
Skoufias, E. (1994). Using Shadow Wages to Estimate Labor Supply of Agricultural
Households. American Journal of Agricultural Economics, 76(2), 215.
https://doi.org/10.2307/1243623
Smale, M., Just, R. E., & Leathers, H. D. (1994). Land Allocation in HYV Adoption Models:
An Investigation of Alternative Explanations. American Journal of Agricultural
Economics, 76(3), 535. https://doi.org/10.2307/1243664
55
StataCorp. (2013). Stata 13 Base Reference Manual: ivregress postestimation —
Postestimation tools for ivregress: College Station, TX: Stata Press.
Stephen, L., Zubeda, M., & Hugo, D. G. (2014). The use of improved maize varieties in
Tanzania. African Journal of Agricultural Research, 9(7), 643–657.
https://doi.org/10.5897/AJAR11.065
Stock, J. H., & Yogo, M. (2003). Testing for weak instruments in linear IV regression.
Tanzania National Bureau of Statistics. (2016). Tanzania mainland child labour survey
2014: Analytical report. Dar es Salaam: National Bureau of Statistics.
Tanzanian National Bureau of Statistics. (2011). Tanzania National Panel Survey Report:
Round 1,2008-2009.
United Nations Department of Economics and Social Affairs. (2015). The Millenium
Develpoment Goals Report, United Nations. Retrieved from
http://www.un.org/millenniumgoals/2015_MDG_Report/pdf/MDG%202015%20rev%20(
July%201).pdf
United Republic of Tanzania-Mainland. (2014a). Education for all 2015 national review.
Retrieved from http://unesdoc.unesco.org/images/0023/002314/231484e.pdf
United Republic of Tanzania-Mainland (Ed.). (2014b). Education for all 2015 national
review. Handbook of Agricultural Economics: Elsevier.
Vanlauwe, B., Wendt, J., Giller, K. E., Corbeels, M., Gerard, B., & Nolte, C. (2014). A
fourth principle is required to define Conservation Agriculture in sub-Saharan Africa:
The appropriate use of fertilizer to enhance crop productivity. Field Crops Research,
155, 10–13. https://doi.org/10.1016/j.fcr.2013.10.002
Verkaart, S., Munyua, B. G., Mausch, K., & Michler, J. D. (2017). Welfare impacts of
improved chickpea adoption: A pathway for rural development in Ethiopia? Food
Policy, 66, 50–61. https://doi.org/10.1016/j.foodpol.2016.11.007
Wheeler, S. A., Zuo, A., Bjornlund, H., Mdemu, M. V., van Rooyen, A., & Munguambe, P.
(2016). An overview of extension use in irrigated agriculture and case studies in south-
eastern Africa. International Journal of Water Resources Development, 1–15.
https://doi.org/10.1080/07900627.2016.1225570
Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data: MIT
Press. Retrieved from http://www.jstor.org/stable/j.ctt5hhcfr
World Bank. (2006). Enhancing agricultural innovation:how to go beyond the strethening
of research systems. Washington DC. Retrieved from Agricultural and Rural
Development,World Bank website:
http://siteresources.worldbank.org/INTARD/resources/enhancing_Ag_innovtion.pdf
World Food Programme. (2017). What is food security? | WFP | United Nations World
Food Programme - Fighting Hunger Worldwide. Retrieved from
https://www.wfp.org/node/359289
56